{
  "meta": {
    "title": "AI Process Integration ROI Library",
    "as_of": "2026-09-25",
    "doc_url": "https://claude.ai/code/artifact/fc67e363-83ee-4810-b701-544cc25ec12c",
    "evidence_bar": "A case enters only if a named company (or a rigorous study) states a quantified before/after outcome for a specific business process. Anonymous claims and vendor aggregates without a named customer are excluded.",
    "confidence_tiers": {
      "High": "Peer-reviewed or randomized field study, or a company figure reproduced consistently over time",
      "Medium-high": "Named company, named executive, stated baseline and timeframe; vendor may host the story",
      "Medium": "Named company and metric but missing baseline or timeframe; self-reported",
      "Low-medium": "Pilot-stage, projected, or survey-of-users figures"
    }
  },
  "summary": [
    "Roughly 88% of large organizations use AI somewhere, yet only 37% attribute any EBIT impact to it and about 6% attribute 5% or more (McKinsey, Aug 2026).",
    "The companies that capture value pick a narrow, high-volume process with a clean baseline, redesign the workflow (73% of high performers vs 25% of others), and measure against the pre-AI baseline for at least a quarter.",
    "Back-office finance and operations produce the most reliable quantified returns; budgets skew to sales and marketing despite better ROI in operations and finance (MIT NANDA).",
    "Headcount headlines have a poor track record: Klarna, Commonwealth Bank and Duolingo each walked back or reframed AI-for-people announcements within a year.",
    "Buy or embed for the first wave (vendor tools succeed about twice as often as internal builds), expect 3-12 month payback on good cases and nothing on pilots without a baseline."
  ],
  "changelog": [
    {"date": "2026-09-25", "change": "Monthly refresh: no new qualifying named-company case or macro survey edition confirmed via primary-source verification in the 2026-09-18–2026-09-25 window (this session's network access blocked WebFetch on most source domains, including mckinsey.com, bcg.com, gartner.com, nber.org, coupa.com, appzen.com, vic.ai, pactum.com and most trade press; candidates below need re-verification). Sources checked: McKinsey State of AI, Stanford AI Index (Economy), BCG AI value reports, Deloitte State of AI in the Enterprise, Gartner newsroom, U.S. Census BTOS, Ardent Partners Payables Place, APQC, NBER/SSRN, Microsoft/AWS/Google Cloud customer stories, HighRadius/Coupa/AppZen/Rossum/Vic.ai/Pactum/Zip case hubs, CFO Dive, Customer Experience Dive, The Register AI, JPMorgan/Amazon/Salesforce/Klarna/IBM earnings coverage.", "section": "All"},
    {"date": "2026-09-25", "change": "Reversal/update caveats added to Klarna, Commonwealth Bank of Australia and Salesforce cases (2026 developments: CBA's Jul 2026 redundancies and Fair Work Commission dispute, Salesforce's Jul 2026 KeyBanc downgrade and Aug 2026 Agentforce revenue disclosure, Klarna's evolving headcount-equivalent figure)", "section": "Customer-facing & revenue"},
    {"date": "2026-09-25", "change": "Reversal/update caveats added to JPMorgan Chase (Dimon's Jul 2026 Q2 earnings-call remarks) and IBM Client Zero/FP&A cases (unreconciled $4.5B aggregate figure flagged)", "section": "Finance & back office"},
    {"date": "2026-09-25", "change": "Added peer-reviewed-publication citations and a countervailing-study caveat to the METR RCT, Choi & Xie, Brynjolfsson/Li/Raymond and BCG/Harvard Dell'Acqua cases; re-verified last_verified date on 23 High/Medium-high confidence cases with no other changes found", "section": "People & IT"},
    {"date": "2026-09-25", "change": "Automated monthly refresh enabled; site data file created", "section": "—"},
    {"date": "2026-09-18", "change": "Initial library: 9 macro studies, 45 cases across finance, customer-facing, operations, and people/IT", "section": "All"}
  ],
  "benchmarks": [
    {"metric": "AP cost per invoice", "typical": "$12.88", "best_in_class": "$2.78", "source": "Ardent Partners AP Metrics That Matter 2025 (N=212)", "url": "https://www.datocms-assets.com/80283/1744404602-ardent-partners-ap-metrics-that-matter-in-2025-pagero-final.pdf"},
    {"metric": "AP processing time", "typical": "17.4 days", "best_in_class": "3.1 days", "source": "Ardent Partners AP Metrics That Matter 2025", "url": "https://www.datocms-assets.com/80283/1744404602-ardent-partners-ap-metrics-that-matter-in-2025-pagero-final.pdf"},
    {"metric": "AP straight-through rate", "typical": "23.4%", "best_in_class": "49.2%", "source": "Ardent Partners AP Metrics That Matter 2025", "url": "https://www.datocms-assets.com/80283/1744404602-ardent-partners-ap-metrics-that-matter-in-2025-pagero-final.pdf"},
    {"metric": "AP exception rate", "typical": "22.0%", "best_in_class": "9.0%", "source": "Ardent Partners AP Metrics That Matter 2025", "url": "https://www.datocms-assets.com/80283/1744404602-ardent-partners-ap-metrics-that-matter-in-2025-pagero-final.pdf"},
    {"metric": "AP cost per invoice (all-firm average)", "typical": "$9.84 / 8.2 days", "best_in_class": "79% lower cost, 79% faster", "source": "Ardent State of ePayables 2025", "url": "https://payablesplace.ardentpartners.com/2026/01/state-of-epayables-part-nine-ap-benchmarks-and-best-in-class-performance/"},
    {"metric": "AP cost per invoice (median, N=5,846)", "typical": "$6.00", "best_in_class": "—", "source": "APQC Open Standards Benchmarking", "url": "https://www.apqc.org/resources/benchmarking/open-standards-benchmarking/measures/total-cost-perform-process-process-19"},
