Fraud Examiners, Investigators and Analysts

13-2099.04

Median wage $81,100/yr132,130 employed (US)Rank #352 of 923 scored · top 38% by substitution

Obtain evidence, take statements, produce reports, and testify to findings regarding resolution of fraud allegations. May coordinate fraud detection and prevention activities.

Sub-scores

0–100 · band = confidence interval from rater disagreement

Substitution32

Exposure30

Augmentation67

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

23 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

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%30

panel mean rating 2.2/5 → substitution pressure 30/100

Technical feasibility todayw 20%31

panel mean rating 2.2/5 → substitution pressure 31/100

Cost vs. human wagew 15%33

panel mean rating 2.3/5 → substitution pressure 33/100

Adoption barriersw 20%inverted — strong barriers lower the score32

panel mean rating 3.7/5 (barrier strength) → substitution pressure 32/100

Sector adoption velocityw 10%37

panel mean rating 2.5/5 → substitution pressure 37/100

Task breakdown (23 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.

Create and maintain logs, records, or databases of information about fraudulent activity.

CI 79–79 · exposure 75 · augmentation 75 · importance 4.6/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Financial services, insurance, and large retail firms are actively deploying AI-assisted case management and fraud logging systems; adoption is well underway in digitized, regulated sectors with high fraud volumes.
Sector adoption velocity	claude-sonnet-5	4/5	Fraud investigation sits within financial services and compliance functions, sectors with above-average, fast-moving AI and automation adoption for case management and data logging.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI can dramatically assist human examiners by auto-populating records, flagging related cases, cross-referencing patterns, and maintaining audit trails, allowing investigators to focus on analysis and decision-making rather than clerical work.
Augmentation potential	claude-sonnet-5	4/5	AI substantially speeds up and improves consistency of record creation and categorization, letting investigators focus on analysis while still reviewing/validating entries.
Task automatability	claude-haiku-4-5-20251001	4/5	Data entry, log structuring, and database updates for fraud records are highly automatable; AI can extract fraud signals from documents, categorize incidents, and populate structured records with minimal human intervention, achieving >50% time savings. However, the task requires judgment about what constitutes relevant 'fraudulent activity' that may warrant human review before logging, preventing a perfect 5.
Task automatability	claude-sonnet-5	4/5	Structured data entry, log maintenance, and database updates from case notes or reports are highly amenable to automation using current AI tools including document parsing and structured output generation.4/5 rather than 5/5 because occasional judgment on categorization/classification of ambiguous fraud types may still need human review.
Adoption barriers	claude-haiku-4-5-20251001	2/5	While fraud examiners often hold investigative certifications (CFE, etc.), the act of maintaining logs and records itself is not legally restricted to licensed individuals; organizations may prefer human oversight for high-stakes cases, but no hard regulatory barrier mandates it.
Adoption barriers	claude-sonnet-5	2/5	Some organizational and data-governance/compliance requirements (audit trails, data accuracy standards) create mild friction, but there is no licensing requirement mandating a human perform record-keeping itself.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	Automated logging and database maintenance via AI (including OCR, entity extraction, and structured data ingestion) cost a fraction of a human examiner's time to manually record and file each fraud case, easily a 10x+ cost advantage.
Cost vs. human wage	claude-sonnet-5	5/5	Automated data logging and database maintenance via software/AI is vastly cheaper than manual clerical entry by a trained fraud examiner, often reducing cost by an order of magnitude or more.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed products (document processing systems, case management platforms with AI enrichment, fraud detection tools) already perform record creation and database population at scale in financial and insurance sectors. Minor friction remains around edge cases and validation, but core functionality is production-ready.
Technical feasibility today	claude-sonnet-5	4/5	Deployed case-management and fraud-detection platforms (e.g., SAS, Actimize, ServiceNow-based case tools) already auto-populate and maintain structured fraud logs from alerts and case data in production environments today.

Maintain knowledge of current events and trends in such areas as money laundering and criminal tools and techniques.

CI 59–92 · exposure 75 · augmentation 100 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Financial services, compliance, and law enforcement organizations are actively deploying threat intelligence platforms and AI-driven news monitoring; adoption is accelerating in digitized sectors, though some smaller or traditional institutions lag.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and compliance sectors are moderately fast adopters of AI research tools, though full integration into fraud investigation workflows is still maturing.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI excels at keeping human fraud examiners informed by delivering curated, real-time alerts and synthesized summaries of emerging threats, patterns, and techniques—dramatically boosting an analyst's ability to stay current without the manual overhead.
Augmentation potential	claude-sonnet-5	5/5	AI tools significantly enhance an investigator's ability to stay current by aggregating, summarizing, and flagging relevant trends far faster than manual research alone.
Task automatability	claude-haiku-4-5-20251001	5/5	AI can continuously monitor, aggregate, and synthesize news, regulatory filings, academic papers, and threat intelligence from multiple sources to maintain current knowledge of fraud trends, money laundering techniques, and criminal methods—delivering structured summaries and alerts with minimal human oversight.
Task automatability	claude-sonnet-5	3/5	AI can summarize news, threat reports, and regulatory updates on money laundering trends, covering much of the information-gathering, but synthesis into professional judgment and knowledge retention still requires human curation.
Adoption barriers	claude-haiku-4-5-20251001	2/5	While organizations may prefer human analysts for strategic judgment and interpretation, there are no legal or licensing barriers preventing AI from performing the knowledge-maintenance function itself; adoption is mainly organizational preference and integration friction rather than hard regulatory gates.
Adoption barriers	claude-sonnet-5	2/5	No licensing requirement mandates a human perform this specific knowledge-maintenance task, though professional certification standards may implicitly expect continuing education by the individual.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	Automated monitoring via API-driven intelligence feeds and LLM processing costs substantially less than employing humans to manually read, track, and summarize evolving fraud and money laundering trends across multiple sources—typically orders of magnitude cheaper per maintained knowledge area.
Cost vs. human wage	claude-sonnet-5	4/5	AI-based summarization and monitoring tools are far cheaper than dedicating analyst hours to continuous manual research, though some oversight cost remains.
Technical feasibility today	claude-haiku-4-5-20251001	5/5	Deployed systems (OSINT platforms, threat intelligence aggregators, news monitoring services, and LLM-based summarization tools) already perform knowledge maintenance at scale in financial crime and cybersecurity organizations, reliably tracking emerging patterns and techniques.
Technical feasibility today	claude-sonnet-5	3/5	Products like news aggregators, AI research assistants, and specialized compliance intelligence platforms exist and are used, but they don't yet reliably replace expert curation of nuanced criminal typologies without human review.

Prepare written reports of investigation findings.

