Tax Examiners and Collectors, and Revenue Agents
13-2081.00Median wage $62,370/yr56,610 employed (US)Rank #145 of 923 scored · top 16% by substitution
Determine tax liability or collect taxes from individuals or business firms according to prescribed laws and regulations.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution42
Exposure43
Augmentation66
Substitution — the headline: capability discounted by cost, barriers and adoption.
Exposure — technical capability alone, regardless of whether anyone deploys it.
Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.
Tasks on the substitution scale
21 rated tasks, binned by substitution score.
Position among all scored occupations
Distribution of 923 occupation scores; the marker is this occupation.
Tasks with substitution ≥ 70
19%
Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.
Why this score
The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.
Task automatabilityw 35%43
panel mean rating 2.7/5 → substitution pressure 43/100
Technical feasibility todayw 20%43
panel mean rating 2.7/5 → substitution pressure 43/100
Cost vs. human wagew 15%48
panel mean rating 2.9/5 → substitution pressure 48/100
Adoption barriersw 20%inverted — strong barriers lower the score33
panel mean rating 3.7/5 (barrier strength) → substitution pressure 33/100
Sector adoption velocityw 10%39
panel mean rating 2.6/5 → substitution pressure 39/100
Task breakdown (21 tasks)
Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.
Enter tax return information into computers for processing.
92CI 92–92 · exposure 100 · augmentation 63 · importance 3.8/5 · click for rater detail
Enter tax return information into computers for processing.
92| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Tax agencies (IRS, state revenue departments) and large tax preparation firms have already deployed substantial automation for return intake and data entry; adoption is accelerating as cloud and RPA tools mature in the financial services sector. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Government tax agencies and financial services broadly have adopted automated data capture and e-filing systems, though public sector procurement can lag private industry somewhat. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI assists human tax examiners by pre-populating fields and flagging inconsistencies, reducing manual keying effort and error rates. The human remains in the loop for review and exception handling, boosting throughput moderately. |
| Augmentation potential | claude-sonnet-5 | 4/5 | Even where full automation isn't complete, AI-assisted extraction and validation tools significantly speed up examiners' data entry and review workflows. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Data entry of tax return information is highly structured, rule-based, and well-suited to OCR and document parsing systems. Current AI can reliably extract and input standard tax form data (names, numbers, filing status) with >50% time savings via automated workflows, requiring minimal human review for straightforward returns. |
| Task automatability | claude-sonnet-5 | 5/5 | Data entry from tax documents into processing systems is a structured, repetitive task well within the capability of OCR/IDP and RPA systems combined with LLM-based extraction, meeting the 50% time-saving bar easily. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | There are minimal legal barriers to automating data entry itself; however, organizations require oversight to catch errors before downstream processing, and legacy system integration can create friction. No licensing requirement prevents substitution. |
| Adoption barriers | claude-sonnet-5 | 2/5 | Some oversight is needed for accuracy and fraud detection, but data entry itself carries no licensing requirement or legal mandate for human performance. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated document processing and data entry costs pennies per return at scale, while manual entry by a tax examiner costs $20–50+ per return in loaded labor. AI is orders of magnitude cheaper for this routine clerical task. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Automated OCR/data extraction pipelines cost fractions of a cent per document versus the loaded wage of a human keying in data, an order of magnitude or more cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Mature products like cloud-based tax processing platforms, RPA tools, and document parsing services (e.g., AWS Textract, UiPath) are deployed in production by tax agencies and firms today, reliably ingesting millions of returns with high accuracy on structured fields. |
| Technical feasibility today | claude-sonnet-5 | 5/5 | Mature commercial products (e.g., intelligent document processing, e-filing systems, tax software like TurboTax/Intuit backends) already perform automated data capture and entry at scale in production. |
Send notices to taxpayers when accounts are delinquent.
83CI 74–92 · exposure 87 · augmentation 50 · importance 4.6/5 · click for rater detail
Send notices to taxpayers when accounts are delinquent.
83| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Tax agencies and revenue departments have been automating routine notice issuance for over a decade, with widespread adoption in state and federal systems. Public records and agency modernization initiatives show that most high-volume tax collection operations now use automated or semi-automated notice systems. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Tax administration and revenue collection agencies have long since automated bulk notice generation and are increasingly adopting AI-enhanced personalization and triggers, reflecting fast-moving digitization in financial/government back-office functions. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist human revenue agents by flagging unusual cases, generating draft notices for review, and organizing delinquency data. However, for routine high-volume notices, augmentation is less relevant because the task is largely automatable end-to-end. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist in refining notice language, prioritizing cases, and personalizing communications, but much of the core sending function is already fully automated rather than augmented. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can reliably identify delinquent accounts, generate personalized notice content, and initiate bulk mail or email delivery with minimal manual oversight. The main tasks—data validation, template selection, and dispatch—are well within automation capability, though human review of edge cases may be needed. |
| Task automatability | claude-sonnet-5 | 5/5 | Generating and sending delinquency notices based on account status is a highly templated, rules-based task that off-the-shelf automation and AI-driven document generation systems can fully execute with minimal human intervention. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Regulatory requirements (e.g., proof of service, record-keeping, taxpayer rights disclosures) and agency preferences for human oversight create moderate friction. However, no hard licensing requirement prevents the agency from automating the notice itself; compliance can be built into the system design. |
| Adoption barriers | claude-sonnet-5 | 2/5 | While tax agencies have compliance and legal-formatting requirements for notices, the actual generation/sending is already routinely automated in practice with limited need for individualized human sign-off. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI-driven notice systems cost pennies per notice (data retrieval, template rendering, email/print dispatch), whereas human notice-sending involves salary, oversight, and turnaround time. The per-unit cost is at least an order of magnitude cheaper with automation. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Automated notice generation and dispatch costs a tiny fraction of a cent per notice compared to a human agent manually drafting and sending each one. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Production systems for automated notice generation and delivery are deployed across tax agencies, collection platforms, and financial services. While some jurisdictions retain manual review steps, mature products reliably handle routine delinquency notices at scale with acceptable error rates. |
| Technical feasibility today | claude-sonnet-5 | 5/5 | Government and financial institutions already deploy mature automated notice systems (rule engines plus generated correspondence) at scale for delinquent account communications. |
Check tax forms to verify that names and taxpayer identification numbers are correct, that computations have been performed correctly, or that amounts match those on supporting documentation.
83CI 74–92 · exposure 87 · augmentation 75 · importance 4.5/5 · click for rater detail
Check tax forms to verify that names and taxpayer identification numbers are correct, that computations have been performed correctly, or that amounts match those on supporting documentation.
83| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Revenue agencies (IRS, state tax boards) and large accounting firms have actively deployed automated form verification and document scanning systems for years. Adoption is measurable and ongoing in the public sector and professional services, though smaller tax practices lag. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Tax administration is a heavily digitized, high-volume domain where automated matching (e.g., IRS document matching, e-file validation) has been standard practice for years, reflecting fast, deep adoption typical of finance/professional services. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI systems that flag inconsistencies, highlight suspect values, or pre-populate comparison tables substantially speed human reviewers' ability to spot errors and approve compliant forms. The human remains in the loop for judgment calls, but productivity gains are significant. |
| Augmentation potential | claude-sonnet-5 | 4/5 | Even where human review remains for edge cases or appeals, AI-driven flagging of discrepancies dramatically speeds up and focuses the examiner's attention on true exceptions rather than routine matches. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can reliably extract names, identification numbers, and perform computational verification through OCR and document comparison. The task involves rule-based validation against supporting documentation—core strengths of modern AI. However, complex edge cases or ambiguous cross-references may require human judgment, preventing a full 5 rating. |
| Task automatability | claude-sonnet-5 | 5/5 | This is a rules-based verification task involving matching data fields, checking arithmetic, and cross-referencing documents—well within current OCR/data-validation and AI capabilities to automate end-to-end with substantial time savings. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Tax examination has regulatory oversight and quality standards, but verification of forms and arithmetic is not itself legally gatekept to licensed professionals in most jurisdictions. Agencies must maintain accuracy and audit trails, requiring some human oversight, but no hard barrier prevents automation of the checking step itself. |
| Adoption barriers | claude-sonnet-5 | 2/5 | While final enforcement actions or audits may require an authorized agent's sign-off, the verification/matching step itself is largely administrative and already automated without licensing requirements blocking the check itself. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | The per-document cost of automated form checking (OCR + computation + database matching) is orders of magnitude cheaper than paying a human examiner to manually verify each field. Bulk processing at marginal cost favors AI heavily. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Automated matching and computation-checking software costs a small fraction of a cent per form compared to a human examiner's loaded wage for manual review of the same volume. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature document processing and data extraction tools (e.g., specialized tax software, RPA platforms, LLM-based document readers) are deployed in production by tax authorities and accounting firms today. Error rates on straightforward verification have fallen substantially, though complex or unusual filings still require review. |
| Technical feasibility today | claude-sonnet-5 | 5/5 | Tax agencies and payroll/accounting software already deploy automated validation systems (e-file checks, IRS matching programs, TurboTax-style error detection) that reliably perform this exact verification at scale in production. |
Maintain records for each case, including contacts, telephone numbers, and actions taken.
82CI 72–92 · exposure 87 · augmentation 88 · importance 4.3/5 · click for rater detail
Maintain records for each case, including contacts, telephone numbers, and actions taken.
82| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Government and financial services sectors are actively digitizing case management and record-keeping; modern tax agencies increasingly use integrated systems that automate logging and contact maintenance. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Government revenue agencies are generally slower adopters of AI than private finance, but case-management automation and RPA are increasingly piloted and deployed in this sector. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can assist by auto-populating records from emails and calls, suggesting next actions, and organizing case histories, substantially raising examiner productivity while they focus on judgment-heavy tax analysis. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI tools for automatic call logging, transcription, and structured note generation significantly boost examiner productivity while they remain responsible for case judgment and follow-up. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Recording and organizing structured data (contacts, phone numbers, actions) is a straightforward data entry and management task that AI systems can fully automate with minimal human intervention, easily achieving 50%+ time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 4/5 | Recording contacts, phone numbers, and actions taken is structured data entry that AI-assisted systems (CRM integrations, voice-to-text logging, automated case notes) can largely handle with human review, meeting the time-saving threshold for most of the workflow. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | While no legal restriction prevents automation of record-keeping itself, some oversight and quality assurance may be expected, and organizational legacy systems can create adoption friction, but these are surmountable barriers. |
| Adoption barriers | claude-sonnet-5 | 2/5 | Record-keeping itself carries no licensing requirement, though data accuracy, privacy, and audit-trail integrity in tax/revenue contexts create moderate compliance friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Database systems and automated logging cost negligibly compared to the loaded wage of a tax examiner, delivering an order of magnitude or greater cost advantage. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated logging and transcription tools cost a small fraction of a human examiner's time spent manually documenting each contact, though integration and oversight add some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Multiple deployed CRM and case management systems already reliably automate record maintenance, contact logging, and action tracking at scale in organizational settings including government agencies. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | CRM and case-management platforms with AI-assisted note-taking, call transcription, and auto-logging are widely deployed in government and collections agencies today, though some manual verification and formatting persists. |
Determine appropriate methods of debt settlement, such as offers of compromise, wage garnishment, or seizure and sale of property.
51CI 25–78 · exposure 58 · augmentation 88 · importance 4.5/5 · click for rater detail
Determine appropriate methods of debt settlement, such as offers of compromise, wage garnishment, or seizure and sale of property.
51| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Tax authorities are highly digitized, data-driven organizations with strong incentives to automate collections; major IRS and state revenue departments have already deployed automated settlement-recommendation systems, though human authorization remains mandatory. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are typically slow adopters of AI for consequential enforcement decisions due to regulatory, fairness, and audit requirements, though some analytics tools are piloted. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI systems substantially augment revenue agents by pre-screening cases, calculating settlement ranges, prioritizing high-value accounts, and drafting settlement notices, freeing agents to focus on complex negotiations and hardship review while maintaining legal accountability. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by analyzing taxpayer financial data, predicting collectability, and recommending options, significantly speeding up the agent's decision process while the agent retains final authority. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | AI systems can analyze financial records, calculate debt-settlement options, and recommend appropriate methods (compromise offers, garnishment thresholds, asset seizure prioritization) based on taxpayer financial data and regulatory guidelines, achieving substantial time savings over manual case review. |
| Task automatability | claude-sonnet-5 | 2/5 | Determining the appropriate debt settlement method requires judgment about taxpayer circumstances, legal constraints, and discretion that AI can inform but not fully replace end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Legal and regulatory barriers are substantial: tax collection authority, enforcement of seizure/garnishment orders, and debt-settlement decisions typically require a licensed agent or supervisor signature and judicial oversight in contested cases, limiting full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Legal authority to seize property, garnish wages, or approve compromises typically requires an authorized government official's decision, creating strong statutory and due-process barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Inference and case analysis on standard financial documents costs pennies per case; human revenue agents require six-figure salaries and significant training, making AI cost per case at least 100× lower for equivalent analytical output. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Given the need for human review, legal sign-off, and case-specific judgment, AI reduces some analysis costs but does not yet substantially undercut the loaded cost of a trained revenue agent for this decision. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Tax software and revenue-management platforms (used by IRS and state agencies) already implement rule-based decision engines for settlement method recommendation; these are deployed in production but typically require human review and authorization rather than fully autonomous operation. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | While AI decision-support tools exist for flagging risk and suggesting collection strategies, no deployed product autonomously determines and executes settlement method selection in production tax agencies. |
Notify taxpayers of any overpayment or underpayment, and either issue a refund or request further payment.