    {"metric": "AR: share of AI users reducing DSO", "typical": "99% reduced DSO; 75% by 6+ days", "best_in_class": "—", "source": "Wakefield for Billtrust (N=500, >$250M firms; vendor-commissioned)", "url": "https://www.billtrust.com/news/study-finds-ai-in-accounts-receivable-reduces-dso"}
  ],
  "macro": [
    {"study": "McKinsey State of AI", "date": "2026-08", "sample": "1,719 respondents, 97 countries, 36% from $1B+ firms", "adoption": "~9 in 10 use AI in at least one function; 44% scaling enterprise-wide", "return": "37% attribute any EBIT impact (flat vs 2025); 6% high performers (5%+ of EBIT)", "finding": "40% of $1B+ firms scaling agents (up from 27%); 73% of high performers redesigned workflows vs 25%; 20% constrained by AI operating costs", "url": "https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"},
    {"study": "Stanford AI Index, Economy chapter", "date": "2026-04", "sample": "Aggregates McKinsey and academic studies", "adoption": "88% of organizations use AI (78% in 2024); 79% use gen AI", "return": "Cost decreases most cited in software engineering and manufacturing (56%); revenue gains most cited in marketing and sales (67%)", "finding": "Global corporate AI investment $581.7B in 2025, up ~130%; most reported gains under 10% cost or 5% revenue", "url": "https://hai.stanford.edu/assets/files/ai_index_report_2026_chapter_4_economy.pdf"},
    {"study": "BCG Widening AI Value Gap", "date": "2025-09", "sample": "1,250 senior executives, 9 industries", "adoption": "—", "return": "5% future-built, 35% scalers, 60% laggards with minimal gains", "finding": "Leaders 1.7x revenue growth, 1.6x EBIT margin, 3.6x 3-yr TSR; 70% of AI value in core functions, IT only 13%", "url": "https://www.bcg.com/press/30september2025-ai-leaders-outpace-laggards-revenue-growth-cost-savings"},
    {"study": "BCG Building business value with AI investment", "date": "2026", "sample": "Follow-on analysis", "adoption": "Leaders invest 1.7% of revenue in AI vs 0.8%", "return": "Leaders report 3x cost reduction and 2.7x ROIC", "finding": "10-20-70 rule: 10% algorithms, 20% technology and data, 70% people, process and behavior change", "url": "https://www.bcg.com/publications/2026/building-business-value-with-ai-investment"},
    {"study": "MIT NANDA GenAI Divide", "date": "2025-07", "sample": "300+ disclosed initiatives, 52 interviews, 153 surveys", "adoption": "90% of workers use personal AI tools daily; 40% of firms have official subscriptions", "return": "95% see no measurable return; ~5% of integrated pilots extract millions in value", "finding": "Vendor-built tools succeed about twice as often as internal builds; $2-10M/yr documented replacing outsourced support and document review", "url": "https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx"},
    {"study": "Deloitte State of AI in the Enterprise", "date": "2026-01", "sample": "3,235 leaders, 24 countries, Aug-Sep 2025", "adoption": "60% of workers have sanctioned AI tools (under 40% prior year)", "return": "Only 25% have moved 40%+ of pilots into production", "finding": "30% redesigning key processes; 37% surface-level use; 21% have mature agent governance", "url": "https://www.deloitte.com/us/en/about/press-room/state-of-ai-report-2026.html"},
    {"study": "Gartner finance AI survey", "date": "2025-11", "sample": "183 CFOs and senior finance leaders", "adoption": "59% of finance functions use AI; AP automation 37%, anomaly detection 34%", "return": "91% report low or moderate impact initially", "finding": "Finance adoption steady rather than accelerating", "url": "https://www.gartner.com/en/newsroom/press-releases/2025-11-18-gartner-survey-shows-finance-ai-adoption-remains-steady-in-2025"},
    {"study": "Gartner agentic AI prediction", "date": "2025-06", "sample": "3,412 webinar poll", "adoption": "19% significant agent investment, 42% conservative", "return": "—", "finding": "40%+ of agentic AI projects canceled by end-2027; only ~130 of thousands of agent vendors are real", "url": "https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027"},
    {"study": "U.S. Census BTOS", "date": "2026-05", "sample": "Biweekly national survey of employer firms", "adoption": "17-20% of all firms; 37% of firms with 250+ employees; 39.7% Information, 33.9% Finance and Insurance, ~14% Retail", "return": "—", "finding": "Adoption rising only among firms with 20+ employees; Fed (Apr 2026) puts worker-level gen AI use at ~41%", "url": "https://www.census.gov/library/stories/2026/05/ai-use-businesses.html"},
    {"study": "Humlum & Vestergaard, NBER", "date": "2025-04", "sample": "25,000 workers, 7,000 Danish workplaces, 11 occupations", "adoption": "Chatbots widespread; most employers encourage use", "return": "Average time savings 2.8%; no significant effect on earnings or hours", "finding": "Individual chatbot use without process change yields small, unmonetized gains", "url": "https://marginalrevolution.com/marginalrevolution/2025/04/large-language-models-small-labor-market-effects.html"}
  ],
  "cases": [
    {"id": 1, "company": "GameStop", "industry": "Retail", "size": "~$4B", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "Invoice capture, matching and e-invoicing", "tool": "Coupa", "result": "750,000 hand-keyed invoices/yr → 82% higher first-time match, 70% shorter processing time, 20% lower AP headcount, 66% more e-invoicing", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "No baseline days or period stated", "url": "https://www.coupa.com/customers/gamestop/", "last_verified": "2026-09-18"},
    {"id": 2, "company": "Veolia shared service center", "industry": "Utilities", "size": "€40B+ group, 30 entities", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "AI invoice capture", "tool": "Rossum", "result": "87.5% less time per document, 8x throughput, 90% automation rate, manual workload down 90%", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "No baseline hours or timeframe", "url": "https://rossum.ai/customer-stories/veolia/", "last_verified": "2026-09-18"},