CI 50–85 · exposure 70 · augmentation 88 · importance 4.7/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Fraud and investigations operate in regulated financial and legal domains with slower digitization than tech; pilot adoption is visible but production displacement remains modest, with many organizations still using templated human-authored reports.
Sector adoption velocity	claude-sonnet-5	3/5	Fraud investigation sits within financial services and compliance, sectors with moderate-to-fast AI adoption for documentation tasks, but full workflow integration into investigative reporting is still emerging.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI dramatically accelerates report drafting, synthesis of evidence, and formatting while investigators focus on analysis and conclusions; this assistive use case is already widespread and measurably boosts analyst productivity.
Augmentation potential	claude-sonnet-5	4/5	AI substantially speeds up drafting, summarizing evidence, and organizing findings into report templates, meaningfully boosting investigator productivity while they retain final review and sign-off.
Task automatability	claude-haiku-4-5-20251001	5/5	Current AI systems can generate comprehensive written reports from investigation data, structured findings, and templates with minimal human intervention, easily achieving >50% time savings while maintaining quality through fact-checking and review frameworks.
Task automatability	claude-sonnet-5	3/5	AI can draft structured investigation reports from case notes and evidence summaries, but synthesizing findings, ensuring accuracy of legal/financial conclusions, and final judgment calls still require substantial human review, limiting full end-to-end automation.
Adoption barriers	claude-haiku-4-5-20251001	2/5	Some organizational friction exists around liability (who signs the report?) and regulatory expectations that a human reviewer approve findings, but no licensing requirement mandates a human author the initial draft, allowing AI-assisted workflows.
Adoption barriers	claude-sonnet-5	3/5	Reports often support legal or regulatory proceedings requiring signed attestation by a qualified examiner, creating moderate friction, though the drafting itself isn't strictly licensed work.
Cost vs. human wage	claude-haiku-4-5-20251001	5/5	AI-driven report generation costs orders of magnitude less than human analyst time once data is structured; a single LLM inference call costs cents while an analyst hour costs $50–150+ loaded.
Cost vs. human wage	claude-sonnet-5	3/5	AI drafting reduces drafting time significantly, but the need for human oversight, fact-checking, and liability review keeps overall cost roughly comparable rather than an order of magnitude cheaper.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Multiple deployed products (generative AI platforms, legal tech solutions, and investigation management systems) reliably generate professional investigation reports in production; minor limitations exist around highly nuanced conclusions but basic factual reporting is mature.
Technical feasibility today	claude-sonnet-5	3/5	Generative AI tools are used in production for drafting reports and summarizing evidence in legal/compliance workflows, but reliability on complex fraud narratives with citations and evidentiary chains still requires human verification and editing.

Gather financial documents related to investigations.

CI 50–70 · exposure 62 · augmentation 88 · importance 4.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Large law firms, forensic accounting practices, and financial institutions have rapidly adopted AI-powered e-discovery and document automation in investigation workflows, with widespread pilot and production deployment over the past 5 years. Smaller practices lag, but the trend is strong in information-intensive sectors.
Sector adoption velocity	claude-sonnet-5	3/5	Financial and legal investigative sectors are adopting AI-assisted document review and e-discovery tools at a moderate pace, with pilots and partial production use but not yet fully autonomous workflows.
Augmentation potential	claude-haiku-4-5-20251001	5/5	AI document gathering tools dramatically enhance investigator productivity by surfacing relevant documents, flagging anomalies, and organizing evidence hierarchically, allowing humans to focus on analysis and interpretation rather than manual collection and sorting. This augmentation is industry-standard in modern fraud investigation.
Augmentation potential	claude-sonnet-5	4/5	AI substantially speeds up locating, organizing, and flagging relevant financial documents, greatly aiding investigators even though final verification and legal gathering steps remain human-led.
Task automatability	claude-haiku-4-5-20251001	4/5	AI systems can reliably locate, retrieve, and organize financial documents from email, file servers, and databases with high consistency, achieving substantial time savings. However, some human judgment may still be needed for complex document sourcing across fragmented legacy systems or non-standard formats.
Task automatability	claude-sonnet-5	3/5	AI tools can search, request, and organize financial documents from databases and structured sources, but gathering often requires subpoenas, negotiating with institutions, or physical retrieval that AI cannot fully handle end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	3/5	Legal and compliance requirements around chain of custody, admissibility, and audit trails create some friction in automation; however, established e-discovery workflows already incorporate AI, and no licensing requirement mandates human-only document gathering. Organizational friction around trust in AI-selected documents remains moderate.
Adoption barriers	claude-sonnet-5	3/5	Gathering financial records often involves legal authority (subpoenas, court orders) and chain-of-custody requirements, creating moderate procedural and regulatory friction even though no license is strictly required to use AI tools.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	AI-driven document gathering (via e-discovery platforms and automated retrieval) costs a fraction of manual document collection by human investigators, with one-time platform investment amortized across many cases. Cost per document gathered is typically 10–50× cheaper than human manual review and retrieval.
Cost vs. human wage	claude-sonnet-5	3/5	AI can cut time spent on searching and compiling documents significantly, but human oversight, legal requests, and verification still add substantial cost, making the ratio moderate rather than dramatically cheaper.
Technical feasibility today	claude-haiku-4-5-20251001	4/5	Deployed e-discovery and document management platforms with AI-powered classification and retrieval are widely used in legal and forensic investigation contexts, with proven track records in production environments. Minor gaps remain in handling obscure file types or highly encrypted archives.
Technical feasibility today	claude-sonnet-5	3/5	Document management and e-discovery products, plus AI-driven data extraction tools, are deployed in fraud investigation workflows, but they typically assist rather than autonomously complete the full gathering process including legal requests and chain-of-custody.

Analyze financial data to detect irregularities in areas such as billing trends, financial relationships, and regulatory compliance procedures.

CI 53–56 · exposure 50 · augmentation 88 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Financial services, insurance, and government agencies have rapidly adopted AI-driven anomaly detection and compliance automation over the last 3–5 years, with many in production use. Adoption is driven by regulatory pressure, scale economics, and digitized datasets.
Sector adoption velocity	claude-sonnet-5	4/5	Financial services and insurance sectors are aggressive adopters of AI-based fraud analytics, with mature production deployments in transaction monitoring and compliance screening.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI excels at surfacing anomalies and flagging high-risk patterns at scale, substantially improving an analyst's ability to prioritize investigation and triage cases. Humans remain essential for judgment and validation, but AI transforms investigator productivity by filtering and ranking.
Augmentation potential	claude-sonnet-5	5/5	AI substantially enhances investigators' ability to surface irregularities across large datasets, prioritize leads, and detect patterns humans would miss manually, while examiners retain interpretive and decision-making control.
Task automatability	claude-haiku-4-5-20251001	3/5	AI can automate detection of anomalies in structured financial data (outliers, pattern deviations) and flag suspicious trends, but requires human expertise to contextualize findings, investigate root causes, and distinguish fraud from legitimate irregularities. This covers roughly half the analytical workload with significant setup for data integration and rule tuning.
Task automatability	claude-sonnet-5	3/5	AI/ML tools can flag statistical anomalies, unusual billing patterns, and outliers in structured financial data with significant time savings, but interpreting findings, establishing intent, and building a case still requires human judgment and contextual investigation.
Adoption barriers	claude-haiku-4-5-20251001	3/5	Regulatory frameworks (SOX, AML/KYC) may require human sign-off on fraud determinations and regulatory findings, and liability for false positives (customer false flags, regulatory missteps) creates audit and compliance friction. However, AI tools are widely permitted as screening and decision-support, not blocked by licensing.
Adoption barriers	claude-sonnet-5	3/5	No licensing requirement mandates a human perform data analysis itself, but regulatory reporting, legal admissibility of findings, and liability for wrongful accusations create meaningful oversight requirements.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	AI-powered anomaly detection and automated reporting systems cost substantially less than analysts per analysis cycle (inference + cloud infrastructure is cheap), though integration and tuning require upfront investment. At scale, the cost per detection is often 5–10× lower than manual review.
Cost vs. human wage	claude-sonnet-5	3/5	Automated anomaly detection is cheap per transaction at scale, but the need for skilled human review, tuning, and investigation of flagged cases keeps blended costs closer to parity for full task completion.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Deployed AI tools exist for financial anomaly detection and compliance monitoring in banking and insurance, but they operate with material false-positive rates and often require domain-specific configuration. Products work reliably on narrow, well-defined datasets but struggle with novel fraud patterns or complex regulatory contexts.
Technical feasibility today	claude-sonnet-5	3/5	Deployed fraud-detection platforms (transaction monitoring, anomaly detection systems) are widely used in banking and insurance, but they generate leads and alerts rather than completing the analysis independently, and false positive rates remain material.

Train others in fraud detection and prevention techniques.

