{ "reviewed_on": "2026-10-25", "status": "Conditional workflow scenarios; task hours and coverage are assumptions, not measured industry allocations.", "rounding": "Outward to five percentage points; exact hour arithmetic remains available.", "review_changes": [ "Audit 65% retained as an unvalidated automation sensitivity; reference range 25–35%.", "Amazon 2025 discovery case held because it uses 2024 technology and unresolved timing boundaries.", "Tax research averages used once, without extra coverage/success discounting." ], "sources": { "accounting-microsoft_collections_2026_insidetrack": { "id": "accounting-microsoft_collections_2026_insidetrack", "title": "Microsoft Inside Track: Streamlining finance cash collection at Microsoft with AI", "url": "https://www.microsoft.com/insidetrack/blog/streamlining-finance-cash-collection-at-microsoft-with-ai/", "publication_date": "2026-06-04", "evidence_period": "Editor note says the internal data were gathered during writing; exact observation window, model versions and launch dates are unstated.", "status": "retain", "kind": "First-party internal deployment report", "reported_result": "Up to 60% lower inquiry handling time using inline suggestions, call summaries and automatically drafted replies.", "transfer": "Closer to human-assisted collections than the autonomous email PDF. Supports retrieval/context/drafting as saving mechanisms. The maximum is not an average or endpoint estimate.", "limits": "SAP/Dynamics consolidation, standardization and automated communications accompany AI assistance. No sample, task timer or fixed-output quality study is disclosed. Do not transfer its matching-accuracy or resolution-speed metrics into labor percentages.", "verified_this_pass": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note.", "project_ids": [ "AI-02" ], "evidence_type": "First-party internal deployment report", "date_verification": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note.", "research_status": "new_recent_primary_reference", "plain_summary": "Up to 60% lower inquiry handling time using inline suggestions, call summaries and automatically drafted replies.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary page: visible June 4, 2026 date; handling-time result in operating-results paragraph; period caveat in editor note." }, "accounting-microsoft_collections": { "id": "accounting-microsoft_collections", "title": "Microsoft Frontier Finance: Collections Email Agent", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/Collections-Email-Agent.pdf", "publication_date": "Document created 2026-05-08; public by 2026-05-11 according to completed source review", "evidence_period": "Measurement dates and model version unstated.", "status": "retain", "kind": "First-party internal deployment case", "reported_result": "About 60,000 incoming emails/month, about 55% resolved automatically and 36,000 annual hours saved.", "transfer": "Demonstrates intent handling and routine reply preparation at scale. Autonomous sending exceeds this project's approval boundary.", "limits": "No baseline human-hours or time-study denominator. Neither 55% automated volume nor the annual hours supplies a percentage saving for a reviewed assistant.", "verified_this_pass": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded.", "project_ids": [ "AI-02" ], "evidence_type": "First-party internal deployment case", "date_verification": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded.", "research_status": "retain_as_illustrative", "plain_summary": "About 60,000 incoming emails/month, about 55% resolved automatically and 36,000 annual hours saved.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party PDF text; document date inherited from source-review-2026-10-25/review.json, whose PDF metadata check is recorded." }, "accounting-microsoft_finance_2025_ga": { "id": "accounting-microsoft_finance_2025_ga", "title": "Microsoft: Empowering Finance with an AI Assistant in Microsoft 365 Copilot", "url": "https://www.microsoft.com/en-us/dynamics-365/blog/it-professional/2025/10/20/empowering-finance-with-an-ai-assistant-in-microsoft-365-copilot/", "publication_date": "2025-10-20", "evidence_period": "Product availability in October 2025; unnamed pilots have unstated measurement periods.", "status": "retain", "kind": "Product announcement", "reported_result": "ERP-grounded draft customer replies for staff review and Excel reconciliation matching/exception assistance.", "transfer": "Confirms both specified human-reviewed workflow shapes are technically available.", "limits": "Days-to-hours pilot wording has no auditable human-time denominator. Supplies no task percentage or endpoint.", "verified_this_pass": "Live primary page with visible date.", "project_ids": [ "AI-02", "AI-03" ], "evidence_type": "Product announcement", "date_verification": "Live primary page with visible date.", "research_status": "retain_as_workflow_reference", "plain_summary": "ERP-grounded draft customer replies for staff review and Excel reconciliation matching/exception assistance.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary page with visible date." }, "accounting-ted_taylor_2026": { "id": "accounting-ted_taylor_2026", "title": "OpenAI: Franchisee reconciles cash in 94% less time with ChatGPT Work", "url": "https://openai.com/index/ted-taylor-restaurant/", "publication_date": "2026-10-20", "evidence_period": "Current customer use described in October 2026; exact deployment/measurement dates and model version unstated. No known older experiment is identified.", "status": "retain", "kind": "Vendor-authored named customer case", "reported_result": "Ted Taylor compares cash transactions with bank deposits monthly across three restaurants. Reported time falls from three hours to five to ten minutes; the article says he reviews results before implementing.", "transfer": "Direct cash-reconciliation feasibility reference. From stated times alone the reported reduction is 94.4–97.2%; use the raw minutes rather than extending its headline to other workflows.", "limits": "No transaction count, exception mix, timed sample, error rate, independent reviewer or breakdown of preparation/review is disclosed. It is unclear whether five to ten minutes includes every retained check. The catalog's two-dataset controlled reconciliation has a more explicit review standard.", "verified_this_pass": "Live primary article, visible date and back-office reconciliation paragraph.", "project_ids": [ "AI-03" ], "evidence_type": "Vendor-authored named customer case", "date_verification": "Live primary article, visible date and back-office reconciliation paragraph.", "research_status": "new_recent_primary_reference", "plain_summary": "Ted Taylor compares cash transactions with bank deposits monthly across three restaurants. Reported time falls from three hours to five to ten minutes; the article says he reviews results before implementing.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article, visible date and back-office reconciliation paragraph." }, "accounting-chatham_2026": { "id": "accounting-chatham_2026", "title": "OpenAI: Chatham scales its capital markets expertise with OpenAI", "url": "https://openai.com/index/chatham-financial/", "publication_date": "2026-10-02", "evidence_period": "Early measurement reported October 2026; exact dates unstated; GPT-5.6 generation explicitly identified.", "status": "retain", "kind": "Vendor-authored named customer early measurement", "reported_result": "Trade-validation review reportedly falls from about 30 minutes to under four; evidence is assembled, terms compared and discrepancies flagged.", "transfer": "Supports substantial reduction in narrow comparison/evidence preparation; can inform feasibility of the upper reconciliation scenario.", "limits": "Trade validation differs from ledger reconciliation. No sample, error rate or timing protocol; comparison against experienced reviewers is ongoing. It does not justify removing independent review.", "verified_this_pass": "Live primary article and visible date.", "project_ids": [ "AI-03" ], "evidence_type": "Vendor-authored named customer early measurement", "date_verification": "Live primary article and visible date.", "research_status": "retain_as_adjacent_reference", "plain_summary": "Trade-validation review reportedly falls from about 30 minutes to under four; evidence is assembled, terms compared and discrepancies flagged.