    {"id": 3, "company": "Fugro", "industry": "Geo-data services", "size": "~€2B", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "AI invoice capture", "tool": "Rossum", "result": "2 min → 35 sec per invoice on 300,000 invoices/yr (70% faster)", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://rossum.ai/customer-stories/", "last_verified": "2026-09-18"},
    {"id": 4, "company": "Diesel Direct", "industry": "Fuel distribution", "size": "Private", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "Autonomous invoice processing on Great Plains", "tool": "Vic.ai", "result": "65% less processing time, 84% no-touch rate, 99% accuracy at month six, AP close 4 → 3 days, 3 planned FTE → 2.5", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "Company size unclear, may be under $50M", "url": "https://vic.ai/resources/case-study-large-us-distribution-company-processes-76-percent-of-invoices-without-any-human-touch", "last_verified": "2026-09-18"},
    {"id": 5, "company": "Purple", "industry": "Mattress manufacturing", "size": "~$500M", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "AI-assisted AP", "tool": "Stampli", "result": "Invoice processing 8 → 3 days; backlog cut 50% in 3 months; approvals 63% faster", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://www.stampli.com/blog/ap-automation/accounts-payable-automation-case-studies/", "last_verified": "2026-09-18"},
    {"id": 6, "company": "Centerfield", "industry": "Adtech", "size": "1,000+ staff", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "AP and global payments", "tool": "Tipalti", "result": "20 weeks of AP work eliminated per year; close shortened 12 days", "source_type": "Vendor-reported", "confidence": "Low-medium", "caveats": "AI share of the gain unclear", "url": "https://tipalti.com/blog/accounts-payable-automation-case-studies/", "last_verified": "2026-09-18"},
    {"id": 7, "company": "Wolt / Eurofins / Adyen / Eurowag", "industry": "Various (delivery, labs, payments, fleet)", "size": "$50M+ each", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "AI invoice capture", "tool": "Rossum", "result": "Wolt 60% straight-through on 50,000 invoices/month; Eurofins +30% AP productivity across 21 countries; Adyen 70% automation across 23 countries; Eurowag 70% across 60 entities", "source_type": "Vendor-reported", "confidence": "Low-medium", "caveats": "Headline claims, no baselines", "url": "https://rossum.ai/customer-stories/", "last_verified": "2026-09-18"},
    {"id": 8, "company": "IBM (Client Zero)", "industry": "IT services", "size": "$62B", "function": "Finance & back office", "subprocess": "Accounts payable", "ai": "Journal processing, invoice review and procurement agents", "tool": "watsonx", "result": "~90% projected cycle-time reduction in journals, ~$600K possible annual savings; 26,000 procurement hours saved/yr; 95%+ first-pass invoice review accuracy across 18 of 19 formats", "source_type": "Company statement", "confidence": "Medium", "caveats": "'Projected' and 'possible' wording on finance items. A Jan 2026 press report (Federal News Network) cites a larger aggregate 'Client Zero' figure of $4.5B in annualized productivity savings over three years; relationship to this specific AP/journal figure is unclear and unverified against a primary source.", "url": "https://www.ibm.com/case-studies/ibm-client-zero", "last_verified": "2026-09-25"},
    {"id": 9, "company": "Danone North America", "industry": "Food", "size": "$8B+", "function": "Finance & back office", "subprocess": "Accounts receivable & deductions", "ai": "AI deductions management on 1.1M claims/yr and touchless cash application", "tool": "HighRadius", "result": "$25.5M/yr invalid deductions recovered; analyst productivity +75%; deduction days 45 → 20; cash-app baseline 268 hours/week manual posting → 100% touchless, 75% lower cost", "source_type": "Vendor-reported", "confidence": "Medium-high", "caveats": "Named director quoted, baselines given, multi-year", "url": "https://www.highradius.com/resources/case-studies/danone/", "last_verified": "2026-09-25"},
    {"id": 10, "company": "DXP Enterprises", "industry": "Industrial distribution", "size": "$1B+", "function": "Finance & back office", "subprocess": "Accounts receivable & collections", "ai": "AI-prioritized collections worklist for 15,000+ customers, remittance prediction", "tool": "HighRadius", "result": "DSO down 20 days; collector productivity +40%; $20M added cash flow; 90% straight-through cash application; 600K+ invoices auto-posted/yr", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "CFO quoted, no timeframe", "url": "https://www.highradius.com/resources/value-creation/dxp-collections/", "last_verified": "2026-09-18"},
    {"id": 11, "company": "Ferrero North America", "industry": "CPG", "size": "$12B group", "function": "Finance & back office", "subprocess": "Accounts receivable & collections", "ai": "AI worklist prioritization and touchless dunning", "tool": "HighRadius", "result": "Average days delinquent down 67%; DSO down 28%; 1,000+ hours/yr saved on correspondence", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://www.highradius.com/resources/case-studies/ferrero/", "last_verified": "2026-09-18"},