CI 32–67 · exposure 45 · augmentation 75 · importance 3.9/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Financial services and compliance-heavy sectors are piloting AI-assisted training tools, but adoption remains in the early-to-middle phase. Legacy preferences for expert-led instruction and slower procurement in risk-sensitive organizations slow deployment.
Sector adoption velocity	claude-sonnet-5	3/5	Corporate learning and development functions in finance/professional services are adopting AI-assisted content generation moderately, though live instruction remains largely human-led.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI excels at generating training scenarios, automating quiz grading, personalizing learning paths, and providing case summaries, allowing human fraud experts to focus on interactive mentoring and judgment-based discussions. This significantly amplifies trainer productivity.
Augmentation potential	claude-sonnet-5	4/5	AI can significantly help trainers by generating case studies, quizzes, updated fraud typologies, and personalized learning materials, boosting productivity while the trainer still delivers instruction.
Task automatability	claude-haiku-4-5-20251001	4/5	AI can automate much of the content creation, case study generation, and delivery of standardized fraud detection training materials. However, interactive feedback, scenario discussion, and mentoring require human judgment, so full end-to-end replacement falls slightly short of the 50% threshold in all contexts.
Task automatability	claude-sonnet-5	2/5	Live training delivery involves adaptive facilitation, audience engagement, and Q&A that current AI cannot fully replicate end-to-end, though content creation portions could be automated.05
Adoption barriers	claude-haiku-4-5-20251001	2/5	While some organizations may prefer human-delivered training for compliance signoff, there is no legal mandate that training must be delivered by a human. Regulatory bodies generally accept documented, repeatable training regardless of delivery method.
Adoption barriers	claude-sonnet-5	3/5	No licensing requirement to train others, but organizational preference for experienced human trainers and accreditation bodies for certifications create moderate friction.
Cost vs. human wage	claude-haiku-4-5-20251001	4/5	AI-driven training platforms are substantially cheaper per learner than live instructor-led training once content is created. Overhead for platform maintenance is minimal compared to the loaded cost of expert fraud examiners delivering repeated training cohorts.
Cost vs. human wage	claude-sonnet-5	2/5	Developing customized AI training content still requires substantial human subject-matter input and review, so cost savings versus a trainer's loaded wage are modest, not order-of-magnitude.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	LLM-based training platforms and AI tutoring systems exist and perform competently in generating training content and quizzes, but they lack the nuance, real-time adaptation, and authority presence that organizations expect from fraud-specific training. Deployments are emerging but not yet mainstream in this domain.
Technical feasibility today	claude-sonnet-5	2/5	AI tools can generate training materials and simulate scenarios, but no deployed product autonomously delivers full fraud-training programs in production at scale.

Document all investigative activities.

CI 43–50 · exposure 50 · augmentation 75 · importance 4.6/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	While some larger law firms and corporate compliance teams are piloting AI-assisted documentation, the fraud investigation sector remains relatively conservative, with many smaller and mid-market investigation firms relying on manual processes; adoption in production is not yet widespread.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and corporate investigation units are increasingly adopting AI drafting and summarization tools, though broader fraud examination sectors adopt more slowly due to compliance concerns.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI can meaningfully assist investigators by auto-populating templates, flagging missing information, organizing evidence chronologically, and summarizing large document sets, substantially raising the productivity of a human who remains responsible for accuracy and legal compliance.
Augmentation potential	claude-sonnet-5	4/5	AI substantially speeds up documentation by auto-generating drafts, structuring findings, and organizing evidence trails, while investigators retain responsibility for accuracy and judgment calls.
Task automatability	claude-haiku-4-5-20251001	3/5	AI can automatically log and summarize investigative activities from unstructured sources (emails, notes, meeting transcripts) and generate structured documentation, but human judgment on what constitutes a complete, legally sufficient record and what details merit emphasis typically requires oversight, limiting time savings to roughly 50%.
Task automatability	claude-sonnet-5	3/5	AI can draft summaries, timelines, and structured reports from notes, transcripts, and case data, but ensuring completeness, accuracy, and legal defensibility of investigative records still requires human review and judgment.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Fraud investigation documentation often supports legal proceedings and regulatory filings, where a licensed investigator's or attorney's signature and judgment on completeness are typically required; many jurisdictions impose liability on the documented record, creating asymmetric error costs that deter full automation.
Adoption barriers	claude-sonnet-5	3/5	Investigative documentation often must meet evidentiary and regulatory standards, requiring an accountable human investigator to certify accuracy, creating moderate procedural barriers to full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	3/5	The all-in cost of AI document automation (inference, integration with case management systems, and required human oversight for legal sufficiency) approximates the cost of a junior investigator or paralegal performing this task, especially given liability concerns in fraud investigations.
Cost vs. human wage	claude-sonnet-5	3/5	AI can cut drafting time substantially, but the need for human verification of factual accuracy and chain-of-custody integrity keeps costs comparable rather than dramatically cheaper.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Products exist that can parse and organize investigative data and generate documentation summaries (e.g., legal AI platforms, case management tools with AI logging), but error rates in capturing critical details and the need for human review to ensure compliance with evidence standards mean production systems are not fully reliable without significant human validation.
Technical feasibility today	claude-sonnet-5	3/5	AI writing assistants and case-management tools with generative summarization are deployed in some fraud/compliance units, but reliability for legally sensitive documentation remains inconsistent and human editing is standard.

Review reports of suspected fraud to determine need for further investigation.

CI 29–53 · exposure 38 · augmentation 88 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	While financial services use AI for fraud scoring and alert triage, actual investigation initiation decisions remain human-controlled in most organizations; adoption of AI for this specific gate is still limited and cautious.
Sector adoption velocity	claude-sonnet-5	4/5	Financial services and insurance sectors are aggressive adopters of AI-based fraud detection and case triage tools, with mature deployment pipelines already in production.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI systems effectively assist fraud examiners by summarizing reports, ranking risk, flagging patterns, and prioritizing cases, allowing investigators to focus effort on substantive cases while maintaining human control over investigation decisions.
Augmentation potential	claude-sonnet-5	5/5	AI substantially improves the speed and accuracy of initial screening, flagging anomalies and prioritizing cases so investigators can focus attention more efficiently.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can flag high-risk patterns and summarize fraud reports, but determining investigation necessity requires judgment on liability, evidentiary standards, and organizational risk tolerance that remain difficult for AI to reliably assess at equal quality today.
Task automatability	claude-sonnet-5	3/5	AI can triage and score reports using pattern recognition and NLP to flag likely fraud, but determining true investigative need often requires contextual judgment, corroboration, and institutional knowledge that current systems only partially replicate.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Fraud investigation decisions often trigger legal, compliance, and regulatory consequences; financial institutions and law enforcement typically require licensed or authorized personnel to formally initiate investigations, creating both legal and reputational barriers to full automation.
Adoption barriers	claude-sonnet-5	3/5	No licensing mandate strictly requires human review, but liability, regulatory expectations for documented investigative judgment, and error-cost asymmetry create moderate friction against full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	3/5	ML-based anomaly detection and report summarization is becoming cost-comparable to manual review labor, but oversight and integration overhead for fraud decisions remain substantial.
Cost vs. human wage	claude-sonnet-5	3/5	AI-assisted screening reduces analyst hours per report, but licensing, model maintenance, and required human review keep costs closer to parity rather than order-of-magnitude cheaper for the full task.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	While some fraud-detection products exist, they primarily score risk or flag anomalies rather than making reliable end-to-end decisions about investigation initiation; deploying such decisions in production carries material liability and error costs.
Technical feasibility today	claude-sonnet-5	3/5	Fraud detection and case-prioritization tools are deployed in banking, insurance, and payments, but they typically assist human triage rather than autonomously deciding which cases warrant investigation.

Design, implement, or maintain fraud detection tools or procedures.