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article and visible date." }, "accounting-microsoft_sox": { "id": "accounting-microsoft_sox", "title": "Microsoft Frontier Finance: SOX Change Management Testing Agent", "url": "https://cdn-dynmedia-1.microsoft.com/is/content/microsoftcorp/microsoft/mcaps/documents/fy26/SOX-Change-Management-Testing-Agent.pdf", "publication_date": "File modified 2026-05-07; public by 2026-05-11 according to completed source review", "evidence_period": "Deployment/measurement dates and model version absent; listed technology includes Bing Chat. A 2026 file date does not establish recent deployment technology.", "status": "retain", "kind": "First-party internal deployment case", "reported_result": "850 annual hours saved. Two agents extract ticket attributes, prepopulate rule checks and mark the matching source evidence.", "transfer": "Direct match to ticket evidence extraction and annotation; supports preparing standardized source-linked workpapers.", "limits": "No baseline hours, ticket volume, correction rate or timed sample. Cannot infer a percentage or establish current-model speed. Auditor verification remains in the scenario.", "verified_this_pass": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit.", "project_ids": [ "AI-08" ], "evidence_type": "First-party internal deployment case", "date_verification": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit.", "research_status": "retain_as_illustrative", "plain_summary": "850 annual hours saved. Two agents extract ticket attributes, prepopulate rule checks and mark the matching source evidence.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Downloaded first-party PDF with curl and extracted its text with pdftotext after web reader failed; date provenance remains the existing metadata audit." }, "accounting-provectus_johnson_lambert": { "id": "accounting-provectus_johnson_lambert", "title": "Provectus: Johnson Lambert reinvents insurance auditing with generative AI", "url": "https://www.provectus.com/case-studies/johnson-lambert-audit/", "publication_date": "2026-04-20, first-party metadata verified in completed source review", "evidence_period": "Exact measurement period absent; listed Claude 3.7 Sonnet and Gemini 2.5 Pro establish post-2024 technology.", "status": "retain", "kind": "Vendor-authored named customer deployment", "reported_result": "50% lower document-processing time; auditors inspect extracted tables against originals and correct values.", "transfer": "Relevant human-checked extraction analogue for the lower audit preparation scenario.", "limits": "PDF financial-report extraction differs from ticket testing. No timed sample or task breakdown. Other broader audit headlines conflict; do not use them or convert an efficiency increase into an equal time reduction. Its 50% document-processing result is neither a universal cap nor a guaranteed stage effect.", "verified_this_pass": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check.", "project_ids": [ "AI-08" ], "evidence_type": "Vendor-authored named customer deployment", "date_verification": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check.", "research_status": "retain_as_adjacent_reference", "plain_summary": "50% lower document-processing time; auditors inspect extracted tables against originals and correct values.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary article confirms workflow, named model generations and narrower result; publication date inherited from recorded first-party metadata check." }, "accounting-afm_controls": { "id": "accounting-afm_controls", "title": "Netherlands AFM: 12 Building blocks for controlled use of auditing tools", "url": "https://www.afm.nl/~/profmedia/files/rapporten/2025/12-building-blocks-for-controlled-use-of-auditing-tools.pdf", "publication_date": "2025-12", "evidence_period": "2025 supervisory guidance; older firm-selection data are unused.", "status": "retain", "kind": "Regulator guidance", "reported_result": "No productivity result. Requires reliable input, verifiable outputs and continuing auditor responsibility.", "transfer": "Supports retaining substantive review and evidence tracing. It neither fixes review hours nor prevents review from getting faster through organized evidence.", "limits": "No numerical effect; none used.", "verified_this_pass": "Live primary PDF.", "project_ids": [ "AI-03", "AI-08" ], "evidence_type": "Regulator guidance", "date_verification": "Live primary PDF.", "research_status": "retain_as_requirements_reference", "plain_summary": "No productivity result. Requires reliable input, verifiable outputs and continuing auditor responsibility.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live primary PDF." }, "planning-pwc_cfo_advisor": { "title": "PwC CFO Advisor implementation case", "url": "https://cee.pwc.com/case-studies/how-cfo-advisor-enhances-and-automates-financial-analysis-and-storytelling-with-ai.html", "publication_date": "2026-04-01", "date_verification": "Live first-party HTML: pwcReleaseDate=2026-04-01T17:45:00.000+03:00; pwcLastModifiedDate=2026-04-01T14:00:19.185Z.", "evidence_period": "Deployment and observation calendar unstated; operating implementation described in the 2026 case.", "evidence_type": "Consultancy-authored anonymous TMT customer case", "reported_result": "About 90% less variance-analysis effort across approximately 20 configured analysis types, with controllers and FP&A editing and approving drafts.", "use": "90% applies only to eligible recurring variance preparation in the AI-04 scenario; it is not reduced by an invented efficiency multiplier.", "limits": "No baseline hours, sample, timing protocol or quality comparison. Tool consolidation and fewer report pages accompany the result. The scenario assumes that stage result can transfer at unchanged output; it is not demonstrated for the scoped workflow. Ambiguous FTE wording is unused.", "checked": "2026-10-25", "id": "planning-pwc_cfo_advisor", "research_status": "retain", "status": "retain", "kind": "Consultancy-authored anonymous TMT customer case", "plain_summary": "About 90% less variance-analysis effort across approximately 20 configured analysis types, with controllers and FP&A editing and approving drafts.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party HTML: pwcReleaseDate=2026-04-01T17:45:00.000+03:00; pwcLastModifiedDate=2026-04-01T14:00:19.185Z." }, "planning-natura_sap": { "title": "SAP: Natura gross-margin application", "url": "https://news.sap.com/2026/07/natura-co-transforming-finance-generative-ai/", "publication_date": "2026-07-08", "date_verification": "Visible first-party article date and explicit August 2025 production go-live.", "evidence_period": "Production since August 2025; use in Ecuador described July 2026. Originating early-2024 prototype is excluded from quantitative support.", "evidence_type": "Named vendor/customer production case", "reported_result": "ERP-connected gross-margin revenue/cost explanations with human interpretation and decision control; no numeric human-time saving.", "use": "Supports the AI-04 workflow and approved-calculation boundary; supplies no task saving rate.", "limits": "Six months describes implementation. No labor baseline or saving is reported.", "checked": "2026-10-25", "id": "planning-natura_sap", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer production case", "plain_summary": "ERP-connected gross-margin revenue/cost explanations with human interpretation and decision control; no numeric human-time saving.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible first-party article date and explicit August 2025 production go-live." }, "planning-clickup_pigment": { "title": "Pigment: ClickUp Analyst Agent case", "url": "https://www.pigment.com/customer-stories/clickup", "publication_date": "2026-09-03", "date_verification": "Live first-party JSON-LD: datePublished=2026-09-03T18:56:58.573Z; dateModified=2026-09-03T18:53:21.520Z.", "evidence_period": "Monthly operating use described in 2026; deployment/measurement months unstated. An agent launch article visibly published September 2025 also describes the result, so this is not assumed to originate in 2026.", "evidence_type": "Named vendor/customer deployment case", "reported_result": "Reported 1-1.5 days saved per person monthly across six finance staff on first-draft budget-variance reports and driver identification.", "use": "Independent named workflow support for AI-05; saved days are not converted to a percentage.", "limits": "No baseline reporting hours, hours/day or timing protocol. Separate Modeler Agent claims are excluded.", "checked": "2026-10-25", "id": "planning-clickup_pigment", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer deployment case", "plain_summary": "Reported 1-1.5 days saved per person monthly across six finance staff on first-draft budget-variance reports and driver identification.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party JSON-LD: datePublished=2026-09-03T18:56:58.573Z; dateModified=2026-09-03T18:53:21.520Z." }, "planning-docker_pigment": { "title": "Pigment: Docker monthly close commentary and review", "url": "https://www.pigment.com/customer-stories/docker", "publication_date": "2026-09-11", "date_verification": "Live first-party JSON-LD: datePublished=2026-09-11T19:10:43.460Z; dateModified=2026-09-11T17:15:48.513Z.", "evidence_period": "Current monthly operating use in a 2026 case; exact measurement months unstated.", "evidence_type": "Named vendor/customer deployment; customer self-estimate", "reported_result": "Estimated 30-50% less time on monthly close commentary and review. The process covers actual-versus-budget analysis, narrative drafting and packaging; staff refine judgment and audience context. Analyst Agent and Claude/MCP are used.", "use": "Transfer 30% and 50% to the combined commentary/packaging/review block in AI-05, where the task boundary is closest. Do not assign this result separately to each constituent task.", "limits": "No timed sample, baseline hours, unchanged-output test or detailed correction boundary. Final pack release approval is separately retained. Modeler Agent build/formula results are excluded.", "checked": "2026-10-25", "id": "planning-docker_pigment", "research_status": "retain", "status": "retain", "kind": "Named vendor/customer deployment; customer self-estimate", "plain_summary": "Estimated 30-50% less time on monthly close commentary and review. The process covers actual-versus-budget analysis, narrative drafting and packaging; staff refine judgment and audience context. Analyst Agent and Claude/MCP are used.