    {"id": 12, "company": "Yaskawa America", "industry": "Industrial automation", "size": "$3.6B", "function": "Finance & back office", "subprocess": "Accounts receivable & collections", "ai": "Collections prioritization and auto deduction coding", "tool": "HighRadius", "result": "DSO down 5.5 days; A/R productivity +60%; collection effectiveness ~90%; $12K/yr card fees saved", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://www.highradius.com/resources/case-studies/yaskawa/", "last_verified": "2026-09-18"},
    {"id": 13, "company": "Konica Minolta", "industry": "Technology", "size": "$8B+", "function": "Finance & back office", "subprocess": "Treasury & cash forecasting", "ai": "AI cash forecasting", "tool": "HighRadius", "result": "Manual effort 80+ hours/month → under 15 minutes; 98.6% forecast accuracy; cash volatility down 15%; $30M cash added; $1.6M/yr interest saved", "source_type": "Vendor-reported", "confidence": "Medium-high", "caveats": "Named treasury director; award-verified Dec 2024", "url": "https://www.highradius.com/resources/case-studies/konica-minolta/", "last_verified": "2026-09-25"},
    {"id": 14, "company": "Walmart Canada", "industry": "Retail", "size": "$600B+", "function": "Finance & back office", "subprocess": "Procurement & supplier negotiation", "ai": "Autonomous tail-spend supplier negotiation chatbot; 3-month pilot, 89 suppliers, 5 buyers", "tool": "Pactum", "result": "Agreement with 64% of suppliers (target 20%); 1.5% savings on negotiated spend; payment terms extended 35 days on average; 11 days per negotiation; expanded to US, Chile, South Africa and freight rates", "source_type": "HBR case co-authored with company and vendor", "confidence": "High", "caveats": "2022 pilot; Pactum now cites 3% average gain and 68% close rate across clients", "url": "https://talkinglogistics.com/2023/05/01/negotiating-with-a-chatbot-a-walmart-procurement-case-study/", "last_verified": "2026-09-25"},
    {"id": 15, "company": "Dollar Tree", "industry": "Retail", "size": "~$30B", "function": "Finance & back office", "subprocess": "Procurement & purchase orders", "ai": "AI procurement intake and orchestration", "tool": "Zip", "result": "Spend under oversight 13% → 40%+ of $5B non-product spend; cycle times down 70%; $100M potential savings identified", "source_type": "Press, vendor-supplied", "confidence": "Medium", "caveats": "'Potential' savings", "url": "https://www.digitalcommerce360.com/2025/12/29/zip-reports-6-billion-customer-savings-ai-procurement/", "last_verified": "2026-09-18"},
    {"id": 16, "company": "SUEZ", "industry": "Water / waste", "size": "€9B", "function": "Finance & back office", "subprocess": "Procurement & supplier negotiation", "ai": "P2P with AI-powered supplier negotiations", "tool": "Coupa", "result": "58% first-time invoice match; annual savings £6M → £15M; 3-month ROI on AI negotiations", "source_type": "Vendor-reported", "confidence": "Low-medium", "caveats": "From vendor roundup; primary page not verified", "url": "https://www.coupa.com/blog/ap-automation-case-studies/", "last_verified": "2026-09-18"},
    {"id": 17, "company": "Takeda", "industry": "Pharma", "size": "50,000+ staff, 63 countries", "function": "Finance & back office", "subprocess": "Expense audit", "ai": "AI expense audit at 100% coverage", "tool": "AppZen", "result": "400,000 audits/yr, 63% auto-approved, 4,000 hours saved per quarter, sampling → full coverage", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://www.appzen.com/resources/case-studies/how-takeda-is-transforming-global-expense-auditing-with-ai", "last_verified": "2026-09-25"},
    {"id": 18, "company": "Databricks", "industry": "Software", "size": "9,000 staff", "function": "Finance & back office", "subprocess": "Expense audit", "ai": "AI expense audit on Emburse", "tool": "AppZen", "result": "12 months: 136K audits, 72% auto-approved, 3,000 hours saved/yr, $483K wasteful spend removed", "source_type": "Vendor-reported", "confidence": "Medium-high", "caveats": "Named head of global T&E", "url": "https://www.appzen.com/databricks-improved-travel-expense-management/", "last_verified": "2026-09-25"},
    {"id": 19, "company": "Owens Corning", "industry": "Building materials", "size": "19,000+ staff", "function": "Finance & back office", "subprocess": "Expense audit", "ai": "AI expense audit on SAP Concur", "tool": "AppZen", "result": "74% auto-approval; 2,000 audit hours saved/yr; audit turnaround cut by one full day", "source_type": "Vendor-reported", "confidence": "Medium", "caveats": "", "url": "https://www.appzen.com/how-owens-corning-enhanced-audit-quality", "last_verified": "2026-09-18"},
    {"id": 20, "company": "U.S. Venture", "industry": "Energy / auto distribution", "size": "5,500+ staff", "function": "Finance & back office", "subprocess": "Close & reconciliation", "ai": "Bank and card statement reconciliation", "tool": "Microsoft 365 Copilot for Finance", "result": "30+ hours/month saved; 80% less time on US Bank reconciliations; 4-week setup, ~3 months in use as of Jan 2025", "source_type": "Vendor-reported", "confidence": "Medium-high", "caveats": "Named practitioner, specific process", "url": "https://www.microsoft.com/en/customers/story/20600-us-venture-microsoft-365-copilot-for-finance", "last_verified": "2026-09-25"},
    {"id": 21, "company": "Microsoft finance org", "industry": "Software", "size": "$245B", "function": "Finance & back office", "subprocess": "Close & reconciliation", "ai": "AR reconciliation and FP&A data reconciliation", "tool": "Copilot", "result": "20 minutes saved per account on AR reconciliation; weekly FP&A reconciliation 1-2 hours → 10 minutes", "source_type": "Press, self-reported", "confidence": "Medium", "caveats": "Vendor reporting on its own use", "url": "https://www.cfodive.com/news/microsoft-finance-copilot-generativeai-software/727176/", "last_verified": "2026-09-18"},