49CI 30–67 · exposure 55 · augmentation 75 · importance 4.4/5 · click for rater detail
Notify taxpayers of any overpayment or underpayment, and either issue a refund or request further payment.
49| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Tax agencies and revenue departments are government-heavy, regulated sectors with slow digitization. While they use audit-support software, actual adoption of AI agents for autonomous notification and refund issuance is minimal and cautious. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Government agencies have automated basic reconciliation and notice generation for years, but full modernization is slow and uneven due to legacy systems and budget constraints. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly assists by automatically flagging calculation errors, generating draft notifications, and organizing case data, raising examiner productivity substantially. The human examiner remains essential for legal authority and complex judgment, but AI transforms efficiency on the routine parts. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI and automated systems significantly speed up identification of discrepancies and drafting of notices, letting agents focus on complex or contested cases. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can calculate overpayments/underpayments and draft notifications, the task requires legal authority to issue official refunds or demands, which demands human review and authorization. Current systems cannot independently perform the full task end-to-end at the >50% time-saving threshold without substantial human oversight. |
| Task automatability | claude-sonnet-5 | 4/5 | Calculating balances and generating notification/refund or payment-request letters is largely rule-based and can be automated with tax software integrated with notification systems, though edge cases need human review. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax law and revenue regulations typically require that licensed tax examiners or revenue agents personally sign and authorize refund issuance and payment demands; liability and legal standing requirements create hard adoption barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | Government tax authority actions require legal accuracy and proper authorization; while much is already automated under statutory authority, disputed or complex cases require human agent sign-off, creating moderate barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI can reduce manual calculation and letter-drafting costs, but the need for tax professional review, verification, and legal authority to issue demands/refunds means human labor remains substantial. All-in cost is comparable to or slightly below human-only workflow. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated batch processing of notices and refunds is vastly cheaper per case than manual review by a revenue agent, though system maintenance and oversight add some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products exist to flag discrepancies and generate preliminary notifications, but no deployed system independently issues final refunds or legally binding payment demands without human agent sign-off. Material error rates and requirement for licensed review limit production reliability. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Tax agencies already use automated systems (e.g., IRS notices, e-filing reconciliation) to detect discrepancies and auto-generate refund or balance-due notices at scale, though exceptions still route to human agents. |
Answer questions from taxpayers and assist them in completing tax forms.
49CI 30–67 · exposure 50 · augmentation 75 · importance 4.2/5 · click for rater detail
Answer questions from taxpayers and assist them in completing tax forms.
49| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Government tax agencies and compliance-heavy sectors have been slow to deploy autonomous tax AI in production due to regulatory caution, public accountability requirements, and risk aversion around tax law errors, though pilot programs are increasing. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Private tax-prep companies have adopted AI assistants fairly quickly, but government tax agencies (IRS and state revenue departments) are slower and more cautious, producing a mixed, moderate adoption pace. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI systems demonstrably assist human tax agents by drafting responses, suggesting relevant forms, identifying potential deductions, and automating routine fact-gathering—substantially raising productivity while the agent remains accountable for accuracy and compliance. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools substantially speed up and improve human agents' ability to answer questions and guide form completion by providing instant lookups, drafting responses, and pre-filling data, while humans remain for verification and complex cases. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can draft standard responses and fill simple tax forms, the diverse nature of taxpayer questions, requirement for contextual judgment, and need to clarify complex individual circumstances makes end-to-end automation with 50% time savings unlikely without significant human oversight and exception-handling. |
| Task automatability | claude-sonnet-5 | 4/5 | Answering routine tax questions and helping fill out standard forms is largely conversational and rule-based, which chatbots and LLM-based assistants already handle well for common scenarios, though edge cases and complex returns still need human judgment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax advice and form completion are subject to regulatory oversight; unauthorized practice of tax law is prohibited in most jurisdictions, and tax agencies typically require licensed or government-employed agents to provide binding guidance—creating legal and liability barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | While government revenue agents have official authority and some interactions require verified human judgment or legal accountability, general taxpayer assistance and form help is not tightly restricted to licensed humans and is already partly self-service. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-driven customer service costs are lower per interaction, but the need for human agent oversight, compliance review, and handling of escalations means the all-in cost per fully resolved taxpayer case remains comparable to or sometimes higher than human-only handling. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated chat/form-assistance tools cost a small fraction of a per-interaction human agent's loaded wage, especially at high volume, though oversight and complex-case routing add some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed chatbots and AI assistants can handle routine tax questions and form-filling guidance in production, but material limitations exist around edge cases, error recovery, and liability—necessitating human verification for non-trivial inquiries. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Tax software (TurboTax, H&R Block, IRS chatbots) and AI-powered tax assistants are deployed at scale for basic Q&A and form guidance, but they still have material error rates on nuanced or unusual tax situations requiring escalation to humans. |
Review filed tax returns to determine whether claimed tax credits and deductions are allowed by law.
47CI 45–49 · exposure 50 · augmentation 75 · importance 4.2/5 · click for rater detail
Review filed tax returns to determine whether claimed tax credits and deductions are allowed by law.
47| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | The IRS and tax-prep firms are adopting AI-assisted tools for initial screening and categorization, but adoption remains pilot-heavy and focused on high-volume, low-complexity returns. Production-scale displacement of core determination work is limited by liability and regulatory oversight requirements. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Tax agencies (e.g., IRS, state revenue departments) are adopting analytics and automated flagging tools, but full agentic determination remains rare due to legal and procedural constraints, placing this in middling adoption territory. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools augment examiners substantially by automating document triage, flagging high-risk items, comparing returns to benchmarks, and suggesting applicable rules, allowing the human examiner to focus on substantive judgment. This represents meaningful productivity lift while maintaining human oversight. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools substantially speed up return screening, cross-referencing deduction eligibility rules, and flagging anomalies, greatly boosting examiner productivity while humans retain final determination authority. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can assist with pattern matching, flagging unusual deductions, and cross-referencing simple rules against return data, but the interpretative task of determining 'allowance by law' often requires contextual judgment about edge cases, taxpayer circumstances, and regulatory nuance that currently demands human review. Roughly half the routine compliance work could be automated, but the most complex determinations would remain. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can flag deductions/credits against rules and cross-check data, but complex cases involving ambiguous law, taxpayer intent, or supporting documentation review still require human judgment, limiting full end-to-end automation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | A licensed tax examiner or certified professional must review, approve, and take responsibility for determinations affecting refunds and liability exposure. Regulatory and legal liability asymmetry (disallowed credits can trigger penalties) and IRS compliance protocols create high barriers to unsupervised automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Determinations affecting taxpayer liability often require legal authority, due process, and agency sign-off by authorized personnel, creating strong regulatory and liability barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI-assisted review (software licensing, inference, oversight) approaches the cost of junior-level human reviewer labor, but full oversight and liability remain with humans. The cost ratio is roughly comparable to having a lower-wage reviewer perform initial filtering. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated rule-checking and anomaly detection is vastly cheaper per return than manual examiner review, especially for high-volume, straightforward filings, though complex audits still need costly human oversight. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Tax software and audit-assistance platforms exist and are deployed at scale (e.g., IRS tools, commercial tax software with compliance checks), but they operate within constrained rule sets and still require human review for non-standard returns, mixed income scenarios, and contestable interpretations. Error rates remain material for complex cases. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Tax software and government systems already use rule-based and ML flagging to screen returns for compliance issues, but these systems have material false-positive/negative rates and rarely make final determinations without human review. |
Contact taxpayers by mail or telephone to address discrepancies and to request supporting documentation.
38CI 25–51 · exposure 38 · augmentation 63 · importance 4.2/5 · click for rater detail
Contact taxpayers by mail or telephone to address discrepancies and to request supporting documentation.
38| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption in the public tax sector is slow; the IRS and state agencies remain conservative, using legacy systems and prioritizing human relationships with taxpayers. Private tax firms show faster tech adoption, but even there, contact and discrepancy resolution remain heavily human-driven due to compliance and relationship requirements. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are generally slower adopters of AI-driven taxpayer contact systems compared to private-sector information/finance firms, constrained by legacy systems and compliance requirements. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by drafting contact templates, flagging common discrepancies, organizing documentation requests, and summarizing taxpayer histories—useful productivity aids for the agent. However, the interaction itself (phone call, negotiating responses, exercising judgment) remains primarily human-centered, limiting transformative potential. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can draft correspondence, flag discrepancies, and pre-fill documentation requests, significantly speeding up the examiner's workflow while the human retains decision authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Initial contact and standard discrepancy notifications could be partially automated via mail merge or templated email, but the task fundamentally requires interpreting taxpayer situations, responding to nuanced questions, and exercising judgment about documentation requests—capabilities current AI cannot reliably do at equal quality. Perhaps 20–30% time savings on routine outreach, but not the ≥50% threshold. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can draft and even send standardized discrepancy notices and follow-up requests, but judgment about which discrepancies warrant contact, tone calibration, and handling taxpayer responses still requires human oversight for a large share of cases. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax examination is a regulated government function; IRS policy and law require human revenue agents or authorized personnel to make substantive determinations, request documentation in contexts involving dispute or audit, and sign off on collection actions. Liability and legal authority create hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | Formal notices often require authorized agency sign-off and must meet legal/procedural standards, and taxpayers may dispute or need human clarification, creating moderate friction despite no strict licensing requirement for the contact itself. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Human revenue agents earn substantial salaries ($50k+); the all-in cost of AI contact (infrastructure, oversight, error recovery, and inevitable human review) remains comparable to or higher than direct human outreach for non-trivial discrepancies. Only high-volume, routine notices approach cost parity. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Automated letter generation and basic phone/chat triage cost far less per contact than an agent's time, though oversight and escalation paths add some cost back. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Automated tax outreach systems exist (templated letters, basic IVR systems), but they handle only the most routine, low-stakes contacts. Real-world tax examination requires understanding context, handling disputes, and building compliance—tasks where current products show material error rates and narrow scope. No deployed system does the full task reliably. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Chatbots, IVR systems, and automated notice-generation are deployed in tax agencies today, but complex or disputed cases are routed to human agents, so reliability is narrow in scope. |
Review selected tax returns to determine the nature and extent of audits to be performed on them.
37CI 30–45 · exposure 42 · augmentation 75 · importance 4.0/5 · click for rater detail
Review selected tax returns to determine the nature and extent of audits to be performed on them.
37| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Major tax authorities (IRS, HMRC) have piloted and deployed AI-assisted audit selection, but adoption remains primarily in screening and triage roles rather than full automation. Pilot projects are common; production deployment at scale is still limited and evolving. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are typically slow adopters of AI due to legacy systems, procurement cycles, oversight requirements, and public accountability concerns, despite some pilots in fraud detection. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools that surface high-risk patterns, compute audit-priority scores, and organize return data can meaningfully amplify examiner productivity by reducing manual search and comparison work. A human examiner using AI-assisted prioritization and risk flagging can process more returns with better targeting than manual review alone. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-driven analytics and anomaly detection substantially help examiners prioritize and scope audits faster by surfacing risk indicators, even though final decisions remain human-driven. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can assist with flagging returns for audit risk factors (unusual deductions, income mismatches) and categorizing complexity, but the task requires nuanced judgment about audit scope and strategy that depends on context, intent assessment, and organizational policy. Current systems cannot reliably perform the full end-to-end decision-making at 50% time savings without human review. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can flag anomalies, cross-check figures, and score returns for audit risk using pattern recognition, but final determination of audit scope requires judgment about legal nuance, taxpayer history, and discretion that current systems only partially replicate.ed |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax administration is heavily regulated and audits must be defensible and compliant with tax law. Liability for incorrect audit selection, taxpayer appeals, and legal exposure mean human experts typically retain final sign-off authority. Organizational culture and governance frameworks in revenue agencies also favor human accountability. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax audit determinations carry significant legal, due-process, and privacy implications, generally requiring an authorized government employee to make or approve audit-scope decisions, creating strong regulatory and liability barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Implementing and maintaining audit-selection AI systems has significant upfront cost and ongoing integration overhead. The marginal cost of human review remains low for moderate-volume operations, and oversight requirements mean AI handles only part of the workflow, making the all-in cost not substantially lower than human labor. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Automated screening is cheap per return compared to a human examiner's time, but building, validating, and maintaining compliant scoring models plus required human review keeps blended costs closer to parity than an order-of-magnitude gain. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Tax software and revenue agencies use rule-based systems and increasingly ML models to score audit risk and recommend return selection, but these are typically used as screening tools with human examiners making final decisions. No fully deployed system replaces the human reviewer's judgment on audit extent and strategy. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Tax agencies (e.g., IRS DIF scoring, fraud detection systems) already use statistical/ML models to triage returns, but these are narrow risk-scoring tools rather than full systems that determine audit nature and extent autonomously. |
Maintain knowledge of tax code changes, and of accounting procedures and theory to properly evaluate financial information.