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Live first-party JSON-LD: datePublished=2026-09-11T19:10:43.460Z; dateModified=2026-09-11T17:15:48.513Z." }, "planning-bcg_finance_roi_2025": { "title": "BCG: How to Get ROI from AI in the Finance Function", "url": "https://www.bcg.com/publications/2025/how-finance-leaders-can-get-roi-from-ai", "publication_date": "2025-06-04", "date_verification": "Visible first-party article date; separate finance survey conducted March 2025.", "evidence_period": "Consumer-goods deployment timing unstated; described in the 2025 article.", "evidence_type": "Consultancy-authored anonymous customer transformation", "reported_result": "50% shorter report-generation time after constructing a driver-tree model, integrating data and adding a GenAI query interface.", "use": "Triangulation only: demonstrates the model-backed answer mechanism.", "limits": "Bundled model/data transformation and unclear elapsed-versus-human-time boundary. No numerical rate is transferred because these projects already have usable data and a validated model.", "checked": "2026-10-25", "id": "planning-bcg_finance_roi_2025", "research_status": "retain", "status": "retain", "kind": "Consultancy-authored anonymous customer transformation", "plain_summary": "50% shorter report-generation time after constructing a driver-tree model, integrating data and adding a GenAI query interface.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible first-party article date; separate finance survey conducted March 2025." }, "planning-amazon_finance_discovery_2025": { "id": "planning-amazon_finance_discovery_2025", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-built-an-ai-assistant-using-amazon-bedrock-and-amazon-kendra-to-support-analysts-for-data-discovery-and-business-insights/", "title": "How Amazon Finance built an AI assistant using Amazon Bedrock and Amazon Kendra to support analysts for data discovery and business insights", "publication_date": "2025-08-28", "date_verification": "Visible dateline and article:published_time / datePublished HTML time: 2025-08-28T12:14:22-08:00.", "evidence_period": "Measurement and deployment calendar dates unstated; explicitly uses Claude 3 Sonnet, a 2024-era model. The 2025 publication does not establish a 2025 experiment or current-model performance.", "status": "hold", "kind": "First-party finance implementation account", "result": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "denominator_and_limits": "The 30% benchmark and 85% before/after descriptions have no reconciled task/sample boundary. Over 50 business queries; no human sample size, timing protocol, calendar or complete checking labor. Data discovery/knowledge search, rather than a full finance workflow; 8/10 retrieval is not accepted-answer coverage.", "quantitative_use": "Task-level feasibility for expanded metrics/datasets/precedents scope. Report both time claims or only explicit before/after minutes with a boundary caveat. Do not choose 85% as a general measured effect or set workflow endpoints from it. Model and measurement calendar remain qualified.", "project_ids": [ "AI-06" ], "evidence_type": "First-party finance implementation account", "reported_result": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "limits": "The 30% benchmark and 85% before/after descriptions have no reconciled task/sample boundary. Over 50 business queries; no human sample size, timing protocol, calendar or complete checking labor. Data discovery/knowledge search, rather than a full finance workflow; 8/10 retrieval is not accepted-answer coverage.", "reason": "Claude 3 Sonnet is 2024 technology; measurement dates are unknown and the article reports unreconciled 30% and 85% search/discovery results.", "research_status": "hold", "plain_summary": "Reports 30% less search time in analyst benchmarking, and separately 45–60 minutes to 5–10 minutes to find relevant information (called 85% improvement). Also reports roughly 8/10 first-attempt retrievals.", "verified_date_evidence": "Visible dateline and article:published_time / datePublished HTML time: 2025-08-28T12:14:22-08:00." }, "planning-fm_knowledge_2026": { "id": "planning-fm_knowledge_2026", "url": "https://www.microsoft.com/en/customers/story/26503-fm-azure", "title": "FM empowers 1,500+ engineers with trusted, AI-assisted access to engineering knowledge using Azure OpenAI", "publication_date": "2026-05-28", "date_verification": "Visible 5/28/2026 dateline.", "evidence_period": "First two production months, without calendar dates. Model version and timing protocol unstated.", "status": "retain", "kind": "Vendor/customer estimate; adjacent engineering knowledge search", "result": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "denominator_and_limits": "No baseline duration or full checking-time boundary. Estimated minutes cannot be converted to a percentage.", "quantitative_use": "Supports potentially substantial lookup savings and source-grounded implementation; neither calibrates a general percentage nor caps lookup savings.", "project_ids": [ "AI-06" ], "evidence_type": "Vendor/customer estimate; adjacent engineering knowledge search", "reported_result": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "limits": "No baseline duration or full checking-time boundary. Estimated minutes cannot be converted to a percentage.", "research_status": "retain", "plain_summary": "Estimated 6–10 minutes saved per search; 17,000 queries in the first two production months across more than 1,500 engineers.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible 5/28/2026 dateline." }, "planning-amazon_regulatory_rag_2026": { "id": "planning-amazon_regulatory_rag_2026", "url": "https://aws.amazon.com/blogs/machine-learning/how-amazon-finance-streamlines-regulatory-inquiries-by-using-generative-ai-on-aws/", "title": "How Amazon Finance streamlines regulatory inquiries by using generative AI on AWS", "publication_date": "2026-05-12", "date_verification": "Visible 12 MAY 2026 dateline.", "evidence_period": "May 2026 account names Claude Sonnet 4.5; exact go-live/observation dates unstated.", "status": "retain", "kind": "First-party technical deployment account", "result": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "denominator_and_limits": "Retrieval number compares system configurations; no human-hours result. Regulatory inquiries differ from ongoing tax monitoring.", "quantitative_use": "Mechanism only; no numeric human-time factor is derived from retrieval latency.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "First-party technical deployment account", "reported_result": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "limits": "Retrieval number compares system configurations; no human-hours result. Regulatory inquiries differ from ongoing tax monitoring.", "research_status": "retain", "plain_summary": "Finds historical documents and precedents and supports regulatory-response preparation. Parallel retrieval falls from 10 seconds to under 2 seconds.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Visible 12 MAY 2026 dateline." }, "planning-tr_cocounsel_corporate": { "id": "planning-tr_cocounsel_corporate", "url": "https://tax.thomsonreuters.com/en/products/cocounsel-tax/corp", "title": "CoCounsel Tax: The AI tax assistant for in-house teams", "publication_date": "2026-06-29", "date_verification": "Primary HTML w-published-date and analytics publishDate=2026-06-29; visible survey footnote states November–December 2025.", "evidence_period": "Self-reported user survey November–December 2025.", "status": "retain", "kind": "Vendor user survey", "result": "Users report an average 32% less tax-research time.", "denominator_and_limits": "Sample size, task mix and review boundary not disclosed; not independently timed or verified. Research differs from scanning five sources and obtaining approval for every relevant update.", "quantitative_use": "Adjacent research reference used to illustrate the lower read/draft scenario assumption; no causal/full-workflow claim.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "Vendor user survey", "reported_result": "Users report an average 32% less tax-research time.", "limits": "Sample size, task mix and review boundary not disclosed; not independently timed or verified. Research differs from scanning five sources and obtaining approval for every relevant update.", "research_status": "retain", "plain_summary": "Users report an average 32% less tax-research time.