    {"id": 22, "company": "IBM FP&A", "industry": "IT services", "size": "$62B", "function": "Finance & back office", "subprocess": "FP&A", "ai": "AI-assisted planning and forecasting", "tool": "Planning Analytics, watsonx", "result": "~40% FP&A productivity gain since 2020; ~95% forecast accuracy; $200M+ annual business value claimed since 2023", "source_type": "Partner-published company claim", "confidence": "Low-medium", "caveats": "A 2026 press report (Federal News Network, Jan 2026) cites a larger 'Client Zero' aggregate figure of $4.5B in annualized productivity savings over three years; unclear whether this supersedes or sits alongside the $200M+ FP&A figure above (unverified against a primary source).", "url": "https://revelwood.com/what-happens-when-ibm-uses-its-own-tech-to-transform-finance/", "last_verified": "2026-09-25"},
    {"id": 23, "company": "Choi & Xie, Stanford GSB (JAR 2026)", "industry": "Accounting services / SMEs", "size": "277 accountants, 79 SME clients", "function": "Finance & back office", "subprocess": "Bookkeeping & close", "ai": "AI-assisted bookkeeping and close", "tool": "—", "result": "+18% weekly client support per SD of AI use (59% top vs bottom users); ~9% of accountant time shifted from data entry; monthly close 7.5 days shorter; ledger granularity +12%", "source_type": "Peer-reviewed", "confidence": "High", "caveats": "SME context, not $50M+ corporates. Now formally published as Choi & Xie, 'Human + AI in Accounting: Early Evidence from the Field,' Journal of Accounting Research (2026).", "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5240924", "last_verified": "2026-09-25"},
    {"id": 24, "company": "LogicMonitor", "industry": "Software", "size": "~$200M+", "function": "Finance & back office", "subprocess": "Contract review", "ai": "AI first-pass review of MSAs, NDAs and procurement contracts", "tool": "Unnamed", "result": "Initial review time down ~90%; $100K/month ROI claimed; 50-70% shorter review cycles", "source_type": "Gartner-published, company-supplied", "confidence": "Medium", "caveats": "Figures internally inconsistent (90% vs 50-70%)", "url": "https://www.gartner.com/en/legal-compliance/trends/ai-powered-contract-review-to-accelerate-legal-operations", "last_verified": "2026-09-18"},
    {"id": 25, "company": "JPMorgan Chase (COiN and enterprise AI)", "industry": "Banking", "size": "$180B", "function": "Finance & back office", "subprocess": "Contract review / enterprise operations", "ai": "Commercial loan agreement review; cash-flow intelligence; LLM Suite", "tool": "In-house", "result": "360,000 lawyer-hours/yr of review automated (2017); cash-flow tool cut manual work ~90%; ~150,000 weekly internal AI users, 450+ use cases; CEO cites ~$2B benefit on ~$2B spend (Oct 2025)", "source_type": "Press / CEO statement", "confidence": "Medium", "caveats": "Aggregate estimate, not audited; guides to at least 10% operations headcount decline. On the 14 Jul 2026 Q2 earnings call, Dimon added more specific detail — AI has cut 30-40% of headcount in 'some' units — while cautioning 'you don't uniquely benefit from AI' since competitors see similar cost reductions; CFO Jeremy Barnum flagged rising gen-AI compute costs in H2 2026 that will offset some labor savings. Core ~$2B figure not revised as of Sep 2026; Q3 2026 earnings (mid-Oct) not yet reported.", "url": "https://www.artificialintelligence-news.com/news/jpmorgan-chase-ai-strategy-2025/", "last_verified": "2026-09-25"},
    {"id": 26, "company": "Brynjolfsson, Li & Raymond (QJE 2025)", "industry": "Software (Fortune 500, anonymized)", "size": "5,179 support agents", "function": "Customer-facing & revenue", "subprocess": "Customer support", "ai": "Chat agent-assist with suggested responses", "tool": "GPT-based", "result": "Issues resolved per hour +14% average, +34% for novice agents, ~0 for the most experienced; higher customer sentiment; lower attrition", "source_type": "Peer-reviewed field study", "confidence": "High", "caveats": "Firm anonymized by design; data 2020-2021. Formally published in the Quarterly Journal of Economics, vol. 140(2), pp. 889-942 (2025).", "url": "https://www.nber.org/papers/w31161", "last_verified": "2026-09-25"},
    {"id": 27, "company": "Klarna", "industry": "Fintech", "size": "~150M consumers", "function": "Customer-facing & revenue", "subprocess": "Customer support", "ai": "AI assistant for customer chat", "tool": "OpenAI", "result": "2.3M conversations in month one, two-thirds of chats; resolution 11 → 2 minutes; repeat inquiries down 25%; 'equivalent to 700 agents'; $40M projected 2024 profit improvement", "source_type": "Company press release / vendor case", "confidence": "Medium", "caveats": "May 2025: CEO said Klarna 'cut too aggressively' and began rehiring human agents for quality; AI still handles ~two-thirds of tickets in 2026. As of mid-2026 the company describes a stable hybrid model (AI for routine chats, human specialists for complex/premium cases) rather than a further reversal; separate Nov 2025 press reporting cites a larger, distinct productivity claim — AI work 'equivalent to 853 employees' (up from the original 700) with ~$60M projected annual savings — track as an evolving figure, not a confirmation of the original estimate.", "url": "https://www.klarna.com/international/press/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month/", "last_verified": "2026-09-25"},