CI 32–40 · exposure 34 · augmentation 75 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	4/5	Financial services, e-commerce, and payment processors have invested heavily in AI-assisted fraud detection tools over the past 5–10 years; these sectors are digitized and competitive, driving faster experimentation, though full automation remains limited by regulatory and liability barriers.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and fraud units have moderate-to-strong AI adoption for detection algorithms, but the design/maintenance function itself still involves significant pilots and human-led iteration rather than full automation.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI systems excel at ingesting large transaction volumes, scoring risk, and surfacing high-confidence anomalies for investigation; human fraud examiners use these alerts to focus their investigation effort, substantially raising their productivity and reducing false-negative risk.
Augmentation potential	claude-sonnet-5	4/5	AI significantly assists in generating detection rules, anomaly scoring models, and pattern recognition, greatly boosting analyst and engineer productivity while humans retain design authority.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can assist in flagging suspicious patterns and generating rule candidates, designing and maintaining a complete fraud detection system requires domain expertise, legal/regulatory judgment, and ongoing human oversight of detection quality and false-positive tuning. This cannot achieve the 50% time-saving bar end-to-end today.
Task automatability	claude-sonnet-5	2/5	Designing and maintaining fraud detection systems requires domain judgment, regulatory knowledge, and iterative tuning against evolving fraud patterns that current AI cannot autonomously handle end-to-end.",
Adoption barriers	claude-haiku-4-5-20251001	4/5	Fraud detection and investigation operate under regulatory scrutiny (AML, KYC, BSA compliance), require human sign-off on cases with legal/liability implications, and financial institutions must verify decisions; liability for missed fraud is borne by the organization, creating strong incentive to retain human validation.
Adoption barriers	claude-sonnet-5	3/5	While no license is strictly required to build fraud tools, financial institutions face regulatory scrutiny and liability for detection failures, creating organizational caution around fully automating design decisions.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Building and maintaining a fraud detection system requires significant human expertise (data scientists, domain experts, compliance officers) and AI inference/infrastructure costs are non-trivial; the all-in cost is not yet cheaper than hiring experienced fraud investigators and analysts.
Cost vs. human wage	claude-sonnet-5	2/5	Building and maintaining robust fraud detection tools requires specialized data science, engineering, and compliance expertise, so AI reduces but doesn't drastically undercut the cost of skilled human labor.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	AI-powered fraud detection products exist and are deployed in banking and payment systems, but they typically flag cases for human review rather than make autonomous decisions; real production systems rely heavily on human analysts to refine rules and validate detections.
Technical feasibility today	claude-sonnet-5	3/5	Products like fraud analytics platforms (e.g., SAS, Feedzai) with ML components exist and are deployed, but designing and maintaining the overall detection strategy still requires significant human architecture and oversight.

Evaluate business operations to identify risk areas for fraud.

CI 28–37 · exposure 30 · augmentation 75 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Large financial institutions and compliance-heavy sectors pilot anomaly detection and transaction monitoring, but production deployment remains limited by regulatory requirements for human judgment and sign-off. Adoption is well underway in data triage phases but not in final determination.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and corporate risk functions are moderately fast adopters of AI-driven fraud detection tools, though full risk evaluation workflows still rely heavily on human analysts.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI significantly assists by automating data aggregation, highlighting anomalies, surfacing suspicious patterns across large datasets, and accelerating the identification of high-risk areas. Examiners can focus analytical effort on the most promising leads rather than manual screening, substantially raising throughput.
Augmentation potential	claude-sonnet-5	4/5	AI significantly enhances pattern detection, anomaly flagging, and data aggregation, substantially boosting the productivity of human fraud examiners performing this evaluation.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can flag anomalies and highlight suspicious patterns in financial data, but evaluating business operations for fraud requires contextual judgment about organizational culture, incentive structures, and nuanced risk factors that demand human expertise. Current systems lack the holistic understanding needed to achieve 50% time savings at equal quality.
Task automatability	claude-sonnet-5	2/5	Identifying fraud risk areas requires synthesizing operational context, judgment about intent, and organizational nuance that AI can support but not fully replace end-to-end today.assistant
Adoption barriers	claude-haiku-4-5-20251001	4/5	Regulatory requirements (SOX, financial compliance regimes) mandate documented investigation and sign-off by qualified personnel; liability and error-cost asymmetry are high because false negatives expose organizations to major penalties. Many regulators and boards require human-certified fraud assessments.
Adoption barriers	claude-sonnet-5	3/5	While no license is strictly required for risk assessment, organizational accountability, audit trail requirements, and liability for missed fraud create meaningful oversight and sign-off requirements.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI inference for anomaly detection is cheap, but integration into existing compliance workflows, model tuning, and the significant overhead of human review and investigation of false positives make the all-in cost comparable to or potentially higher than hiring skilled fraud examiners for many organizations.
Cost vs. human wage	claude-sonnet-5	2/5	AI tools can process transaction data cheaply, but the human-in-the-loop investigative work, interviews, and holistic risk assessment keep overall costs comparable to skilled human labor.
Technical feasibility today	claude-haiku-4-5-20251001	3/5	Products exist (e.g., anomaly detection engines, transaction monitoring systems) that identify suspicious patterns in structured data, but they produce material false-positive rates and miss sophisticated fraud schemes. Deployed systems operate within narrow scopes and require substantial human review.
Technical feasibility today	claude-sonnet-5	2/5	Some fraud analytics and risk-scoring products exist, but comprehensive evaluation of business operations for fraud risk is still largely human-led with AI as a supplementary tool rather than a reliable standalone product.

Conduct in-depth investigations of suspicious financial activity, such as suspected money-laundering efforts.

CI 28–35 · exposure 30 · augmentation 88 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	3/5	Financial institutions have adopted AI-assisted AML screening at significant scale (common in banking and fintech), but actual in-depth investigation and case closure remains human-led; adoption of AI-driven investigation tools themselves is still in pilot and early production phases across most sectors.
Sector adoption velocity	claude-sonnet-5	4/5	Financial services is a fast-adopting sector for AI-driven fraud detection and AML tools, with widespread production deployment of monitoring and analytics systems industry-wide.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI significantly augments human investigators by rapidly scanning large transaction volumes, surfacing high-risk patterns, cross-referencing databases, and highlighting anomalies that would take humans weeks to identify—substantially raising investigation productivity while the analyst maintains control and judgment.
Augmentation potential	claude-sonnet-5	5/5	AI substantially augments investigators by surfacing patterns, flagging anomalies, and summarizing large transaction datasets, greatly increasing productivity while humans retain judgment and decision authority.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can assist in pattern detection and flagging suspicious transactions at scale, but investigating money-laundering requires understanding complex intent, jurisdiction-specific legal nuance, and integration of disparate evidence sources—tasks that currently demand human judgment and cannot achieve 50% time savings end-to-end with current systems.
Task automatability	claude-sonnet-5	2/5	AI can accelerate transaction analysis and pattern detection but the full in-depth investigation involves judgment, interviews, legal interpretation, and case-building that current systems cannot autonomously complete end-to-end with equal quality.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Regulatory bodies (FinCEN, FATF, national regulators) legally require qualified human investigators to sign off on AML findings; many jurisdictions mandate that suspicious activity reports (SARs) carry human accountability, creating hard barriers to full automation.
Adoption barriers	claude-sonnet-5	4/5	Regulatory frameworks (BSA/AML, SAR filing) require qualified human judgment and accountability for investigative conclusions and legal reporting, creating strong compliance and liability barriers to full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI screening and monitoring tools are relatively inexpensive, but the cost of oversight by skilled investigators, combined with integration and tuning effort, remains comparable to or sometimes exceeds the cost of direct human investigation in small to mid-sized institutions.
Cost vs. human wage	claude-sonnet-5	2/5	AI screening tools reduce initial triage costs substantially, but the full investigative process still requires skilled human investigators, keeping overall cost comparable to human-led work once oversight and case development are included.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	Anti-money-laundering (AML) detection products exist and flag suspicious activity, but they generate high false-positive rates and rely on human analysts to conduct the actual in-depth investigation; no deployed product independently performs full investigations reliably without expert human oversight.
Technical feasibility today	claude-sonnet-5	3/5	AML transaction monitoring and anomaly detection tools are widely deployed in production, but they flag suspicious activity rather than conduct complete investigations, and error rates (false positives) remain material.