37CI 32–41 · exposure 34 · augmentation 75 · importance 4.2/5 · click for rater detail
Maintain knowledge of tax code changes, and of accounting procedures and theory to properly evaluate financial information.
37| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tax and revenue agencies are moderately digitized and have pilot AI tools for research and code tracking, but actual deployment in production evaluations remains limited due to regulatory and liability constraints. Adoption is faster in back-office tax compliance than in revenue examination. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Government tax agencies are moderate adopters of AI, with growing use of research and compliance tools but still cautious, security-conscious rollout typical of public sector environments. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially assists tax examiners and revenue agents by automating research into tax code changes, summarizing accounting standards, and flagging relevant precedents and regulations—transforming the speed and comprehensiveness of knowledge maintenance while the human expert applies judgment and makes final evaluations. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools can efficiently digest and summarize new tax code changes, case law, and accounting standards, significantly speeding up an examiner's ability to stay current while the human validates and applies the knowledge. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Current AI can summarize and retrieve tax code changes and accounting standards, but applying this knowledge to evaluate complex financial information requires contextualized judgment, interpretation of ambiguity, and integration with specific client circumstances that AI systems cannot perform end-to-end with reliability. Human review remains essential. |
| Task automatability | claude-sonnet-5 | 2/5 | Staying current on tax code and accounting theory requires continuous learning and contextual judgment; AI can summarize updates but cannot autonomously 'maintain knowledge' in the way a professional internalizes and applies it to novel cases.rated conservatively. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax examination and revenue assessment functions are heavily regulated; human tax examiners and revenue agents must be licensed/credentialed, and liability for incorrect evaluations falls on the organization and credentialed human. Regulatory frameworks explicitly assign authority to human agents, creating hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No strict licensing requirement to 'stay informed,' but agencies often require certified training, continuing education, and internal sign-off processes that create moderate friction against pure automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While AI can reduce research time for tax code and accounting updates, the cost of maintaining and overseeing AI systems, integrating with existing tax evaluation workflows, and correcting errors is still comparable to or greater than the loaded cost of human researchers and analysts performing this task. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI subscription tools for tax/regulatory monitoring are cheaper than dedicating staff time to research, but human review and interpretation are still needed, keeping costs comparable rather than dramatically lower. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | AI systems (LLMs, specialized tax software with AI components) can assist with tracking regulatory changes and retrieving accounting procedures, but no deployed product reliably performs the full evaluation of financial information against tax code independently. Products exist in narrow domains but with material limitations in edge cases and complex scenarios. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like tax research assistants and legal/regulatory update tools exist and are used to surface changes, but they are supplementary rather than a replacement for the examiner's ongoing professional knowledge maintenance. |
Impose payment deadlines on delinquent taxpayers and monitor payments to ensure that deadlines are met.
34CI 29–40 · exposure 30 · augmentation 75 · importance 4.5/5 · click for rater detail
Impose payment deadlines on delinquent taxpayers and monitor payments to ensure that deadlines are met.
34| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Government tax agencies have adopted automated payment systems and reminders, but adoption is slower and more fragmented than private sector; many jurisdictions still rely on manual processes, and cultural/institutional conservatism in government limits rapid, agency-wide rollout. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI-driven decision systems due to legal, political, and due-process constraints, though some automated notice/payment tracking exists. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI systems can substantially assist human tax examiners by tracking payment status in real time, flagging delinquencies, and recommending deadline adjustments based on risk profiles, freeing agents to focus on complex cases and enforcement decisions while remaining in control. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-driven systems can significantly streamline tracking of deadlines, flagging delinquencies, and generating notices, improving efficiency for human agents managing large caseloads. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can easily generate payment deadlines and send automated reminders, the task requires human judgment on penalties, hardship exceptions, and negotiation with delinquent taxpayers—elements that demand legal and contextual discretion beyond current AI capabilities. |
| Task automatability | claude-sonnet-5 | 2/5 | Setting deadlines per policy and tracking payment status can be partially automated via workflow systems, but decisions on enforcement, negotiation, and exceptions still require human judgment, so full end-to-end automation with equal quality is not yet achievable. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax authorities operate under strict statutory deadlines and legal requirements; enforcement decisions often require documented human authority, audit trails, and potential legal defensibility that regulations typically vest in a licensed or credentialed official, creating compliance and liability friction. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax collection involves legal authority, due process requirements, and accountability for government action against individuals, creating substantial regulatory and liability barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Automated deadline imposition and payment monitoring are dramatically cheaper than employing human tax examiners for these clerical and tracking tasks, with inference and integration costs measured in cents per taxpayer versus loaded human wages in the $50k–$80k range annually. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Automated notice generation and payment monitoring are cheap, but case management, appeals, and legal escalation require human oversight, keeping overall cost roughly comparable when full task scope is considered. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Automated payment tracking and deadline notification systems exist and are deployed in tax agencies, but they handle only the mechanical parts; human revenue agents still make final determinations on enforcement actions and payment arrangements, limiting fully autonomous performance. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Tax agencies use automated reminder and payment-tracking systems, but imposing deadlines and escalation decisions on delinquent accounts typically still involve human agents/collectors making case-specific determinations. |
Confer with taxpayers or their representatives to discuss the issues, laws, and regulations involved in returns, and to resolve problems with returns.