    {"id": 28, "company": "Octopus Energy / Kraken", "industry": "Utility", "size": "Multi-£B", "function": "Customer-facing & revenue", "subprocess": "Customer support", "ai": "Gen-AI drafting of customer emails; 2026 autonomous agent trial", "tool": "Kraken (in-house)", "result": "~35% of customer emails AI-assisted; CSAT ~70% vs lower for unassisted; 6.2M calls summarized; 2026 trial: autonomous agent handled ~8,000 emails/week at 76% CSAT vs 72% for comparable human replies", "source_type": "Company statements over 3 years", "confidence": "Medium-high", "caveats": "'Work of 250 people' is a capacity claim; no layoffs stated", "url": "https://www.techuk.org/resource/case-study-kraken-tech-s-generative-ai-tool-for-customer-service.html", "last_verified": "2026-09-25"},
    {"id": 29, "company": "Vodafone", "industry": "Telecom", "size": "~€37B", "function": "Customer-facing & revenue", "subprocess": "Customer support", "ai": "TOBi / SuperTOBi chatbot", "tool": "Azure OpenAI", "result": "~1M interactions/month, 70% first-time resolution (UK); Portugal appointment booking first-time resolution 15% → 60%; online NPS +14 points", "source_type": "Company statement", "confidence": "Medium", "caveats": "Single-market pilot for the 15→60% figure", "url": "https://www.vodafone.com/news/newsroom/technology/meet-super-tobi-vodafone-s-new-generative-ai-virtual-assistant-now-serving-customers-in-multiple-countries", "last_verified": "2026-09-18"},
    {"id": 30, "company": "Commonwealth Bank of Australia", "industry": "Banking", "size": "A$27B+", "function": "Customer-facing & revenue", "subprocess": "Fraud prevention & customer service", "ai": "AI scam alerts, in-app messaging, annual credit reviews", "tool": "In-house", "result": "Customer scam losses down 50% (Nov 2024) and 76% by Aug 2025; call-centre wait times down 40%; credit reviews ~14 → 2 hours", "source_type": "Company statement", "confidence": "Medium", "caveats": "Multi-control attribution. Aug 2025: reversed 45 AI-justified call-centre cuts after call volumes rose. Jul 2026: announced 276 further redundancies (tech/ops/HR), part of ~800 cuts for the year per the Finance Sector Union, which lodged a Fair Work Commission dispute alleging 'fake redundancies' with roles re-advertised in India; CBA denies this, attributing cuts to automation and restructuring. A separate 1H FY2026 results disclosure cites a >20% fraud-loss reduction, a different framing/period than the 50-76% range above.", "url": "https://www.commbank.com.au/articles/newsroom/2024/11/reimagining-banking-nov24.html", "last_verified": "2026-09-25"},
    {"id": 31, "company": "Salesforce", "industry": "Software", "size": "~$38B", "function": "Customer-facing & revenue", "subprocess": "Customer support", "ai": "Agentforce on own help site", "tool": "Agentforce", "result": "2M conversations since Oct 2024; support headcount ~9,000 → ~5,000 with redeployment; 17% lower support cost", "source_type": "Vendor on its own product", "confidence": "Medium", "caveats": "Headcount figures include redeployment to sales. On 9 Jul 2026, KeyBanc downgraded Salesforce (to Sector Weight), citing a CIO survey finding limited evidence Agentforce is driving growth and no disclosed data showing Agentforce-linked revenue outgrowing overall ACV. Salesforce's Q2 FY2027 results (26 Aug 2026) separately reported Agentforce annualized revenue over $1.5B (+240% YoY) and 7.0B cumulative 'Agentic Work Units' delivered — a company-level KPI, not an independently audited customer outcome.", "url": "https://www.salesforce.com/blog/support-requests-agentforce/", "last_verified": "2026-09-25"},
    {"id": 32, "company": "Bank of America", "industry": "Banking", "size": "~$100B", "function": "Customer-facing & revenue", "subprocess": "Client assistant & internal helpdesk", "ai": "Erica client assistant; Erica for Employees", "tool": "In-house", "result": "2.5B+ client interactions; IT service-desk calls down 50%+; developer efficiency +20%; 23M advisor-assistant interactions in 2024", "source_type": "Company statement", "confidence": "Medium", "caveats": "No methodology", "url": "https://newsroom.bankofamerica.com/content/newsroom/press-releases/2025/04/ai-adoption-by-bofa-s-global-workforce-improves-productivity--cl.html", "last_verified": "2026-09-18"},
    {"id": 33, "company": "DoorDash", "industry": "Delivery platform", "size": "~$10B", "function": "Customer-facing & revenue", "subprocess": "Voice support", "ai": "Gen-AI voice agent for Dasher support", "tool": "Amazon Bedrock", "result": "Agent transfers down 49% vs prior IVR; first-contact resolution +12%; $3M/yr operating savings; built in ~2 months", "source_type": "Vendor case study", "confidence": "Medium", "caveats": "", "url": "https://aws.amazon.com/solutions/case-studies/doordash-bedrock-case-study", "last_verified": "2026-09-18"},
    {"id": 34, "company": "Ally Financial", "industry": "Banking", "size": "~$8B", "function": "Customer-facing & revenue", "subprocess": "Contact-centre after-call work", "ai": "Post-call summarization for 700+ associates", "tool": "Azure OpenAI", "result": "After-call work down 30% (target 50%); accuracy above 85%; 8-week deployment; later rolled out enterprise-wide", "source_type": "Vendor case study", "confidence": "Medium", "caveats": "", "url": "https://www.microsoft.com/en/customers/story/1715820133841482699-ally-azure-banking-en-united-states", "last_verified": "2026-09-18"},
    {"id": 35, "company": "Telstra", "industry": "Telecom", "size": "A$23B", "function": "Customer-facing & revenue", "subprocess": "Contact-centre agent assist", "ai": "One Sentence Summary and Ask Telstra", "tool": "Azure OpenAI", "result": "20% less follow-up contact; 90% of ~100 pilot agents reported time savings", "source_type": "Vendor case study", "confidence": "Low-medium", "caveats": "Survey-based pilot", "url": "https://www.microsoft.com/en/customers/story/1740058425924206437-telstra-telecommunications-azure-openai-service", "last_verified": "2026-09-18"},