Advise businesses or agencies on ways to improve fraud detection.

CI 25–32 · exposure 25 · augmentation 75 · importance 4.0/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Adoption of AI-driven fraud detection is concentrated in large financial institutions and tech firms; most SMEs and agencies rely on traditional investigator-led assessment. Actual displacement of advisory roles remains limited despite emerging analytics tools, reflecting organizational preference for human expert judgment on sensitive compliance matters.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and fraud analytics are adopting AI tools for pattern detection and reporting assistance at a moderate pace, though full advisory automation remains uncommon.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI tools effectively augment fraud examiners by accelerating pattern detection, surfacing anomalies across large datasets, and suggesting hypotheses for investigation—allowing human experts to focus on high-value judgment calls, client communication, and strategic design rather than manual data sifting.
Augmentation potential	claude-sonnet-5	4/5	AI significantly aids fraud examiners by identifying data patterns, benchmarking controls, and drafting recommendations, meaningfully boosting the human advisor's productivity.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can help generate detection rule suggestions and analyze patterns in historical fraud data, but advising on systemic improvements requires understanding organizational context, risk appetite, and operational constraints that current AI systems cannot reliably assess end-to-end. Human expertise in interpreting business-specific vulnerabilities and designing practical countermeasures remains essential.
Task automatability	claude-sonnet-5	2/5	Advising requires synthesizing organizational context, risk appetite, and judgment about tradeoffs that AI cannot fully replicate end-to-end; AI can draft recommendations but a human must tailor and validate them.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Significant barriers exist: fraud advice often requires sign-off by licensed professionals (CFEs, compliance officers) to establish liability and credibility; regulatory frameworks in financial services demand human accountability for fraud controls; clients typically prefer human expertise for strategic guidance that carries legal and operational weight.
Adoption barriers	claude-sonnet-5	3/5	No strict licensing mandate for this advisory role, but liability concerns, client trust, and need for contextual judgment create moderate friction against pure AI substitution.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI infrastructure for data processing and analysis is relatively affordable, but advisory work requires human expert time for client engagement, contextual interpretation, and sign-off. The integrated cost of human-AI advisory remains comparable to or exceeds hiring a qualified fraud consultant.
Cost vs. human wage	claude-sonnet-5	2/5	While AI can cheaply produce draft reports or analysis, the consulting/advisory value requires expert review and customization, keeping all-in cost closer to human-comparable levels.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	While AI tools can assist with data analysis and generate recommendations, no deployed product reliably advises businesses on fraud detection improvement as a standalone end-to-end task. Existing systems produce analysis fragments (anomaly detection, pattern reports) that still require human integration into strategic advisory output.
Technical feasibility today	claude-sonnet-5	2/5	AI tools can generate generic fraud-control recommendations and analyze data patterns, but no deployed product independently advises organizations on fraud detection improvements without expert oversight.

Research or evaluate new technologies for use in fraud detection systems.

CI 25–32 · exposure 25 · augmentation 75 · importance 3.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Fraud detection organizations are cautious and regulated; while they pilot new tech, adoption of AI-driven technology evaluation itself is slow and limited to larger financial institutions and specialized consultancies, not the majority of fraud-examination functions.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and fraud units are moderately fast adopters of AI tools for research support, though full evaluation workflows remain human-led with pilots more common than mature deployment.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI can meaningfully assist by summarizing vendor documentation, identifying relevant academic research, and organizing comparative feature matrices, allowing examiners to focus on strategic fit and risk assessment rather than raw information gathering.
Augmentation potential	claude-sonnet-5	4/5	AI can significantly speed up literature review, competitive analysis, and technology scanning, giving fraud analysts strong productivity gains while they retain final evaluative judgment.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can help scan technical literature and summarize vendor capabilities, the core task requires human judgment to evaluate whether new technologies align with organizational fraud-detection needs, security posture, and risk tolerance. Current AI cannot independently assess suitability or conduct meaningful due diligence.
Task automatability	claude-sonnet-5	2/5	This requires open-ended judgment about emerging technologies, vendor evaluation, and strategic fit that current AI can support with research but not fully perform end-to-end.provide independently.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Strong barriers exist: fraud examiners often require specific certifications (CFE, etc.), organizational risk tolerance dictates technology adoption decisions, and liability concerns mean senior staff must validate technology recommendations before deployment.
Adoption barriers	claude-sonnet-5	3/5	No strict licensing requirement, but organizational risk tolerance and need for accountable decision-making on security-critical systems creates moderate friction against pure AI-driven evaluation.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI-assisted research tools cost money and require expert human review of outputs, making the all-in cost closer to or potentially exceeding what a skilled fraud examiner charges for the same evaluation work, especially given required oversight.
Cost vs. human wage	claude-sonnet-5	2/5	Human expert evaluation involves domain judgment, stakeholder input, and risk assessment that AI cannot cheaply replicate without significant oversight, keeping costs comparable to skilled analyst time.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No deployed product reliably performs end-to-end technology evaluation for fraud detection. Benchmarking tools and literature review assistants exist, but production systems for systematic technology assessment and recommendation remain limited and require significant human oversight.
Technical feasibility today	claude-sonnet-5	2/5	AI tools can summarize research and compare vendor claims, but no deployed product autonomously evaluates and selects fraud-detection technologies for organizations today.

Recommend actions in fraud cases.

CI 25–28 · exposure 25 · augmentation 63 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Fraud detection and investigation remain heavily regulated and risk-averse domains. While pattern-detection tools see adoption, recommendation and action-planning remain human-driven in most financial services and government agencies; adoption of autonomous recommendation systems is nascent.
Sector adoption velocity	claude-sonnet-5	3/5	Financial services and insurance sectors are adopting AI fraud tools quickly, but recommendation generation specifically remains in pilot or decision-support stages rather than full production autonomy.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can usefully assist by summarizing case facts, flagging similar historical patterns, and scoring risk indicators, helping investigators work faster. However, the core recommendation—what enforcement action to pursue—still centers on human judgment, making augmentation meaningful but not transformative.
Augmentation potential	claude-sonnet-5	4/5	AI substantially aids investigators by surfacing patterns, summarizing case files, and suggesting possible actions, meaningfully speeding up the recommendation process while humans retain final judgment.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can analyze fraud patterns and flag suspicious transactions, recommending case actions requires legal judgment, risk assessment, and contextual decision-making tied to investigation outcomes and organizational policy. Current systems can assist in evidence synthesis but cannot independently recommend full action paths with the reliability and accountability required.
Task automatability	claude-sonnet-5	2/5	AI can synthesize case evidence and draft recommendations, but final action recommendations require contextual judgment about legal, business, and human factors that current systems cannot reliably weigh end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Significant regulatory and legal barriers exist: fraud investigations often involve law enforcement partnerships, evidence chain-of-custody requirements, potential criminal prosecution, and liability asymmetry if false recommendations lead to wrongful prosecution or missed enforcement. Organizations require human sign-off on case actions.
Adoption barriers	claude-sonnet-5	4/5	Fraud investigations often have legal and regulatory implications, requiring accountable human judgment and sign-off, creating strong liability-driven barriers to full automation.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	Fraud investigators command high salaries (~$70k–$100k+ loaded), and building reliable recommendation systems with compliance oversight is expensive. Current AI infrastructure (model inference, data integration, legal review overhead) does not achieve cost parity with human investigators for this judgment-heavy task.
Cost vs. human wage	claude-sonnet-5	2/5	AI can cheaply summarize data, but the oversight, verification, and liability review needed for recommendations keeps overall cost closer to human-level rather than an order of magnitude cheaper.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No mature deployed product reliably recommends fraud case actions end-to-end in production at scale. Tools exist for pattern detection and alert generation, but production fraud recommendation systems require human oversight and domain expertise, limiting autonomous deployment.
Technical feasibility today	claude-sonnet-5	2/5	Deployed fraud analytics tools flag anomalies and risk scores, but few products autonomously generate case-specific action recommendations that examiners rely on without heavy review.