25CI 25–25 · exposure 25 · augmentation 75 · importance 4.4/5 · click for rater detail
Confer with taxpayers or their representatives to discuss the issues, laws, and regulations involved in returns, and to resolve problems with returns.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Tax agencies and compliance departments have low historical adoption of autonomous AI agents in client-facing roles; pilots exist for document review and triage, but production displacement of conferral tasks remains minimal. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI for citizen-facing casework due to legal, privacy, and procedural constraints, despite some pilot chatbot deployments. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially assist by summarizing returns, flagging relevant regulations, suggesting precedent cases, and drafting talking points, meaningfully raising a tax professional's productivity while they retain control over negotiations and final decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist agents by summarizing tax law, drafting responses, and flagging relevant regulations during taxpayer conferences, improving speed and consistency while the human remains in charge. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can draft responses and summarize tax code, the core task requires live negotiation, judgment about taxpayer circumstances, and resolution of disputes—activities demanding real-time human interaction, empathy, and discretionary decision-making that current systems cannot perform end-to-end at the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 2/5 | This involves live, interactive negotiation with taxpayers requiring judgment, empathy, and case-specific legal interpretation, which current AI cannot fully replicate end-to-end despite handling FAQs or simple lookups.imum time saving does not meet the 50% threshold for the full task. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax determination and dispute resolution typically require a licensed or authorized representative; regulatory frameworks (IRS rules, state tax codes) impose requirements that a human agent be accountable for conclusions, creating hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax agencies typically require authorized personnel to make binding determinations and legally defensible resolutions, and due process/appeal rights create strong regulatory and liability barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The task requires specialized tax expertise, secure communication, and legal accountability—making AI integration costly relative to the loaded wage of a tax professional, especially when accounting for oversight, error liability, and the need for human sign-off. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI could cheaply handle routine inquiries, but complex conferences requiring human judgment and legal accountability still require costly human oversight, keeping the blended cost ratio closer to parity. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No production systems reliably conduct autonomous tax conferences with taxpayers; AI can assist with document analysis and code lookups, but deployed products do not independently resolve disputes or make binding determinations on tax issues. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Chatbots exist for basic tax FAQs but no deployed product reliably conducts substantive negotiations or dispute resolution conversations with taxpayers about return-specific legal issues. |
Investigate claims of inability to pay taxes by researching court information for the status of liens, mortgages, or financial statements, or by locating assets through third parties.
25CI 25–25 · exposure 25 · augmentation 63 · importance 4.0/5 · click for rater detail
Investigate claims of inability to pay taxes by researching court information for the status of liens, mortgages, or financial statements, or by locating assets through third parties.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Government tax agencies adopt technology slowly; legacy systems dominate, and investigative tasks remain heavily manual. Pilot AI projects exist, but production replacement of investigators is minimal and constrained by bureaucratic and legal friction. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI tools relative to private-sector finance and professional services, with pilots emerging but production-scale agentic investigation still rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by automating document retrieval, organizing court records, flagging financial anomalies, and summarizing asset searches, reducing manual drudgework. However, the core investigative and judgment tasks still require the tax examiner's expertise and sign-off. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully speed up research by scanning court records, public filings, and financial documents, helping examiners more efficiently narrow leads before making judgment calls and initiating third-party inquiries. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can assist with data retrieval and document analysis (court records, financial statements), but the investigative judgment—determining credibility of claims and synthesizing findings into a defensible conclusion—requires human expertise and contextual understanding of tax law. Current AI lacks reliable ability to assess claim legitimacy end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can help retrieve and summarize court/lien records and financial statements, but synthesizing this into an investigative judgment about ability to pay, including locating hidden assets via third parties, requires human investigative reasoning and follow-up actions AI cannot fully perform end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax examiners operate under IRS authority and legal standards; determinations of inability to pay have liability implications for the agency. Regulatory requirements, the need for human sign-off on investigative findings, and reliance on third-party cooperation create material barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax collection actions carry legal authority, due-process requirements, and accountability obligations that generally require a government employee to authorize and document findings, creating substantial procedural and legal barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Document retrieval and OCR are cheap, but the investigative labor—contacting third parties, validating information, cross-referencing legal records—still requires substantial human oversight, making all-in AI cost comparable to or higher than a tax examiner's loaded wage. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI-assisted document search can reduce research time cheaply, but the investigative and verification work still requires skilled human labor and access to non-digitized or restricted records, keeping overall costs comparable to or only modestly below human-only costs. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Products exist for document parsing and public records search, but no deployed end-to-end system reliably conducts tax inability investigations. Manual verification and legal oversight remain necessary, and integration with tax agency workflows is narrow. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some legal/financial research tools and data aggregators exist that can surface lien or asset records, but no deployed product reliably conducts full asset investigations and third-party inquiries autonomously in production tax agency workflows. |
Examine accounting systems and records to determine whether accounting methods used were appropriate and in compliance with statutory provisions.
25CI 25–25 · exposure 25 · augmentation 63 · importance 3.9/5 · click for rater detail
Examine accounting systems and records to determine whether accounting methods used were appropriate and in compliance with statutory provisions.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While government tax agencies and accounting firms are early in piloting AI-assisted review tools, production-scale automation of compliance determinations remains limited; most deployment remains in research or narrow pilot phases rather than broad organizational adoption. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI for substantive determinations, though risk-scoring and audit-selection tools are gradually being piloted. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by surfacing anomalies, extracting data from complex documents, and highlighting potential non-compliance areas, enabling human examiners to focus on judgment calls and statutory interpretation—a supportive but not transformative role. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist examiners by surfacing anomalies, cross-referencing records, and summarizing large accounting datasets, significantly speeding up the human review process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with data extraction and pattern detection in accounting records, the task requires nuanced judgment about the appropriateness of accounting methods and interpretation of complex statutory provisions—both of which remain outside current AI capabilities without substantial human review and decision-making. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can flag anomalies and check some compliance patterns in structured data, but judging whether an accounting method complies with statutory provisions requires nuanced legal-accounting interpretation and evidentiary reasoning that current systems cannot reliably complete end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax examination and compliance certification often require licensed tax professionals or government-credentialed revenue agents; liability for incorrect compliance determinations and regulatory requirements that a human must sign off on the examination create substantial legal and organizational barriers. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Formal determinations of compliance and enforcement actions typically require credentialed revenue agents/examiners with legal authority, creating strong institutional and liability barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-assisted data scanning and flagging may reduce search time, but the expert judgment required to validate findings and ensure statutory compliance still demands experienced human examiners, keeping total cost per task comparable to or higher than human-only approaches. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply screen large volumes of records, but the deep verification and judgment work still requires trained examiners, so blended cost savings are modest rather than order-of-magnitude. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs end-to-end compliance assessment and accounting method evaluation across varied systems and statutory frameworks; existing tools support document review but cannot independently verify appropriateness or render compliance determinations. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some tax authorities use analytics/ML tools to flag returns for audit, but no deployed product independently examines full accounting systems and renders compliance determinations at production scale. |
Examine and analyze tax assets and liabilities to determine resolution of delinquent tax problems.