    {"id": 36, "company": "Lumen Technologies", "industry": "Telecom", "size": "~$13B", "function": "Customer-facing & revenue", "subprocess": "Sales research", "ai": "Seller account research", "tool": "Microsoft 365 Copilot", "result": "Research per account 4 hours → 15 minutes; '$50M revenue over 12 months' is a projection from time saved", "source_type": "Vendor case study", "confidence": "Low-medium", "caveats": "Revenue figure is projected, not measured", "url": "https://www.microsoft.com/en/customers/story/1771760434465986810-lumen-microsoft-copilot-telecommunications-en-united-states", "last_verified": "2026-09-18"},
    {"id": 37, "company": "Amazon (Rufus)", "industry": "Retail", "size": "$600B+", "function": "Customer-facing & revenue", "subprocess": "Shopping assistant / sales lift", "ai": "Gen-AI shopping assistant", "tool": "In-house", "result": "250M users in 2025; 60% higher purchase completion among users (7-day attribution); $10B annualized incremental sales expected", "source_type": "Earnings call", "confidence": "Medium", "caveats": "Company attribution model", "url": "https://www.fortune.com/2025/11/02/amazon-rufus-ai-shopping-assistant-chatbot-10-billion-sales-monetization", "last_verified": "2026-09-18"},
    {"id": 38, "company": "Wendy's (FreshAI)", "industry": "QSR", "size": "~$2.2B", "function": "Customer-facing & revenue", "subprocess": "Drive-thru ordering", "ai": "Drive-thru voice ordering", "tool": "Google Cloud", "result": "100+ locations, tens of thousands of orders/day, 'increased the average check'; no accuracy or speed figure disclosed", "source_type": "Company statement", "confidence": "Low-medium", "caveats": "Adoption metrics only", "url": "https://www.wendys.com/blog/wendysr-square-deal-blog/transforming-ordering-experience-wendys-freshai-update", "last_verified": "2026-09-18"},
    {"id": 39, "company": "Amazon (Shreveport fulfillment center)", "industry": "Retail / logistics", "size": "$600B+", "function": "Operations & supply chain", "subprocess": "Fulfillment", "ai": "Next-generation FC with 10x robotics and AI orchestration", "tool": "In-house", "result": "Fulfillment processing time reduced up to 25%; targeting 25% lower cost to serve at peak; 30% more reliability and engineering roles", "source_type": "Company statement", "confidence": "Medium", "caveats": "'Aiming' language", "url": "https://www.aboutamazon.com/news/operations/amazon-fulfillment-center-robotics-ai", "last_verified": "2026-09-18"},
    {"id": 40, "company": "Google / DeepMind data centers", "industry": "Technology", "size": "—", "function": "Operations & supply chain", "subprocess": "Energy / cooling cost", "ai": "ML control of cooling systems", "tool": "DeepMind", "result": "40% reduction in cooling energy in 2016 tests (15% PUE overhead reduction); ~30% average energy savings under fully autonomous control from 2017", "source_type": "Company statement", "confidence": "Medium-high", "caveats": "Pre-2024; still the reference case for utility-cost AI. 2026 press coverage continues to cite consistent 30-40% cooling-energy savings with no reported reversal.", "url": "https://deepmind.google/blog/safety-first-ai-for-autonomous-data-centre-cooling-and-industrial-control/", "last_verified": "2026-09-25"},
    {"id": 41, "company": "IBM (AskHR / AskIT)", "industry": "IT services", "size": "$62B", "function": "People & IT", "subprocess": "HR and IT service delivery", "ai": "HR and IT service delivery agents", "tool": "watsonx Orchestrate", "result": "94% containment of common HR questions; 80 HR tasks automated; 11.5M interactions in 2024; HR operating cost down 40% over four years; support tickets down 75% since 2016; IT service interactions down 70%", "source_type": "Company self-case study", "confidence": "Medium", "caveats": "CEO says total headcount rose despite 'hundreds' of HR roles replaced", "url": "https://www.ibm.com/case-studies/ibm-askhr", "last_verified": "2026-09-25"},
    {"id": 42, "company": "Chipotle", "industry": "Restaurants", "size": "~$11B", "function": "People & IT", "subprocess": "Recruiting", "ai": "Conversational recruiting assistant 'Ava Cado'", "tool": "Paradox on Workday", "result": "Application to first day 12 → 4 days; application completion 50% → 85%; 30% of interview scheduling now after hours", "source_type": "COO quoted; vendor case consistent", "confidence": "Medium-high", "caveats": "", "url": "https://fortune.com/2026/06/02/chipotle-hiring-ai-assistant-ava-cado-from-12-to-4-days/", "last_verified": "2026-09-25"},
    {"id": 43, "company": "ServiceNow (Now on Now)", "industry": "Software", "size": "~$11B", "function": "People & IT", "subprocess": "IT/HR/customer agents", "ai": "IT/HR/customer agents and developer tools", "tool": "Now Assist", "result": "First 120 days: $10M forecast benefit ($5M+ cost takeout, $4M+ productivity), ~50 FTE of annualized productivity; resolution notes 80% faster; self-service deflection +14%", "source_type": "Vendor on own product", "confidence": "Low-medium", "caveats": "'Forecast' benefit", "url": "https://www.servicenow.com/customers/now-on-now-now-assist.html", "last_verified": "2026-09-18"},
    {"id": 44, "company": "Amazon (Q Developer)", "industry": "Retail / cloud", "size": "$600B+", "function": "People & IT", "subprocess": "Software maintenance", "ai": "Automated Java 8/11 → 17 upgrades across tens of thousands of apps", "tool": "Amazon Q Developer", "result": "~4,500 developer-years saved; $260M/yr savings from performance and security gains; per-app migration from a day+ to minutes", "source_type": "Company statement", "confidence": "Medium", "caveats": "Developer-years is a counterfactual estimate", "url": "https://aws.amazon.com/blogs/devops/amazon-q-developer-just-reached-a-260-million-dollar-milestone", "last_verified": "2026-09-18"},