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Primary HTML w-published-date and analytics publishDate=2026-06-29; visible survey footnote states November–December 2025." }, "planning-tr_copeland_buhl_2026": { "id": "planning-tr_copeland_buhl_2026", "url": "https://tax.thomsonreuters.com/en/insights/case-studies/from-hours-to-minutes-how-cocounsel-reimagined-tax-workflow-efficiency-at-copeland-buhl", "title": "From hours to minutes: How CoCounsel reimagined tax workflow efficiency at Copeland Buhl", "publication_date": "2026-03-09", "date_verification": "Fetched primary HTML: w-published-date=2026-03-09T00:00:00.000Z and analytics publishDate=2026-03-09.", "evidence_period": "14,000+ messages over six months, exact months unstated. CoCounsel Tax & Audit launched March 2025 per separate dated first-party announcement.", "status": "retain", "kind": "Named vendor/customer implementation account", "result": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "denominator_and_limits": "No timed sample, matched task baseline, exact dates or average review duration. 2–3 times faster implies roughly 50–67% shorter research only if the same task and output are compared. Memo-review minutes do not justify eliminating expert tax applicability approval.", "quantitative_use": "Supports a conditional high research-stage reduction around 65%; exact modeled factor remains an assumption.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "Named vendor/customer implementation account", "reported_result": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "limits": "No timed sample, matched task baseline, exact dates or average review duration. 2–3 times faster implies roughly 50–67% shorter research only if the same task and output are compared. Memo-review minutes do not justify eliminating expert tax applicability approval.", "research_status": "retain_with_measurement_limits", "plain_summary": "Research projects reported 2–3 times faster; reviewers save 15–30 minutes per research memo. 67 active licenses.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Fetched primary HTML: w-published-date=2026-03-09T00:00:00.000Z and analytics publishDate=2026-03-09." }, "planning-tr_tax_launch_confirmation_2026": { "id": "planning-tr_tax_launch_confirmation_2026", "url": "https://www.thomsonreuters.com/en-us/posts/innovation/the-real-ai-story-in-tax-and-audit-isnt-adoption-its-impact/", "title": "The Real AI Story in Tax and Audit Isn’t Adoption — It’s Impact", "publication_date": "2026-04-16", "date_verification": "Primary HTML datePublished=2026-04-16T08:00:32+00:00; modified 2026-04-23.", "evidence_period": "States CoCounsel Tax & Audit launched March 2025.", "status": "retain", "kind": "First-party executive article", "result": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "denominator_and_limits": "Does not calendar-date the Copeland Buhl six-month period or provide a controlled time study.", "quantitative_use": "Date corroboration only.", "project_ids": [ "AI-06", "AI-07" ], "evidence_type": "First-party executive article", "reported_result": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "limits": "Does not calendar-date the Copeland Buhl six-month period or provide a controlled time study.", "research_status": "retain_as_product_date_confirmation", "plain_summary": "Confirms March 2025 product launch; this establishes product-era context for the Copeland Buhl account.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "Primary HTML datePublished=2026-04-16T08:00:32+00:00; modified 2026-04-23." }, "planning-tax_plat": { "id": "planning-tax_plat", "url": "https://arxiv.org/abs/2503.03444", "title": "Taxation Perspectives from Large Language Models: A Case Study on Additional Tax Penalties", "publication_date": "2025-03-05; revised 2026-02-08", "evidence_period": "Recent 2025 model snapshots in revised benchmark; exact runs unstated.", "status": "retain", "kind": "Tax reasoning benchmark", "result": "Continuing difficulties applying tax law to facts and conclusions.", "denominator_and_limits": "300 Korean tax-penalty questions; no human-time study.", "quantitative_use": "Supports preserving applicability review; no percentage or correction-hours factor derived.", "project_ids": [ "AI-07" ], "evidence_type": "Tax reasoning benchmark", "reported_result": "Continuing difficulties applying tax law to facts and conclusions.", "limits": "300 Korean tax-penalty questions; no human-time study.", "research_status": "retain_as_accuracy_context", "plain_summary": "Continuing difficulties applying tax law to facts and conclusions.", "reason": "Retained after current source review; observation dates and limits are recorded separately.", "verified_date_evidence": "" } }, "projects": { "AI-02": { "title": "Collections reply assistant", "baseline": "Five supported inquiry types from intake and account lookup through reply drafting, factual verification, correction and approved dispatch. Dispute negotiation and wider receivables management are excluded. The same inbound mix is compared at both endpoints.", "evidence_strength": "Recent first-party assisted-collections report gives an up-to-60% handling-time result; exact catalog range remains an analyst scenario. An autonomous collections case confirms broader capabilities but supplies no valid percentage denominator for reviewed replies.", "blocks": [ { "work": "Read, classify and route incoming inquiries", "baseline_hours": 10, "reduction_percent": [ 20.0, 80.0 ], "after_hours": [ 8, 2 ], "basis": "Lower: staff still read most messages and correct routing. Upper: intent and routing are prepared automatically; staff use the review step below to read the request and confirm it belongs to the supported customer/type.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Retrieve account records, invoice detail and supporting evidence", "baseline_hours": 30, "reduction_percent": [ 33.33333333, 83.33333333 ], "after_hours": [ 20, 5 ], "basis": "Lower: one-third of manual lookup effort disappears but ambiguous identifiers and missing records still require searches. Upper: correctly linked account/invoice evidence is presented together for most cases; five hours remain for unresolved identifiers, stale sources and manual fallback.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Prepare reply text and perform preparer edits", "baseline_hours": 30, "reduction_percent": [ 66.66666667, 90.0 ], "after_hours": [ 10, 3 ], "basis": "Lower: approved templates supply first drafts but material edits remain. Upper: repetitive supported replies are populated from verified records and approved language; three hours cover preparer edits and rejected drafts. Reading and factual approval are counted in the next block.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Verify customer/account facts, correct errors, approve and dispatch every reply", "baseline_hours": 30, "reduction_percent": [ -23.33333333, 0.0 ], "after_hours": [ 37, 30 ], "basis": "Every outgoing reply retains human approval. Lower adds seven hours for checking unfamiliar drafts and repairing errors; upper retains the full existing thirty-hour verification/dispatch budget. No gain from autonomous sending is counted.", "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_finance_2025_ga" ] } ], "source_ids": [ "accounting-microsoft_collections_2026_insidetrack", "accounting-microsoft_collections", "accounting-microsoft_finance_2025_ga" ], "range_percent": [ 25, 60 ], "calculated_percent": [ 25, 60 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [], "uncertainties": [ "Recent first-party assisted-collections report gives an up-to-60% handling-time result; exact catalog range remains an analyst scenario. An autonomous collections case confirms broader capabilities but supplies no valid percentage denominator for reviewed replies." ], "validation_needed": [ "Time the same five inquiry types before and after, tracking collector and approver labor separately", "Stratify by identity ambiguity, record completeness and draft acceptance; record every fallback and correction", "Require unchanged factual accuracy, disclosure controls and resolution quality before treating the high timing budget as achieved" ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": [ "Substantial manual lookup and typing in the starting process", "Read-only current account/invoice evidence joined by reliable customer identifiers", "Narrow approved reply policies and repetitive supported inquiry types", "Per-reply human identity, facts, disclosure and dispatch check", "All repair and fallback time included within the four budgets", "Connected data and draft templates remove 32 preparation hours, while correction and approval add seven: 100 - (8 + 20 + 10 + 37) = 25 hours saved." ], "remaining_hours": 75, "saved_hours_per_100": 25 }, "fast": { "conditions": [ "Substantial manual lookup and typing in the starting process", "Read-only current account/invoice evidence joined by reliable customer identifiers", "Narrow approved reply policies and repetitive supported inquiry types", "Per-reply human identity, facts, disclosure and dispatch check", "All repair and fallback time included within the four budgets", "Reliable customer links and standardized