23CI 20–25 · exposure 25 · augmentation 63 · importance 4.5/5 · click for rater detail
Examine and analyze tax assets and liabilities to determine resolution of delinquent tax problems.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Government tax agencies and revenue departments are historically slow adopters of cutting-edge AI, emphasizing audit trails, compliance, and human accountability. While some jurisdictions pilot automation for routine filing and simple compliance checks, production deployment of AI-driven resolution determination remains limited and cautious. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are typically slower adopters of AI due to legacy systems, procurement cycles, and compliance requirements, though some digitization pilots exist. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by extracting data, summarizing account histories, and highlighting relevant precedents or similar cases, raising analyst productivity in the data-gathering phase. However, the core task of determining resolution strategy still benefits significantly from human domain knowledge and negotiation judgment. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly assist by cross-referencing financial data, flagging inconsistencies, and drafting summaries, improving agent efficiency while humans retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can extract and organize tax data from documents and flag certain red flags in delinquent accounts, the determination of resolution strategies requires judgment about taxpayer circumstances, negotiation readiness, and legal precedent that current systems handle only partially. End-to-end automation with 50% time savings at equal quality is not demonstrated. |
| Task automatability | claude-sonnet-5 | 2/5 | Requires judgment on ambiguous financial records, negotiation strategy, and legal interpretation of tax code that current AI cannot reliably execute end-to-end without heavy human review. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Tax examination and revenue collection are heavily regulated by federal and state tax authorities, with strict procedural and documentation requirements. Revenue agents typically require state licensure or federal appointment, and the final determination of delinquent-account resolution must be signed by an authorized representative, creating hard legal barriers. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax determinations often carry legal and regulatory weight requiring accountable government employees, with formal appeals processes and liability concerns limiting full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-assisted document review and data extraction may reduce labor, but the oversight and verification of resolution determinations by tax professionals remains substantial. Full cost savings do not yet offset the human expertise required, keeping costs roughly comparable or favoring the human. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply flag anomalies, but the complex analysis and decision-making still requires skilled human agents, keeping overall cost comparable to human labor when oversight is included. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed products reliably perform the full task of analyzing tax assets/liabilities and determining delinquent-account resolution in production settings. Tax software handles compliance and calculation, but AI tools do not yet systematically advise on resolution strategies with acceptable accuracy for regulatory use. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some AI-assisted analytics tools exist in revenue agencies for flagging discrepancies, but full determination of delinquent case resolution is not performed autonomously by deployed products. |
Prepare briefs and assist in searching and seizing records to prepare charges and documentation for court cases.
23CI 20–25 · exposure 25 · augmentation 50 · importance 3.8/5 · click for rater detail
Prepare briefs and assist in searching and seizing records to prepare charges and documentation for court cases.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Tax agencies and legal departments have digitized slowly relative to other sectors and remain conservative with automation in enforcement actions due to due-process and accuracy requirements. Pilot projects exist, but production adoption of AI-driven brief preparation remains minimal. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax enforcement and legal sectors are historically slow adopters of AI for enforcement actions, with pilots limited mostly to document review rather than operational seizure/prosecution workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by organizing large document sets, flagging relevant records, and drafting initial documentation summaries that human agents then review and refine. This augmentation is useful but not transformative given the human judgment required throughout. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with drafting briefs, summarizing records, and organizing documentation, meaningfully speeding up parts of the task even though the core enforcement action remains human-led. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with searching records and organizing documentation, the task fundamentally requires legal judgment about what constitutes sufficient evidence for charges and proper seizure procedures. These require human legal expertise and discretionary decision-making that current AI cannot reliably perform end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | Legal brief drafting can be partly AI-assisted, but coordinating physical/electronic search-and-seizure actions and preparing court-admissible documentation requires legal judgment, chain-of-custody rigor, and physical enforcement actions that current AI cannot execute end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong legal barriers exist: licensed attorneys must prepare court briefs, and seizure of records is a regulated law enforcement action requiring human authorization and accountability. Liability exposure for improper seizure creates additional organizational friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Search and seizure actions and court case preparation involve statutory authority, due process, and legal accountability requiring a credentialed government agent/attorney, making this a hard-barrier task. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration costs for document management AI plus required human legal review, combined with liability overhead, likely approach or exceed the cost of a skilled human tax examiner performing the work directly, especially given the high stakes of error. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply draft text, but the overall task involves investigative fieldwork, legal review, and compliance oversight that still requires substantial paid human labor, keeping overall cost savings modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs the full scope of this task in production. Document search and organization tools exist, but no system can independently prepare legal briefs for court cases or make determinations about seizure legality without substantial human oversight and validation. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI drafting tools are used in legal contexts but no deployed product autonomously conducts seizures or assembles court-ready evidentiary packages reliably in production for tax enforcement. |
Collect taxes from individuals or businesses according to prescribed laws and regulations.
20CI 20–20 · exposure 25 · augmentation 63 · importance 4.8/5 · click for rater detail
Collect taxes from individuals or businesses according to prescribed laws and regulations.
20| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Tax agencies and collection bureaus operate with legacy systems and high organizational friction; adoption of AI agents for collection remains experimental. Few government or private collection firms have deployed production AI agents for core collection decisions. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are typically slow adopters of AI due to legacy systems, procurement cycles, and regulatory caution, though some automation of routine correspondence is emerging. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist tax examiners by automatically flagging high-risk accounts, summarizing documentation, and suggesting enforcement pathways, raising efficiency on data-heavy intake. However, the human must retain authority for all legally binding collection actions, limiting transformative potential. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly assist by flagging delinquent accounts, drafting correspondence, analyzing payment histories, and prioritizing cases, greatly improving agent efficiency while humans retain final authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Tax collection involves legal compliance, judgment calls on enforcement discretion, and adversarial negotiation that current AI cannot handle reliably end-to-end. While AI can automate document processing and straightforward case routing, the core task requires human authority to assess liability, issue legal notices, and negotiate settlements—preventing 50% time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Collecting taxes involves case management, negotiation with delinquent taxpayers, judgment calls on payment plans, and enforcement actions that require legal authority and human discretion, limiting full automation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Tax collection is heavily regulated and legally prescribed; only licensed or government-authorized revenue agents can issue assessments, liens, and collection actions. Liability for wrongful collection and statutory error-cost asymmetry create hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Tax collection is a government function requiring statutory authority; only authorized revenue agents can execute liens, levies, audits, and enforcement actions, creating hard legal barriers to full AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-assisted tax collection tools are costly to deploy and require significant human oversight (auditors, appeals specialists, legal review), making per-task cost comparable to or exceeding a tax examiner's loaded wage for complex cases requiring judgment. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While automating payment portals is cheap, the enforcement and judgment-heavy parts of collection still require costly human labor and legal oversight, keeping blended costs relatively high. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI products exist for tax data extraction and compliance flagging, but no deployed system reliably performs the actual collection task (determining enforcement action, validating disputes, executing legal remedies) in production. Pilot automation handles narrow, high-volume routine assessments only. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some automated payment processing and reminder systems exist, but actual collection actions (liens, levies, negotiations, legal enforcement) still require human agents in production government systems. |
Secure a taxpayer's agreement to discharge a tax assessment or submit contested determinations to other administrative or judicial conferees for appeals hearings.