Prepare evidence for presentation in court.

CI 20–20 · exposure 25 · augmentation 63 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Adoption is slow in fraud investigation and law enforcement sectors, which remain highly regulated and resistant to full automation of evidence preparation. While some law firms pilot AI-assisted discovery, production-scale adoption of autonomous evidence preparation for court is rare and limited to narrow document-review phases.
Sector adoption velocity	claude-sonnet-5	2/5	Legal and investigative sectors are historically slow adopters of AI for high-stakes, liability-sensitive tasks, with pilots more common than production deployment for court-facing work.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can meaningfully assist by organizing large document sets, flagging relevant evidence, generating summaries, and organizing materials—raising investigator productivity in the collation and indexing phases. However, the human investigator and legal team must remain central to the final certification and presentation decision-making.
Augmentation potential	claude-sonnet-5	4/5	AI tools substantially assist with organizing documents, searching records, summarizing findings, and drafting reports, meaningfully boosting investigator productivity even though a human must finalize and validate court-ready evidence.
Task automatability	claude-haiku-4-5-20251001	2/5	AI can assist with evidence organization and document review, but preparing court-admissible evidence requires legal judgment, chain-of-custody verification, and compliance with evidentiary rules that demand human expertise and accountability. Partial automation of evidence collation is possible, but the full task cannot meet the 50% time-saving bar without substantial human oversight.
Task automatability	claude-sonnet-5	2/5	AI can help organize and draft summaries of evidence, but preparing legally admissible evidence for court requires judgment about admissibility, chain of custody, and legal strategy that current AI cannot reliably handle end-to-end.'
Adoption barriers	claude-haiku-4-5-20251001	5/5	Strong legal and professional barriers protect this task: the attorney preparing evidence for trial must certify its proper handling, rules of evidence impose strict requirements on how evidence is presented, and courtroom procedure mandates human expert testimony and accountability. Regulatory and liability frameworks effectively require licensed human judgment.
Adoption barriers	claude-sonnet-5	5/5	Court evidence preparation is subject to strict legal rules on admissibility, chain of custody, and often requires sworn testimony or certification by licensed professionals, creating hard regulatory and liability barriers.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI-assisted document review and organization can reduce some labor, but the expertise required to verify admissibility, ensure proper documentation, and manage legal liability means human investigators remain the primary cost driver. Full-service legal compliance oversight remains more expensive than the investigator labor saved.
Cost vs. human wage	claude-sonnet-5	2/5	AI can cut costs on document review/organization portions, but the overall task still requires expensive human legal and investigative oversight, keeping total cost comparable to human-only processes.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	While AI tools exist for document review and summarization, no deployed product reliably handles the end-to-end task of preparing legally compliant, court-ready evidence packages that meet jurisdiction-specific evidentiary standards and withstand adversarial scrutiny. Error rates remain material for high-stakes legal applications.
Technical feasibility today	claude-sonnet-5	2/5	Some e-discovery and document review tools are deployed in legal contexts, but full evidence preparation for court presentation still requires human legal expertise and is not a mature standalone AI product function.

Lead, or participate in, fraud investigation teams.

CI 7–25 · exposure 13 · augmentation 75 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	2/5	Financial services and law enforcement have slow, cautious adoption of AI in investigative roles; pilots are common but production deployment of AI as team participants or leaders remains rare. Regulatory, legal, and organizational risk aversion limit velocity.
Sector adoption velocity	claude-sonnet-5	2/5	Fraud investigation units in finance/insurance are adopting AI for detection and analytics, but adoption of AI into team leadership or investigative coordination roles remains minimal.
Augmentation potential	claude-haiku-4-5-20251001	4/5	AI strongly augments investigation teams by automating data triage, flagging anomalies, organizing evidence, and accelerating pattern discovery. Investigators using modern AI tools can process far larger datasets and refine hypotheses faster, though humans remain essential for judgment and direction.
Augmentation potential	claude-sonnet-5	4/5	AI tools significantly help investigation teams by flagging anomalies, summarizing evidence, and drafting reports, boosting the productivity of the humans who lead and staff these teams.
Task automatability	claude-haiku-4-5-20251001	2/5	While AI can assist with data analysis, pattern detection, and case organization, leading or meaningfully participating in investigation teams requires complex judgment, stakeholder management, cross-functional coordination, and real-time responsiveness that current systems cannot reliably perform end-to-end. AI cannot replace the investigator's role in team dynamics, decision-making, and accountability.
Task automatability	claude-sonnet-5	1/5	Leading or participating in an investigation team requires human coordination, judgment, interpersonal management, and accountability that current AI cannot perform end-to-end.leading a team is inherently a human management function.
Adoption barriers	claude-haiku-4-5-20251001	4/5	Fraud investigations often involve legal authority, evidence admissibility, regulatory compliance (SarbOx, banking regs, criminal procedures), and liability for errors. Many jurisdictions require a qualified human investigator to lead investigations and attest to findings, creating hard barriers to full automation.
Adoption barriers	claude-sonnet-5	4/5	Fraud investigations often involve legal proceedings, chain-of-custody, licensing (e.g., CFE credentials), and liability concerns requiring accountable human leadership and testimony.
Cost vs. human wage	claude-haiku-4-5-20251001	2/5	AI tools for fraud analysis have meaningful infrastructure and integration costs, while team leadership and participation require significant human oversight, judgment, and liability assumption. The all-in cost remains comparable to or higher than a skilled human investigator for this complex coordination task.
Cost vs. human wage	claude-sonnet-5	1/5	There is no AI substitute performing this managerial/collaborative role, so no meaningful cost comparison of AI replacing the human function exists.
Technical feasibility today	claude-haiku-4-5-20251001	2/5	No deployed product reliably leads or participates in fraud investigation teams as a standalone agent. AI tools exist for evidence analysis and anomaly detection, but they operate as narrow inputs to human-led processes rather than as functional team participants or leaders in production settings.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product leads or manages fraud investigation teams; AI tools at best support individual analytic subtasks, not team leadership or participation as a role.

Negotiate with responsible parties to arrange for recovery of losses due to fraud.