replies reduce preparation from 70 to ten hours; every response still gets the original thirty-hour approval/dispatch allowance: 100 - (2 + 5 + 3 + 30) = 60 hours saved." ], "remaining_hours": 40, "saved_hours_per_100": 60 } }, "lower_case": "Substantial manual lookup and typing in the starting process Read-only current account/invoice evidence joined by reliable customer identifiers Narrow approved reply policies and repetitive supported inquiry types Per-reply human identity, facts, disclosure and dispatch check All repair and fallback time included within the four budgets Connected data and draft templates remove 32 preparation hours, while correction and approval add seven: 100 - (8 + 20 + 10 + 37) = 25 hours saved.", "upper_case": "Substantial manual lookup and typing in the starting process Read-only current account/invoice evidence joined by reliable customer identifiers Narrow approved reply policies and repetitive supported inquiry types Per-reply human identity, facts, disclosure and dispatch check All repair and fallback time included within the four budgets Reliable customer links and standardized replies reduce preparation from 70 to ten hours; every response still gets the original thirty-hour approval/dispatch allowance: 100 - (2 + 5 + 3 + 30) = 60 hours saved." }, "AI-03": { "title": "Reconciliation assistant", "baseline": "Prepare and align two datasets, apply three agreed matching rules, investigate up to ten exception types, produce source-linked notes and verify balances/conclusions. No autonomous journal posting. Savings include configured deterministic matching delivered in this project.", "evidence_strength": "New named 2026 cash-reconciliation example reports 180 to five-to-ten minutes; adjacent current-model trade validation reports about thirty to under four. Both are vendor cases with incomplete review and timing methods. They support a potentially large narrow-workflow reduction, not empirical endpoints for the catalog scope.", "blocks": [ { "work": "Prepare and align datasets; check completeness and totals", "baseline_hours": 20, "reduction_percent": [ 25.0, 85.0 ], "after_hours": [ 15, 3 ], "basis": "Lower: exports and mapping still need manual repair. Upper: stable scheduled exports and reusable mappings remove assembly; three hours remain for completeness, control totals and feed exceptions.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Match records and identify genuine differences", "baseline_hours": 35, "reduction_percent": [ 71.42857143, 97.14285714 ], "after_hours": [ 10, 1 ], "basis": "Rules execute over the entire population. Lower retains ten hours to resolve ambiguous candidates and inspect unmatched records. Upper retains one hour to initiate/check the matching result; no one manually compares already matched rows. This is mainly deterministic automation, not an attributed LLM reasoning effect.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Investigate exceptions and draft source-linked notes", "baseline_hours": 25, "reduction_percent": [ 20.0, 76.0 ], "after_hours": [ 20, 6 ], "basis": "Lower: most exceptions need external follow-up and AI mainly organizes records. Upper: a small recurring set of timing, reference and split-payment differences is resolved from the two connected datasets; staff check proposed causes and handle the remaining genuinely unresolved cases. Twenty-five to six is an assumed routine-exception budget, not a measured published stage result.", "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga" ] }, { "work": "Independently verify balances, evidence and resolution decisions", "baseline_hours": 20, "reduction_percent": [ 0.0, 25.0 ], "after_hours": [ 20, 15 ], "basis": "The same reviewer controls and final decisions remain. Upper saves five hours from having matched items, control totals and source-linked evidence organized together; professional judgment and unresolved differences still require people.", "source_ids": [ "accounting-afm_controls" ] } ], "source_ids": [ "accounting-ted_taylor_2026", "accounting-chatham_2026", "accounting-microsoft_finance_2025_ga", "accounting-afm_controls" ], "range_percent": [ 35, 75 ], "calculated_percent": [ 35, 75 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [], "uncertainties": [ "New named 2026 cash-reconciliation example reports 180 to five-to-ten minutes; adjacent current-model trade validation reports about thirty to under four. Both are vendor cases with incomplete review and timing methods. They support a potentially large narrow-workflow reduction, not empirical endpoints for the catalog scope." ], "validation_needed": [ "Use two comparable historical periods, then a shadow production run with the same rules and exception severity", "Measure preparation, matching oversight, investigation and reviewer labor separately", "Count all falsely matched, missed, duplicated and unsupported-explanation cases; time reviewer re-performance and manual fallback", "Replace assumed task shares with observed time and rerun the exact arithmetic" ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": [ "Starting reconciliation still includes significant manual matching", "Two reliable datasets and three reusable, deterministic matching rules", "Upper case has routine exception types with causes traceable within connected records", "Unchanged balances, materiality, evidence requirements and reviewer sign-off", "Every manual repair, unresolved exception and reviewer correction counted", "Preparation fifteen + matching ten + investigation twenty + independent verification twenty = sixty-five human-hours, a thirty-five-hour saving." ], "remaining_hours": 65, "saved_hours_per_100": 35 }, "fast": { "conditions": [ "Starting reconciliation still includes significant manual matching", "Two reliable datasets and three reusable, deterministic matching rules", "Upper case has routine exception types with causes traceable within connected records", "Unchanged balances, materiality, evidence requirements and reviewer sign-off", "Every manual repair, unresolved exception and reviewer correction counted", "Stable feeds, automatic rule matching and mostly recurring supported exceptions leave three + one + six + fifteen = twenty-five human-hours, a seventy-five-hour saving." ], "remaining_hours": 25, "saved_hours_per_100": 75 } }, "lower_case": "Starting reconciliation still includes significant manual matching Two reliable datasets and three reusable, deterministic matching rules Upper case has routine exception types with causes traceable within connected records Unchanged balances, materiality, evidence requirements and reviewer sign-off Every manual repair, unresolved exception and reviewer correction counted Preparation fifteen + matching ten + investigation twenty + independent verification twenty = sixty-five human-hours, a thirty-five-hour saving.", "upper_case": "Starting reconciliation still includes significant manual matching Two reliable datasets and three reusable, deterministic matching rules Upper case has routine exception types with causes traceable within connected records Unchanged balances, materiality, evidence requirements and reviewer sign-off Every manual repair, unresolved exception and reviewer correction counted Stable feeds, automatic rule matching and mostly recurring supported exceptions leave three + one + six + fifteen = twenty-five human-hours, a seventy-five-hour saving." }, "AI-08": { "title": "Audit evidence assistant", "baseline": "Retrieve the fixed change-ticket sample, extract evidence for the same three test criteria, prepare source-linked workpapers, independently verify evidence and reach the auditor's conclusion. Same sample, audit coverage and evidence standard at both endpoints.", "evidence_strength": "Direct SOX extraction/annotation case supplies annual hours without a denominator; recent human-reviewed audit-document case reports fifty percent less processing time. Exact ticket-workflow endpoints remain conditional analyst assumptions.", "blocks": [ { "work": "Locate tickets and gather supporting evidence", "baseline_hours": 20, "reduction_percent": [ 25.0, 50.0 ], "after_hours": [ 15, 10 ], "basis": "Assumed retrieval budget: 20 hours before, 15 with partial assistance or 10 with consistent indexed evidence. The 50% fast saving is a scenario assumption; the document-processing case does not separately time retrieval.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Extract approval, testing and date attributes; populate rule checks", "baseline_hours": 30, "reduction_percent": [ 50.0, 50.0 ], "after_hours": [ 15, 15 ], "basis": "Use the adjacent case’s 50% document-processing result as the reference extraction assumption: 30 hours becomes 15. Independent source verification stays in the final block; this result does not measure the entire audit workflow.