20CI 20–20 · exposure 16 · augmentation 50 · importance 3.6/5 · click for rater detail
Secure a taxpayer's agreement to discharge a tax assessment or submit contested determinations to other administrative or judicial conferees for appeals hearings.
20| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Government tax agencies adopt AI slowly due to bureaucratic procurement cycles, legal conservatism, and public-sector IT constraints; current deployment is limited to supporting tools rather than autonomous negotiation agents. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI for taxpayer-facing negotiation and legal processes, with pilots limited mostly to document review and fraud detection rather than assessment resolution. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can usefully assist by summarizing case files, drafting settlement language, identifying precedent determinations, and organizing contested points for presentation to conferees, meaningfully raising an agent's productivity within a human-led negotiation process. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist by summarizing case history, drafting settlement language, and flagging precedent for appeals, meaningfully supporting the examiner though the human retains negotiation and judgment authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI could draft agreement templates and flag key contested issues, the task fundamentally requires negotiating with taxpayers and securing their actual agreement, which demands legal authority, judgment on settlement discretion, and real-time interpersonal negotiation that current AI systems cannot conduct independently. |
| Task automatability | claude-sonnet-5 | 2/5 | This task involves negotiation, judgment about legal merits, and interpersonal persuasion with taxpayers, which current AI cannot reliably execute end-to-end even with significant setup. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Tax examiners and revenue agents must be government employees or licensed professionals with statutory authority to bind the government and negotiate settlements; liability for improper settlements and regulatory oversight create hard barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Tax assessments and appeals are governed by administrative law requiring authorized government agents to negotiate settlements and represent the agency in judicial/administrative proceedings, creating strong legal and procedural barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI can assist with document preparation and case analysis at low cost, but the core negotiation and agreement-securing work must be performed by licensed revenue agents or attorneys, keeping per-task all-in costs comparable to or higher than human labor alone. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI could draft correspondence or summarize case files cheaply, the core negotiation and appeals decision-making still requires costly human oversight, keeping overall cost comparable to human-only workflows. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product today reliably performs independent taxpayer negotiation, legal settlement authority, or binding agreement execution; these tasks remain firmly within human legal and administrative domain in production systems. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously secures taxpayer agreements or manages appeals routing; this remains a human negotiation and legal process handled by trained agents. |
Direct service of legal documents, such as subpoenas, warrants, notices of assessment, and garnishments.
10CI 0–20 · exposure 8 · augmentation 38 · importance 4.3/5 · click for rater detail
Direct service of legal documents, such as subpoenas, warrants, notices of assessment, and garnishments.
10| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | This task involves mandatory legal compliance and physical presence; adoption of AI in service of legal documents is virtually nonexistent, as courts and revenue agencies rely on human process servers and licensed agents by law and necessity. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are typically slower adopters of AI for legally sensitive enforcement actions, with pilots limited mostly to document generation, not directing service. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist with scheduling optimization, locating addresses, or case management, but the critical act of service delivery remains fundamentally human; augmentation value is limited to support functions rather than the core task. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist by generating documents, tracking deadlines, and managing case files, but the core act of directing legal service still requires human oversight and decision-making. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Directing service of legal documents requires physical delivery, personal verification of receipt, and legal compliance with jurisdiction-specific rules that are context-dependent and require human judgment and presence. |
| Task automatability | claude-sonnet-5 | 2/5 | Directing physical/legal service of documents involves coordination, legal compliance, and often in-person or certified delivery that current AI cannot fully execute end-to-end.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Legal service of process is governed by strict statutes and court rules; only authorized process servers or government agents can lawfully execute service, and improper service is grounds for legal challenge and case dismissal. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Legal service of process is often governed by statute requiring specific procedures, authorized personnel, or certified delivery, creating strong regulatory and liability barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | This task inherently requires human agents (process servers or government employees) to travel, locate individuals, and execute service; AI has no cost advantage in performing the core requirement of in-person legal document delivery. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI could help draft or track documents, the actual directing and legal service process still requires human labor and process servers, limiting cost savings. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No AI system can perform the end-to-end legal service of documents, which requires physically locating individuals, verifying identity, confirming delivery, and often handling hostile or complex interpersonal situations that demand human judgment and authority. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product autonomously directs legal service of subpoenas, warrants, or garnishments; this remains a human administrative/legal function today. |
Participate in informal appeals hearings on contested cases from other agents.
10CI 0–20 · exposure 8 · augmentation 50 · importance 3.8/5 · click for rater detail
Participate in informal appeals hearings on contested cases from other agents.
10| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Tax administration remains heavily regulated and process-bound; appeals hearings are adversarial and legal in nature, sectors where automation is laggard and human professionals remain gatekeepers. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Government tax agencies are historically slow adopters of AI for adjudicative and interpersonal functions, with pilots focused on document review rather than hearing conduct. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist by pre-hearing document review or case preparation, but the hearing participation itself demands human judgment and presence; augmentation is limited to supporting activities rather than the core task. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can strongly assist by summarizing case history, prior rulings, and precedent, and drafting talking points, meaningfully speeding hearing preparation even though the hearing itself stays human-led. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires real-time judgment, legal reasoning, and adversarial interaction in a hearing context where human credibility and decision-making authority are essential. No current AI system can conduct or participate meaningfully in appeals hearings without a licensed human representative. |
| Task automatability | claude-sonnet-5 | 2/5 | This involves live hearing participation, judgment about disputed facts, negotiation, and representing agency positions in real time—AI can prepare briefing materials but cannot conduct the hearing itself end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Tax appeals hearings are regulated by statute; only licensed tax professionals can represent the government position, and the process requires human judgment, legal standing, and accountability that cannot be delegated to AI. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Appeals hearings typically require an authorized government representative with delegated authority to speak for the agency and make binding determinations, creating a strong institutional/legal barrier to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The task requires a skilled tax examiner or revenue agent (high-wage professional) to participate. AI assistance today does not replace the human participant and would add cost rather than reduce it. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI could cheaply assist with case file review and argument drafting, but the hearing itself still requires a paid human agent, so overall cost savings are limited to prep work. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product can independently conduct or meaningfully participate in formal or informal appeals hearings; such proceedings require human expertise, legal judgment, and the authority to make binding decisions or recommendations. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product conducts or substitutes for a human agent in informal appeals hearings; this remains a human-mediated adversarial/discretionary process. |
Related occupations — Business & Financial Operations
How to read this
A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.
What would change this score
New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.