CI 0–20 · exposure 8 · augmentation 50 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Fraud investigation remains a compliance and legal-sensitive function in heavily regulated sectors; adoption of autonomous negotiation AI is minimal. Organizations continue to rely on human investigators for authoritative settlement discussions.
Sector adoption velocity	claude-sonnet-5	2/5	Fraud investigation functions in finance are adopting AI for detection and analytics, but the negotiation/recovery subtask itself sees little to no AI deployment.
Augmentation potential	claude-haiku-4-5-20251001	3/5	AI can assist by analyzing settlement data, drafting proposal language, and identifying leverage points, but the human investigator remains the primary negotiator. These tools offer moderate productivity gains without replacing human judgment and authority.
Augmentation potential	claude-sonnet-5	3/5	AI can help by summarizing case files, drafting settlement proposals, or suggesting negotiation strategies, but the negotiation execution remains human-led.
Task automatability	claude-haiku-4-5-20251001	2/5	Negotiation inherently requires judgment, relationship-building, and understanding of nuanced human interests that current AI systems cannot reliably execute end-to-end. While AI could draft communications or suggest settlement positions, the persuasion, authority, and legal liability of actual negotiation remain firmly human domains.
Task automatability	claude-sonnet-5	1/5	Negotiation with responsible parties requires real-time interpersonal judgment, persuasion, and adaptive strategy that current AI cannot autonomously execute end-to-end for legal/financial settlements.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Negotiation on fraud recovery involves legal authority, contractual commitment, and liability that generally require a licensed investigator or attorney to conduct or sign off on. Regulatory and organizational barriers to autonomous AI negotiation are substantial.
Adoption barriers	claude-sonnet-5	4/5	Negotiating recovery of fraud losses often involves legal liability, contractual authority, and sometimes licensed professionals (attorneys, certified fraud examiners) whose judgment and accountability cannot be delegated to AI.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	The AI cost of generating negotiation scripts or recommendations would not achieve sufficient time savings to undercut the loaded cost of a trained fraud examiner who must oversee and be liable for the outcome.
Cost vs. human wage	claude-sonnet-5	2/5	While AI could support drafting or case analysis cheaply, the negotiation itself still requires a skilled human negotiator, so overall cost savings are minimal to none.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed product performs autonomous negotiation with responsible parties on fraud recovery today. Current AI can assist with document drafting or analysis, but production systems require human negotiators to conduct the actual negotiation.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product conducts autonomous fraud-loss recovery negotiations with responsible parties; this remains firmly in the human domain of legal/financial dispute resolution.

Interview witnesses or suspects and take statements.

CI 0–11 · exposure 0 · augmentation 50 · importance 4.7/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Investigation and law enforcement agencies operate under strict evidentiary and chain-of-custody rules; adoption of autonomous interviewing systems is virtually non-existent because legal and institutional requirements mandate human conduct and accountability.
Sector adoption velocity	claude-sonnet-5	2/5	Fraud investigation units are adopting AI for data analysis and documentation but have been slow to change core interview practices due to legal and procedural constraints.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI can assist marginally by summarizing prior statements, flagging inconsistencies in transcripts, or suggesting follow-up questions, but the human investigator remains entirely responsible for conducting the interview itself. The augmentation value is limited.
Augmentation potential	claude-sonnet-5	4/5	AI tools can transcribe, summarize, flag inconsistencies, and help prepare questions, meaningfully boosting investigator efficiency even though humans conduct the interview itself.
Task automatability	claude-haiku-4-5-20251001	1/5	Interviewing witnesses or suspects requires real-time social judgment, rapport-building, credibility assessment, and adaptive questioning based on subtle cues—capabilities that current AI systems cannot perform reliably end-to-end. While AI can draft interview templates or transcribe, it cannot conduct the interview itself or replace the human investigator's critical role.
Task automatability	claude-sonnet-5	1/5	Interviewing requires reading human demeanor, adapting questioning strategy in real time, building rapport, and exercising judgment about credibility—capabilities current AI cannot reliably perform end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Jurisdictional law and investigative standards require a human investigator to conduct interviews; statements must be taken by or under the direct supervision of a licensed investigator or law enforcement officer to be admissible. This is a hard legal and procedural barrier.
Adoption barriers	claude-sonnet-5	4/5	Legal admissibility, chain-of-custody, and procedural/evidentiary rules typically require a qualified human investigator to conduct and attest to interviews, creating strong institutional and legal barriers.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	The cost of a failed or unusable interview (legal liability, case dismissal, warrant rejection) far exceeds savings from any AI system; human investigator wages are the lower-cost option when accounting for legal and evidentiary requirements.
Cost vs. human wage	claude-sonnet-5	2/5	While AI transcription is cheap, the actual interviewing and judgment work must still be done by a trained human investigator, so overall cost savings are minimal.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No production systems exist that independently conduct witness or suspect interviews at legal or investigative standard. AI chatbots cannot establish trust, read deception, adapt to evasion, or produce legally admissible statements without human interviewer control and presence.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product conducts investigative interviews or takes formal witness/suspect statements autonomously; this remains a human-led activity with AI only assisting in transcription or note-taking.

Coordinate investigative efforts with law enforcement officers and attorneys.

CI 0–7 · exposure 0 · augmentation 38 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Law enforcement and legal sectors remain highly regulated and resistant to AI-driven autonomous coordination; adoption of AI in investigative workflows remains minimal and heavily overseen by humans.
Sector adoption velocity	claude-sonnet-5	2/5	Fraud investigation units use some AI tools for data analysis, but the interpersonal coordination with external legal/law enforcement bodies sees minimal AI adoption.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI could assist by scheduling meetings, summarizing case details, or flagging document inconsistencies, but coordination inherently requires human judgment and relationship management that AI cannot substantially elevate.
Augmentation potential	claude-sonnet-5	3/5	AI can help draft case summaries, track communications, and organize evidence to support coordination, but the core liaison work is human-driven.
Task automatability	claude-haiku-4-5-20251001	1/5	Coordinating investigative efforts requires real-time negotiation, relationship-building, and complex decision-making with human stakeholders (law enforcement, attorneys). Current AI cannot meaningfully participate in or replace the strategic and interpersonal aspects of such coordination.
Task automatability	claude-sonnet-5	1/5	Coordinating with law enforcement and attorneys requires relationship-building, negotiation, judgment about legal strategy, and real-time interpersonal interaction that current AI cannot perform end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Legal and regulatory frameworks require human investigators and attorneys to take direct responsibility for coordination decisions. Authority, liability, and legal standing cannot be delegated to or mediated solely by AI systems.
Adoption barriers	claude-sonnet-5	4/5	Legal privilege, chain-of-custody requirements, and law enforcement protocols mean human professionals with authority and accountability must handle this liaison work.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	AI coordination tools, if they existed at scale, would require significant human oversight and would not reduce the need for the fraud examiner's involvement, making them more expensive than direct human coordination.
Cost vs. human wage	claude-sonnet-5	1/5	There is no viable AI substitute performing this coordination role, so cost comparison favors the human entirely.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed product reliably orchestrates cross-agency investigative coordination. This fundamentally requires human judgment, accountability, and legal authority that AI systems cannot exercise independently.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product manages cross-organizational coordination with external legal and law enforcement parties; this remains a human relational and procedural function.

Conduct field surveillance to gather case-related information.

CI 0–5 · exposure 0 · augmentation 38 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Fraud investigation remains a human-centered, legally bounded profession; there is no measurable adoption of autonomous AI agents for field surveillance in production environments, as regulatory and liability barriers prevent deployment.
Sector adoption velocity	claude-sonnet-5	1/5	Field investigation work remains a low-digitization, physical-world task with minimal AI agent deployment in this specific surveillance function.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI can assist with post-surveillance analysis (video review, pattern detection in footage) but offers minimal augmentation to the core task of conducting field surveillance itself, which fundamentally requires human judgment, presence, and legal authority in the physical world.
Augmentation potential	claude-sonnet-5	3/5	AI can assist with route planning, license plate recognition, video/image analysis, and organizing surveillance data afterward, but the on-the-ground observation task itself is not meaningfully augmented in real time.
Task automatability	claude-haiku-4-5-20251001	1/5	Field surveillance requires sustained physical presence in real environments, adaptive decision-making in response to unpredictable human behavior, and real-time judgment about what constitutes relevant evidence—capabilities far beyond current AI systems. Current AI cannot legally or practically conduct outdoor surveillance operations.
Task automatability	claude-sonnet-5	1/5	Field surveillance requires physical presence, mobile observation, discretion, and real-time judgment calls that current AI systems cannot perform end-to-end; there is no substitute for a human physically tailing or observing a subject.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Field surveillance is heavily regulated by privacy laws, trespassing statutes, and investigative licensing requirements; a licensed investigator must typically conduct or directly supervise the surveillance, and evidence chain-of-custody requirements legally bind the human to the investigation.
Adoption barriers	claude-sonnet-5	4/5	Surveillance often has legal requirements (licensing for private investigators, evidentiary chain-of-custody, privacy law compliance) that necessitate an accountable human actor performing or supervising the activity.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	Field surveillance demands human presence, decision-making, and accountability; autonomous AI surveillance systems capable of gathering legally admissible evidence would require significant custom development and oversight infrastructure, making total cost per case higher than human investigator labor.
Cost vs. human wage	claude-sonnet-5	1/5	AI cannot perform the core physical task, so any 'AI cost' would require extensive human-operated hardware and oversight, making it more expensive than simply using a human investigator.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed AI product performs autonomous field surveillance; this task requires autonomous agents with legal authority to conduct investigative observation, which does not exist in production. Research prototypes for video analysis exist but cannot replace the human investigator's presence and judgment in the field.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product performs autonomous physical field surveillance for fraud investigations; drone/camera-assisted monitoring exists in narrow security contexts but not as a general investigative surveillance product.