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Draft and edit source-linked workpapers and annotations", "baseline_hours": 20, "reduction_percent": [ 40.0, 50.0 ], "after_hours": [ 12, 10 ], "basis": "Assumed workpaper preparation: 20 hours becomes 12 with substantial editing or 10 with consistent templates and source links. No published writing-time percentage is transferred.", "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert" ] }, { "work": "Independently verify original evidence, resolve exceptions and reach the conclusion", "baseline_hours": 30, "reduction_percent": [ -10.0, 0.0 ], "after_hours": [ 33, 30 ], "basis": "All thirty baseline auditor-review hours remain at the upper endpoint. Lower adds three hours to resolve omitted/conflicting attributes and annotations. Prepopulated pass/fail fields do not determine the final audit conclusion.", "source_ids": [ "accounting-afm_controls" ] } ], "source_ids": [ "accounting-microsoft_sox", "accounting-provectus_johnson_lambert", "accounting-afm_controls" ], "range_percent": [ 25, 35 ], "calculated_percent": [ 25, 35 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [], "uncertainties": [ "Direct SOX extraction/annotation case supplies annual hours without a denominator; recent human-reviewed audit-document case reports fifty percent less processing time. Exact ticket-workflow endpoints remain conditional analyst assumptions." ], "validation_needed": [ "Time matched manual and assisted preparation on the same fifty historical cases, with held-out adjudication", "Record attribute omissions, wrong evidence links, false passes and missing/conflicting approvals separately", "Measure preparer and independent auditor labor including correction and source re-performance; retain the same evidential standard", "Test whether preparation fits the reference 35–42 hour budget per 100 baseline hours, while preserving the full auditor-review and correction effort." ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": "Partly indexed evidence and substantial editing leave 15 retrieval + 15 extraction + 12 drafting + 33 independent-review/correction hours: 75 remain, 25 saved.", "remaining_hours": 75, "saved_hours_per_100": 25 }, "fast": { "conditions": "Indexed evidence, consistent formats and reusable workpapers leave 10 retrieval + 15 extraction + 10 drafting + 30 independent-review hours: 65 remain, 35 saved.", "remaining_hours": 65, "saved_hours_per_100": 35 } }, "lower_case": "Partly indexed evidence and substantial editing leave 15 retrieval + 15 extraction + 12 drafting + 33 independent-review/correction hours: 75 remain, 25 saved.", "upper_case": "Indexed evidence, consistent formats and reusable workpapers leave 10 retrieval + 15 extraction + 10 drafting + 30 independent-review hours: 65 remain, 35 saved.", "near_automatic_sensitivity": { "label": "Near-automatic evidence preparation", "saving_percent": 65, "before_hours": [ 20, 30, 20, 30 ], "after_hours": [ 1, 2, 2, 30 ], "conditions": "Complete indexed evidence and reliable template population reduce preparation from 70 to 5 hours. All 30 independent-review hours remain. This is an unvalidated automation scenario, beyond the quantified case reference." } }, "AI-04": { "title": "Margin analysis assistant", "baseline": "The same recurring margin questions across three approved datasets and five dimensions, using existing validated calculations; retrieve results, explain drivers, supply context, verify and approve the source-linked answer.", "evidence_strength": "Conditional scenario informed by a recent variance-analysis case; no measured end-to-end margin-answer result.", "blocks": [ { "work": "Retrieve approved calculation outputs and sources", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; no numeric retrieval saving is claimed from the qualitative Natura case.", "source_ids": [ "planning-natura_sap" ], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Model-backed driver decomposition and first source-linked explanation", "baseline_hours": 45, "eligible_baseline_hours": [ 20, 45 ], "eligible_task_saving_rate_pct": [ 90, 90 ], "after_hours": [ 27, 4.5 ], "saving_rate_pct": [ 40, 90 ], "basis": "The constrained case has 20 of these 45 baseline hours eligible for a configured recurring analysis; the remaining 25 hours run manually. The fast case covers all 45 hours. PwC's 90% stage result is applied unchanged only to eligible hours: after = (45 - eligible_hours) + eligible_hours * 0.10. Eligibility is an analyst assumption about work hours, not the proportion of questions or a measured adoption rate.", "source_ids": [ "planning-pwc_cfo_advisor" ], "reduction_percent": [ 40.0, 90.0 ] }, { "work": "Investigate missing business context and assess causal explanation", "baseline_hours": 20, "after_hours": [ 20, 20 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; retain all contextual judgment. This is operating analysis, not data repair or finance-model development.", "source_ids": [ "planning-natura_sap" ], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Verify figures and citations, correct answers and approve", "baseline_hours": 25, "incremental_correction_hours": [ 3, 0 ], "after_hours": [ 28, 25 ], "saving_rate_pct": [ -12, 0 ], "basis": "Retain all 25 baseline review hours. The constrained scenario explicitly adds three correction hours per 100 baseline hours; the fast scenario fits corrections inside existing review. Neither correction assumption is an empirical result or a discount to PwC's stage rate.", "source_ids": [], "reduction_percent": [ -12.0, 0.0 ] } ], "source_ids": [ "planning-pwc_cfo_advisor", "planning-natura_sap", "planning-bcg_finance_roi_2025" ], "range_percent": [ 15, 45 ], "calculated_percent": [ 15, 40.5 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [], "uncertainties": [ "The 45/20/25/10 work mix and 20-versus-45 eligible hours are unmeasured scenario assumptions. PwC's rate also accompanies process/tool/report changes; equal-output transfer is conditional. Natura validates the use case without quantifying savings. The display range is not a technology ceiling." ], "validation_needed": [ "Time the same 20 reference questions before/after, separately logging the four blocks and corrections.", "Classify eligible recurring work by baseline human hours before applying the saving formula.", "Require equivalent figures, citations, explanation support and output volume; include failed answers and manual fallback." ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": "Twenty baseline hours of recurring bridges/decompositions are within configured coverage; the rest of the analytical stage remains manual. All baseline context and validation remain; three hours of extra corrections are charged.", "remaining_hours": 85, "saved_hours_per_100": 15 }, "fast": { "conditions": "All 45 preparation hours are recurring, model-backed questions with stable definitions, supported drivers and approved sources. The configured stage reaches the reported PwC effect, while the full 45 hours of context plus review remain.", "remaining_hours": 59.5, "saved_hours_per_100": 40.5 } }, "lower_case": "Twenty baseline hours of recurring bridges/decompositions are within configured coverage; the rest of the analytical stage remains manual. All baseline context and validation remain; three hours of extra corrections are charged.", "upper_case": "All 45 preparation hours are recurring, model-backed questions with stable definitions, supported drivers and approved sources. The configured stage reaches the reported PwC effect, while the full 45 hours of context plus review remain.", "higher_coverage_sensitivity": { "purpose": "Expose a faster case when a larger share of the same scoped question workload is configured; the displayed range is not a technology ceiling.", "baseline_task_mix_hours": { "retrieve_approved_outputs": 10, "eligible_model_backed_variance_preparation_and_first_explanation": 55, "business_context_and_causal_investigation": 10, "verification_corrections_and_approval": 25 }, "eligible_task_saving_pct": 90, "after_hours": 50.5, "saving_pct": 49.5, "outward_rounded_upper_pct": 50, "arithmetic": "10 + 55*0.10 + 10 + 25 = 50.5; (100-50.5)/100 = 49.5%", "condition": "An organization whose same approved-question workload spends 55 baseline hours on recurring, configured model-backed variance preparation and only 10 on additional context could reach this scenario. All 25 baseline checking/correction/approval hours remain. This is a different explicit baseline mix, not a newly measured saving or added implementation scope.", "source_ids": [ "planning-pwc_cfo_advisor" ] } }, "AI-05": { "title": "Management reporting assistant", "baseline": "Prepare the same monthly 10-page management pack and ten variance narratives from two approved sources for one audience, including numerical checking, contextual edits, corrections and release approval.", "evidence_strength": "Recent named customer self-estimate closely matches commentary and review; the complete pack percentage still depends on an assumed task mix.", "blocks": [ { "work": "Refresh approved inputs and perform source/tie-out checks", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption. Sources are usable already; retain routine checks and claim no data-transformation benefit.", "source_ids": [], "reduction_percent": [ 0.0, 0.0 ] }, { "work": "Variance commentary, pack narrative/visual placement and finance review/correction", "baseline_hours": 80, "baseline_composition_hours": { "analyze_and_draft_ten_variances": 