    {"id": 45, "company": "Airbnb", "industry": "Travel platform", "size": "~$11B", "function": "People & IT", "subprocess": "Software maintenance", "ai": "LLM pipeline migrating ~3,500 test files", "tool": "In-house", "result": "Estimated 1.5 engineer-years → 6 weeks; 75% migrated in 4 hours, 97% after retries", "source_type": "Engineering blog", "confidence": "High", "caveats": "Single project; 1.5 years is an internal estimate. Distinct from a separate Airbnb customer-support AI initiative (not this case) that 2026 press reports as resolving a rising share of US/Canada support inquiries — not merged here pending primary-source verification.", "url": "https://airbnb.tech/infrastructure/accelerating-large-scale-test-migration-with-llms/", "last_verified": "2026-09-25"},
    {"id": 46, "company": "Cui et al. (Microsoft, Accenture, Fortune 100 firm)", "industry": "Software engineering", "size": "~4,800 developers", "function": "People & IT", "subprocess": "Software engineering", "ai": "GitHub Copilot, staggered rollout", "tool": "GitHub Copilot", "result": "Completed tasks/PRs +26% average; juniors +27-39%, seniors +8-13%; ~60% adoption after a year; quality not measured", "source_type": "Peer-reviewed RCTs (Management Science 2025)", "confidence": "High", "caveats": "Vendor's parent is a study site", "url": "https://mitsloan.mit.edu/ideas-made-to-matter/how-generative-ai-affects-highly-skilled-workers", "last_verified": "2026-09-25"},
    {"id": 47, "company": "Google internal RCT", "industry": "Software engineering", "size": "96 engineers", "function": "People & IT", "subprocess": "Software engineering", "ai": "Internal AI coding assistance", "tool": "In-house", "result": "Task time 114 → 96 minutes (~21% faster); marginal significance", "source_type": "Academic", "confidence": "Medium-high", "caveats": "Single lab task", "url": "https://arxiv.org/abs/2410.12944", "last_verified": "2026-09-25"},
    {"id": 48, "company": "METR RCT", "industry": "Open-source software", "size": "16 expert maintainers, 246 issues", "function": "People & IT", "subprocess": "Software engineering", "ai": "Cursor with Claude 3.5/3.7", "tool": "Cursor", "result": "Developers took 19% longer with AI while believing they were 20% faster", "source_type": "Academic RCT (Jul 2025)", "confidence": "High", "caveats": "Low generalizability; early-2025 tools. In a Feb 2026 post, METR disclosed that its follow-up developer-productivity study suffered from selection bias (many developers refused to join the no-AI control arm), which it says likely biases the AI-speedup estimate downward, and announced a redesign — a reason for caution in generalizing the 19%-slower finding to later 2026 AI-adoption conditions.", "url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/", "last_verified": "2026-09-25"},
    {"id": 49, "company": "Woven by Toyota", "industry": "Automotive software", "size": "Toyota subsidiary", "function": "People & IT", "subprocess": "Software compliance", "ai": "Automated MISRA compliance fixes", "tool": "Azure OpenAI o1", "result": "80% of corrections automated (50% with GPT-4o); 60,000 errors fixed in one module", "source_type": "Vendor case study", "confidence": "Low-medium", "caveats": "POC stage", "url": "https://www.microsoft.com/en/customers/story/24108-woven-by-toyota-inc-azure-openai", "last_verified": "2026-09-18"},
    {"id": 50, "company": "BCG / Harvard (Dell'Acqua et al.)", "industry": "Consulting", "size": "758 consultants", "function": "People & IT", "subprocess": "Knowledge work", "ai": "GPT-4 on consulting tasks", "tool": "GPT-4", "result": "+12.2% tasks, 25.1% faster, 40%+ higher quality inside the model's competence; 19 points less likely to be correct outside it; bottom-half performers +43%, top +17%", "source_type": "Pre-registered field experiment", "confidence": "High", "caveats": "GPT-4 era. Now formally published in Organization Science (Mar 2026). A separate BCG study ('When Using AI Leads to Brain Fry,' 5 Mar 2026, n=1,488 US knowledge workers) found a countervailing effect: 14% report AI-driven cognitive overload, correlated with more errors and higher quit intention — a caution against over-generalizing productivity gains.", "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321", "last_verified": "2026-09-25"},
    {"id": 51, "company": "UK Government Copilot trial", "industry": "Government", "size": "20,000 civil servants, 3 months", "function": "People & IT", "subprocess": "Knowledge work", "ai": "Microsoft 365 Copilot", "tool": "Microsoft 365 Copilot", "result": "26 minutes/day average saved (~2 weeks/yr); 17% saved no time; later DWP evaluation found 19 minutes/day", "source_type": "Government evaluation", "confidence": "Medium", "caveats": "Self-reported time", "url": "https://www.geekwire.com/2025/microsoft-ai-tools-saved-british-government-workers-26-minutes-a-day-new-study-shows/", "last_verified": "2026-09-25"},
    {"id": 52, "company": "Lloyds Banking Group", "industry": "Banking", "size": "~30,000 licenses", "function": "People & IT", "subprocess": "Knowledge work", "ai": "Microsoft 365 Copilot", "tool": "Microsoft 365 Copilot", "result": "46 minutes/day saved (survey of 1,000 users)", "source_type": "Company claim", "confidence": "Low-medium", "caveats": "Survey-based. Separately, Lloyds reports ~£50M in P&L benefit from ~50 GenAI use cases in 2025, targeting >£100M in 2026 (a different metric than the per-user time-saved figure above).", "url": "https://www.theregister.com/2025/10/20/lloyds_banking_copilot/", "last_verified": "2026-09-25"}
  ]
}