Testify in court regarding investigation findings.

CI 0–0 · exposure 0 · augmentation 38 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	This task involves a core legal function that has not and cannot shift to AI automation due to its nature. No sector is adopting AI to testify in place of human fraud examiners.
Sector adoption velocity	claude-sonnet-5	1/5	Legal and judicial systems show essentially no movement toward AI-delivered testimony; this is a highly conservative, human-mandated domain.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI can assist by preparing evidence summaries, organizing findings, or generating visual aids before testimony, but the courtroom testimony itself remains entirely human-driven with limited augmentation potential during the actual act of testifying.
Augmentation potential	claude-sonnet-5	3/5	AI can help investigators prepare testimony by organizing evidence, drafting summaries, and anticipating cross-examination questions, though the testimony itself remains fully human.
Task automatability	claude-haiku-4-5-20251001	1/5	Courtroom testimony requires human judgment, credibility assessment under cross-examination, and legal authority that only a human can provide. AI cannot appear as a witness or provide testimony with legal standing today.
Task automatability	claude-sonnet-5	1/5	Testifying in court requires a live human witness who can be cross-examined, swear an oath, and respond dynamically; no AI system can perform this act itself.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Testifying in court is explicitly a legal function requiring a human witness under oath. Courts require human testimony; regulatory and legal frameworks mandate human presence and accountability, creating an absolute barrier to automation.
Adoption barriers	claude-sonnet-5	5/5	Courtroom testimony requires a qualified, sworn human witness subject to legal rules of evidence, perjury liability, and cross-examination—an absolute legal barrier to automation.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	The task requires a qualified human examiner to appear in court; AI cannot substitute for this human obligation. The cost is entirely determined by human time and presence.
Cost vs. human wage	claude-sonnet-5	1/5	There is no AI substitute cost to compare since the task cannot be performed by AI at all; the human cost is unavoidable.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed AI system can autonomously testify in court. While AI can assist with evidence presentation, the task itself—oral testimony under oath—is legally reserved for human witnesses.
Technical feasibility today	claude-sonnet-5	1/5	No deployed product testifies in court on behalf of a human investigator; this remains legally and practically infeasible today.

Obtain and serve subpoenas.

CI 0–0 · exposure 0 · augmentation 25 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	Legal processes remain highly regulated and conservative; courts and law firms continue to rely on human staff and court systems for subpoena management, with minimal AI displacement in evidence.
Sector adoption velocity	claude-sonnet-5	1/5	Legal service of process is a highly manual, low-digitization function with minimal AI adoption; investigative and legal sectors have been slow to change this specific procedural task.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI could assist by drafting subpoena language, managing databases of recipient information, or tracking service status, but human legal professionals must remain responsible for review, authorization, and proper execution of the subpoena itself.
Augmentation potential	claude-sonnet-5	2/5	AI can help draft subpoena requests, track deadlines, or manage documentation, but offers little assistance for the actual physical/legal act of obtaining and serving them.
Task automatability	claude-haiku-4-5-20251001	1/5	Obtaining and serving subpoenas is a legal and procedural task requiring human judgment, authorization, and personal delivery in many jurisdictions. Current AI systems cannot independently perform the legal actions, personal service requirements, or procedural compliance that this task demands.
Task automatability	claude-sonnet-5	1/5	Serving subpoenas requires physical or legal delivery to a person/entity and often court-mandated procedures that AI cannot execute end-to-end; drafting portions may be assisted but the core act is procedural/legal execution.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Subpoena issuance and service are legally-protected functions; most jurisdictions require a licensed attorney or court officer to authorize and execute subpoenas, and proof of proper service must be certified by a qualified person—creating a hard regulatory barrier to automation.
Adoption barriers	claude-sonnet-5	5/5	Serving subpoenas is a legally regulated act requiring proper authorization, chain-of-custody, and often licensed process servers or court officers, creating hard legal and procedural barriers to automation.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	The cost of implementing an AI system to handle subpoena procedures would exceed the cost of having a paralegal or clerk perform the task, given the legal oversight and error costs that would still be required.
Cost vs. human wage	claude-sonnet-5	1/5	AI cannot perform the physical service or legal authorization step, so there is no meaningful AI cost basis to compare against human service fees for this specific action.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed AI product reliably performs subpoena issuance and service; these are legally-mandated human functions requiring attorney review, proper documentation, and verified service of process that law firms and courts still handle through human personnel.
Technical feasibility today	claude-sonnet-5	1/5	No deployed AI product serves legal process or obtains subpoenas; this remains a human/legal-system function performed by process servers, attorneys, or court officers.

Arrest individuals to be charged with fraud.

CI 0–0 · exposure 0 · augmentation 25 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
Dimension	Model	Rating	Rationale
Sector adoption velocity	claude-haiku-4-5-20251001	1/5	This task involves no AI adoption because it remains a non-automatable legal function reserved for human agents with arrest authority.
Sector adoption velocity	claude-sonnet-5	1/5	Law enforcement and physical custodial functions show essentially no AI adoption for the act of arrest itself.
Augmentation potential	claude-haiku-4-5-20251001	2/5	AI may assist fraud examiners in evidence gathering, pattern analysis, or suspect identification before arrest, but cannot augment the arrest act itself, which requires human judgment and legal authority.
Augmentation potential	claude-sonnet-5	2/5	AI can assist with case-building, evidence analysis, and identifying suspects prior to arrest, but offers no assistance during the physical act of arrest itself.
Task automatability	claude-haiku-4-5-20251001	1/5	Arrest requires legal authority, physical custody, and presence at a location—none of which AI systems can perform. This task is fundamentally incompatible with current AI capabilities and legal frameworks.
Task automatability	claude-sonnet-5	1/5	Physical arrest requires bodily presence, use of authority, and potential use of force, none of which current AI systems can perform end-to-end.
Adoption barriers	claude-haiku-4-5-20251001	5/5	Arrest authority is granted exclusively to licensed law enforcement officers under statute and constitutional law; only a human with specific legal authority can lawfully detain a person.
Adoption barriers	claude-sonnet-5	5/5	Arrest authority is a legally restricted power held only by sworn law enforcement officers, representing a hard legal and constitutional barrier to automation.
Cost vs. human wage	claude-haiku-4-5-20251001	1/5	AI cannot perform this task at any cost since it lacks legal standing and physical agency to execute arrests. The comparison is inapplicable.
Cost vs. human wage	claude-sonnet-5	1/5	There is no AI substitute capable of performing this task at all, so cost comparison favors the human by default.
Technical feasibility today	claude-haiku-4-5-20251001	1/5	No deployed AI product can execute an arrest; this is exclusively a human law-enforcement function backed by statutory and constitutional authority that cannot be delegated to machines.
Technical feasibility today	claude-sonnet-5	1/5	No deployed AI product performs physical arrests; this remains entirely outside the scope of existing AI technology.

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.