35, "assemble_same_ten_page_pack": 20, "finance_context_numeric_source_layout_checks_and_corrections": 25 }, "after_hours": [ 56, 40 ], "saving_rate_pct": [ 30, 50 ], "basis": "Apply Docker's reported 30-50% commentary/review estimate to this combined block. The 35/20/25 split makes the 80-hour analyst work-share assumption auditable; it is not a set of independently measured task effects. Source and scenario both retain human refinement, and review/corrections are included once inside the bundle.", "source_ids": [ "planning-docker_pigment" ], "reduction_percent": [ 30.0, 50.0 ] }, { "work": "Final accountable approval and release of the checked pack", "baseline_hours": 10, "after_hours": [ 10, 10 ], "saving_rate_pct": [ 0, 0 ], "basis": "Analyst task-share assumption; retain the full approval/release step outside the customer commentary/review estimate.", "source_ids": [], "reduction_percent": [ 0.0, 0.0 ] } ], "source_ids": [ "planning-docker_pigment", "planning-pwc_cfo_advisor", "planning-clickup_pigment" ], "range_percent": [ 20, 40 ], "calculated_percent": [ 24, 40 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [], "uncertainties": [ "The 80% commentary/review share is an analyst scenario assumption, not a finding that all reporting teams spend 80% of time there. Docker's 30-50% is one customer's self-estimate and does not establish the full pack rate. ClickUp lacks a baseline denominator; PwC's 90% covers a configured variance stage. Neither is averaged into the model." ], "validation_needed": [ "Log three comparable monthly reporting cycles at the same ten pages and ten narratives.", "Record source checks, combined commentary/assembly/review, final approval and every correction/fallback separately.", "Replace the assumed 80-hour share with observed baseline hours; test whether net block saving meets the scenario after required quality checks." ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": "The same pack is produced; the combined drafting, packaging and review stage achieves the lower Docker self-estimate. Normal business-context edits, numerical/source checks and correction effort fit inside its 56 remaining hours; source checks and final approval retain 20 hours.", "remaining_hours": 76, "saved_hours_per_100": 24 }, "fast": { "conditions": "Stable pack format and supported recurring variances allow the combined block to achieve the upper Docker self-estimate. Review and corrections fit inside its 40 remaining hours; all 20 hours of source checks and final approval remain.", "remaining_hours": 60, "saved_hours_per_100": 40 } }, "lower_case": "The same pack is produced; the combined drafting, packaging and review stage achieves the lower Docker self-estimate. Normal business-context edits, numerical/source checks and correction effort fit inside its 56 remaining hours; source checks and final approval retain 20 hours.", "upper_case": "Stable pack format and supported recurring variances allow the combined block to achieve the upper Docker self-estimate. Review and corrections fit inside its 40 remaining hours; all 20 hours of source checks and final approval remain.", "higher_coverage_sensitivity": { "purpose": "Expose a faster case where existing templates make pack assembly short and configured variance preparation dominates baseline hours; neither Docker 50% nor the displayed 40% workflow endpoint is a capability limit.", "baseline_task_mix_hours": { "source_checks": 10, "eligible_configured_variance_analysis_and_first_drafts": 65, "existing_template_pack_assembly": 5, "finance_context_checks_and_corrections": 15, "final_release_approval": 5 }, "eligible_task_saving_pct": 90, "after_hours": 41.5, "saving_pct": 58.5, "outward_rounded_upper_pct": 60, "arithmetic": "10 + 65*0.10 + 5 + 15 + 5 = 41.5; (100-41.5)/100 = 58.5%", "condition": "For the same ten narratives and ten-page pack, 65 baseline hours are spent on recurring configured variance analysis/first drafts, while an already efficient template requires five assembly hours. Preserve all 15 human context/check/correction hours plus final approval. Task shares are analyst assumptions to be replaced by a local time study; PwC 90% still applies only to the named eligible stage.", "source_ids": [ "planning-pwc_cfo_advisor" ] }, "additional_fast_configuration_sensitivity": { "purpose": "Show the PwC result without using it as the whole-pack rate or an empirical cap.", "baseline_eligible_variance_preparation_hours": 45, "eligible_task_saving_pct": 90, "all_other_work_hours_retained": 55, "after_hours": 59.5, "saving_pct": 40.5, "arithmetic": "45*0.10 + 55 = 59.5; 40.5% saved", "assumption": "An alternative task mix dedicates 45 baseline hours to configured variance analysis and first drafts; all 55 other hours, including corrections and approvals, stay unchanged. Do not add this result to the Docker scenario.", "source_ids": [ "planning-pwc_cfo_advisor" ] } }, "AI-06": { "title": "Finance knowledge assistant", "baseline": "Find metric definitions/approved values, locate datasets, retrieve policies and precedents, prepare cited answers, check authority/current version and applicability, and resolve unsupported questions. Count all requester and finance-owner human work once. No downstream transactions or decisions; ongoing knowledge maintenance is separate.", "evidence_strength": "A 2026 knowledge-search case reports saved minutes without a baseline duration. Reference lookup duration, task mix, coverage and answer-success rates are explicit assumptions; no complete finance-knowledge workflow is measured.", "blocks": [ { "work": "Locate authoritative material: metric definitions/approved values, datasets, policies and precedents", "baseline_hours": 60, "coverage_fraction": [ 0.7, 0.95 ], "successful_fraction_of_covered_work": [ 0.85, 0.95 ], "successful_work_reduction_fraction": [ 0.8, 0.95 ], "basis": "FM estimates 6–10 minutes saved per search, without a baseline duration. This reference assumes a successful manual lookup takes 10 minutes and accepted assisted lookup takes 2 or 0.5 minutes: 8 or 9.5 minutes saved. Those timings, coverage (70%/95%) and success on covered work (85%/95%) are assumptions. They are not FM-measured percentages or precision/recall conversions. Uncovered and unsuccessful work retains its baseline time; all existing checking is separate.", "source_ids": [ "planning-fm_knowledge_2026" ], "after_hours": [ 31.44, 8.5575 ], "reduction_percent": [ 47.6, 85.7375 ] }, { "work": "Assemble and communicate a concise cited answer", "baseline_hours": 20, "coverage_fraction": [ 0.7, 0.95 ], "successful_fraction_of_covered_work": [ 0.85, 0.95 ], "successful_work_reduction_fraction": [ 0.6, 0.9 ], "basis": "Assumed residual human preparation of 40% or 10% for accepted grounded drafts. Template and recurring-answer conditions drive the high case. No cited source measures these exact writing reductions; excluded Noy/Zhang result is not reused.", "source_ids": [ "planning-amazon_regulatory_rag_2026", "planning-tr_copeland_buhl_2026" ], "after_hours": [ 12.86, 3.755 ], "reduction_percent": [ 35.7, 81.225 ] }, { "work": "Check source/version and applicability; resolve unsupported questions and approve answers", "baseline_hours": 20, "reduction_percent": [ 0.0, 0.0 ], "basis": "Preserve all existing human checking, judgment and escalation labor. This is an assumed baseline share, not an observed finance service time study. Count requester and finance-owner work once.", "after_hours": [ 20, 20 ], "source_ids": [] }, { "work": "Additional prompting, failed attempts and AI-specific correction", "baseline_hours": 0, "added_hours": [ 4, 1 ], "basis": "Explicit assumed extra labor per 100 baseline human hours, including attempts that fall back to the original work. In the factors above, unsuccessful/uncovered work retains its full baseline labor. Separate corpus maintenance is excluded.", "after_hours": [ 4, 1 ], "source_ids": [], "reduction_percent": null } ], "source_ids": [ "planning-fm_knowledge_2026", "planning-amazon_regulatory_rag_2026", "planning-tr_copeland_buhl_2026" ], "range_percent": [ 30, 70 ], "calculated_percent": [ 31.7, 66.6875 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [ "Poor coverage, failed retrieval, a weak manual baseline or lengthy numerical/source reconciliation can yield less than 30% or negative savings.", "A service dominated by simple well-covered metric/dataset/policy lookups with very short retained checking can exceed 70%; report its measured task mix rather than raising a universal cap." ], "uncertainties": [ "A 2026 knowledge-search case reports saved minutes without a baseline duration. Reference lookup duration, task mix, coverage and answer-success rates are explicit assumptions; no complete finance-knowledge workflow is measured." ], "validation_needed": [ "Time matched reference questions across metrics, datasets, policies and precedents, including ambiguous, restricted, conflicting and unsupported cases.", "Record total active human minutes, including requester prompting, source verification, finance-owner escalation, failures and corrections; report model wait separately.", "Estimate time-weighted content coverage, successful-answer rate and stage reductions independently, then compare equal answer accuracy/currentness.", "Measure recurring corpus/access/update maintenance separately." ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": "70% coverage and 85% success on preparation work; accepted lookup 80% faster and cited-answer preparation 60% faster; all 20 existing checking/escalation hours retained; add 4 failure/correction hours. Net 31.7 hours saved.", "remaining_hours": 68.3, "saved_hours_per_100": 31.7 }, "fast": { "conditions": "95% coverage and 95% success on preparation work; accepted lookup 95% faster and cited-answer preparation 90% faster; all 20 existing checking/escalation hours retained; add 1 failure/correction hour. Net 66.6875 hours saved.", "remaining_hours": 33.3125, "saved_hours_per_100": 66.6875 } }, "lower_case": "70% coverage and 85% success on preparation work; accepted lookup 80% faster and cited-answer preparation 60% faster; all 20 existing checking/escalation hours retained; add 4 failure/correction hours. Net 31.7 hours saved.", "upper_case": "95% coverage and 95% success on preparation work; accepted lookup 95% faster and cited-answer preparation 90% faster; all 20 existing checking/escalation hours retained; add 1 failure/correction hour. Net 66.6875 hours saved.", "direct_lookup_high_speed_example": { "status": "Illustrative assumption; exact before/after not observed in cited studies", "before_human_minutes": 10, "before_preparation_minutes": 9.5, "before_existing_check_minutes": 0.5, "after_human_minutes": 1, "after_preparation_minutes": 0.5, "after_existing_check_minutes": 0.5, "calculated_saving_percent": 90, "conditions": "One well-covered, unambiguous question; current authorized source; short answer; accepted on first attempt; source checking takes 30 seconds in both methods. A 20-minute baseline with the same 1-minute assisted total would save 95%.", "interpretation": "90%+ task-level savings are entirely possible under these assumptions. The mixed-workflow scenario is lower because of its assumed work mix, coverage and full retained checking, and does not impose a 70% cap on every service." } }, "AI-07": { "title": "Tax monitoring assistant", "baseline": "Monitor five official sources for one tax regime across two jurisdictions, including checks with no relevant change; triage and read updates, extract dates, draft summaries and obtain expert applicability approval. Same update volume/coverage and approved output. No filing, autonomous advice or policy implementation.", "evidence_strength": "Late-2025 self-reported tax-research survey plus named 2026 tax-research implementation account; no measured end-to-end tax monitoring labor result. Conditional estimate.", "blocks": [ { "work": "Check the five official sources, including no-change checks; detect and triage updates", "baseline_hours": 30, "coverage_fraction": [ 0.8, 1.0 ], "successful_fraction_of_covered_work": [ 0.9, 0.98 ], "successful_work_reduction_fraction": [ 0.8, 0.95 ], "basis": "Analyst operational assumptions, without an empirical scan-time anchor. Low: feeds/readable exports cover 80% of baseline scanning work, 90% succeeds and accepted automation removes 80% of its human time. High: all five sources work reliably, 98% succeeds, accepted scan/triage needs 5% of its former time. Human spot checks and false-positive review remain in residual scan time; no-change checks belong to the denominator.", "source_ids": [ "planning-amazon_regulatory_rag_2026" ], "after_hours": [ 12.72, 2.07 ], "reduction_percent": [ 57.6, 93.1 ] }, { "work": "Read relevant updates, extract publication/effective dates and draft cited summaries", "baseline_hours": 30, "basis": "Apply the adjacent reported 32% research saving to the entire 30-hour read/extract/draft block in the constrained case. The fast case assumes 65% less time, consistent with the conditional same-output interpretation of Copeland Buhl’s 2–3-times-faster research claim. Both are scope transfers to monitoring, not observed monitoring effects. No additional coverage or success discount is applied to these real-world research results.", "source_ids": [ "planning-tr_cocounsel_corporate", "planning-tr_copeland_buhl_2026" ], "after_hours": [ 20.4, 10.5 ], "reduction_percent": [ 32.0, 65.0 ] }, { "work": "Check legal applicability, effective dates and completeness; correct and approve summaries", "baseline_hours": 40, "reduction_percent": [ 0.0, 0.0 ], "basis": "Retain all expert review and existing correction labor. Forty hours is an assumed monitoring-workflow share, not a measured result. The PLAT benchmark supplies accuracy context rather than a review-hours factor. No filing or autonomous advice is included.", "after_hours": [ 40, 40 ], "source_ids": [] }, { "work": "Additional failed-attempt, false-positive and AI-specific correction labor", "baseline_hours": 0, "added_hours": [ 3, 1 ], "basis": "Assumed extra human work per 100 baseline hours, including investigating missed or misclassified updates and checking failures. Unsuccessful/uncovered work also keeps its full baseline labor above. Ongoing source-feed maintenance is separate.", "after_hours": [ 3, 1 ], "source_ids": [], "reduction_percent": null } ], "source_ids": [ "planning-tr_cocounsel_corporate", "planning-tr_copeland_buhl_2026", "planning-amazon_regulatory_rag_2026", "planning-tax_plat" ], "range_percent": [ 20, 50 ], "calculated_percent": [ 23.88, 46.43 ], "scenario_method": "before_after_hours", "estimate_status": "analyst_scenario", "assumptions": [ "Baseline task hours, eligible coverage and remaining human-work budgets are scenario assumptions.", "Published task results inform the scenario; no source establishes the complete workflow range.", "Same work volume and required output; manual fallback and corrections are included. Separate upkeep is excluded." ], "outside_band": [ "Source failures, weak archives, heavy legal interpretation or long review/correction can put results below 20% or below zero.", "More than 45% is possible when routine source checks/drafting dominate baseline time and review is a smaller share; 40% retained review here creates a scenario-specific limit, not a universal tax-monitoring cap." ], "uncertainties": [ "Late-2025 self-reported tax-research survey plus named 2026 tax-research implementation account; no measured end-to-end tax monitoring labor result. Conditional estimate." ], "validation_needed": [ "Time scan work on all five sources, including no-change days and irrelevant updates; compare against the actual existing alert workflow.", "Use the same historical and live update sets and require equivalent recall; missing a relevant update is not a saving.", "Time human triage, date extraction, drafting, applicability approval, rejected summaries and corrections separately.", "Track sources/updates that fall back to manual processing and record the extra failed-attempt labor.", "Measure source-feed upkeep, test updates and classification maintenance separately." ], "reasoning": "Re-estimated from retained references with explicit before/after human-hour budgets.", "research_file": "reestimation-2026-10-25/model.json", "scenarios": { "constrained": { "conditions": "Scanning has 80% coverage, 90% success and 80% accepted-task saving; read/draft work uses the reported 32% research effect without another coverage discount. Retain all 40 expert-review hours and add 3 correction hours.", "remaining_hours": 76.12, "saved_hours_per_100": 23.88 }, "fast": { "conditions": "All source checks are covered, with 98% success and 95% accepted-task saving; the research block assumes 65% less time. Retain all 40 expert-review hours and add 1 correction hour.", "remaining_hours": 53.57, "saved_hours_per_100": 46.43 } }, "lower_case": "Scanning has 80% coverage, 90% success and 80% accepted-task saving; read/draft work uses the reported 32% research effect without another coverage discount. Retain all 40 expert-review hours and add 3 correction hours.", "upper_case": "All source checks are covered, with 98% success and 95% accepted-task saving; the research block assumes 65% less time. Retain all 40 expert-review hours and add 1 correction hour." } }, "short_references": { "accounting-microsoft_collections_2026_insidetrack": { "label": "Microsoft collections (2026)", "text": "up to 60% less inquiry-handling time" }, "accounting-ted_taylor_2026": { "label": "Ted Taylor / OpenAI (2026)", "text": "monthly cash reconciliation: 3 hours → 5–10 minutes" }, "accounting-afm_controls": { "label": "AFM audit guidance (2025)", "text": "" }, "planning-docker_pigment": { "label": "Docker / Pigment (2026)", "text": "30–50% less close-commentary and review time, estimated" }, "planning-bcg_finance_roi_2025": { "label": "BCG finance case (2025)", "text": "50% shorter report-generation time" }, "planning-amazon_finance_discovery_2025": { "label": "Amazon Finance / AWS (2025)", "text": "search: 30% less time; discovery: 45–60 → 5–10 minutes" }, "planning-tr_copeland_buhl_2026": { "label": "Copeland Buhl / Thomson Reuters (2026)", "text": "research reported 2–3 times faster" }, "planning-tr_tax_launch_confirmation_2026": { "label": "Thomson Reuters product update (2026)", "text": "" } } }