Real AI deployments in corporate treasury
36 sourced deployments across forecasting, cash positioning, agents and more. Most results are self-reported by vendors or award entries; check the source.
Weekly: what treasury AI actually shipped. One email, every Tuesday.
| Company | Industry | Use case | Vendor | Reported result | Date | Source |
|---|---|---|---|---|---|---|
| Amazon | E-commerce/Technology | ML cash forecasting (60-day global) and cash positioning | In-house build (with Kyriba case study) | Cash-positioning time cut from 8 hours to under 30 minutes | n/d | source |
| Groupement Les Mousquetaires | Retail | Predictive AI cash forecasting (12-week) | In-house models fed from Kyriba | 12-week forecast error cut from 3% to 1%; ~EUR 1M annual savings | Launched 2019 | source |
| Danone | Consumer goods | AI receivables and cash forecasting | HighRadius | 99.3% receivables forecast accuracy; 30% less manual forecasting effort | n/d | source |
| Bishop Lifting | Industrial | AI/ML cash forecasting and bank data integration | HighRadius | 95% overall forecast accuracy; 97% forecasting productivity boost | n/d | source |
| Pavion | Security/Services | AI-agent cash forecasting | HighRadius | Forecast time from 30 hrs/week to under 5 minutes; 94.2% accuracy; 17 AI agents | n/d | source |
| Microsoft | Technology | ML + GenAI (Copilot/Azure OpenAI) collections and cash application | In-house (Dynamics 365) with Accenture | Adam Smith Award 2025 winner (Project MICS) | 2025 | source |
| Siemens (Mexico) | Industrial | AI cash application matching (pilot) | Not disclosed | Base tool expected to automate 80%; AI may raise to 90% | 2025 | source |
| Hindustan Unilever | Consumer goods | AI-enabled working capital and cash forecasting | CashForce + Unilever TMS | End-to-end treasury automation; real-time visibility | 2023 | source |
| Siemens Healthineers | Healthcare | AI liquidity planning | Liquiture | Part of TREfF treasury transformation (Adam Smith 2024) | 2024 | source |
| Sealaska | Diversified holding | ML cash forecasting | Trovata | Forecast horizon from 1 month to 13 weeks-1 year; 10+ hrs/week saved | n/d | source |
| Taboola | Advertising technology | AI agents for cash position and 13-week forecast | Nilus | 95% of transactions auto-tagged; results within 3 months | n/d | source |
| 365 Retail Markets | Retail technology | AI cash flow forecasting | Nilus | Implemented in weeks; replaced Excel forecasting | n/d | source |
| Alloy | Fintech | AI transaction tagging and reconciliation | Nilus | 50+ hours saved per month | n/d | source |
| On | Consumer goods/Apparel | ML cash forecasting and classification | Palm | $350M+ additional cash investments from previously idle cash | n/d | source |
| StackAdapt | Advertising technology | AI agents for cash position and tagging | Nilus | 95%+ of transactions auto-tagged | n/d | source |
| WPP | Advertising | GenAI "Treasurer Bot" for policy validation and treasury Q&A | In-house | Adam Smith Awards 2025 Judges' Choice | 2025 | source |
| ASML | Semiconductors | ML forecast for FX purchase hedging | In-house (open-source) | Forecast accuracy increased from 70% to 96% | 2024 | source |
| Kulicke & Soffa | Semiconductors | ML cash forecasting and anomaly detection | J.P. Morgan Cash Flow Intelligence | Forecasting time decreased from 20 hours per week to approximately three hours (85% reduction) | 2025 | source |
| Coca-Cola | Beverages | Internal ChatGPT instance for treasury analysis tasks | OpenAI (internal instance) | Used e.g. to build CDS spread heat map tooling | 2023 | source |
| FinQuery | Software | AI transaction categorisation and 13-week forecast; agent workflow in beta | Agicap | 95%+ of bank transactions auto-categorised on first run | n/d | source |
| University Medical Partners | Healthcare | AI categorisation and rolling forecasts | Agicap | 85% reduction in categorisation time (2 hours/month to 10-15 minutes) | n/d | source |
| Super Group | Gaming/Entertainment | AI-powered treasury automation (liquidity/FX/payments) | Bond | Case study published on AFP marketplace | n/d | source |
| Ant Group | Fintech | AI co-pilot (Tianluo) and AI treasury agent (TGL-14) | In-house | Automation of 99% of payments and loans | 2025 | source |
| BearingPoint | Consulting | ML payment-collection forecasting | Ripple Treasury (GSmart Ledger) | Forecasts based on customer payment behaviour vs due dates | n/d | source |
| Prysmian | Industrial/Cables | AI/ML cash forecasting | J.P. Morgan Cash Flow Intelligence | Forecast horizon from 30 to 91 days; <1% error rate; saves $100,000 annually | n/d | source |
| Flex | Fintech | AI agents for cash position and runway forecasts | Atlar | 5x first-year ROI; $50M+ idle cash swept to interest-bearing accounts | n/d | source |
| Centalion Group | Commodity trading | LLM-based AI treasury agent for trade finance instrument issuance and amendments (facility capacity checks; human approval) | In-house agent on Komgo Konsole (MCP connector) | Issuance processing time cut by around 55% and amendment processing time by around 80% (first deployment: OBSI syndicated facility) | Oct 2026 | source |
| Airwallex | Fintech/Payments | AI-built liquidity insights tool for safeguarding monitoring; programmatic signal-based cash rebalancing | In-house | No metrics published; treasury director says the team now takes on more sophisticated work with the same team | Oct 2026 | source |
| Newell Brands | Consumer goods | AI summaries and translations of bank documentation (POAs, signatory powers) and account-opening requirements | Not disclosed | Qualitative: centralised team says it relies on AI summaries/translations to manage POAs and signatory powers across countries; not yet used for liquidity forecasting. Via Treasury Career Corner Podcast (Mike Richards) | Oct 2026 | source |
| AstraZeneca | Pharmaceuticals | ML receipt-timing prediction and short-term liquidity forecasting fed live from multiple SAP ECC systems | FIS Quantum + Uniun-Forecasting (Covarius) | Liquidity visibility horizon from ~1 week to up to 5 weeks; Adam Smith Awards 2026 Best Cash Management Solution and Overall Winner. Partner-reported (Covarius): entity coverage 3 to 247; receipt prediction accuracy up 60%+ | 2026 | source |
| Bandwidth | Communications technology | Self-built AI-native treasury platform: cash positioning, forecasting, transaction categorisation, netting, covenant tracking, bank fee analysis | In-house build on Atlar bank connectivity + Claude (via Atlar MCP connector) | AFP 2026 Pinnacle Awards finalist (winner to be announced at AFP 2026, 8-11 Nov); no metrics published | 2026 | source |
| Technology | ML daily cash forecasts plus multi-agent investment workflow (rate collection, investment recommendations, execution agent for trades and ERP entries) | In-house (piloted with HSBC in India and Taiwan) | AFP 2026 Pinnacle Awards finalist (winner to be announced at AFP 2026, 8-11 Nov); no metrics published | 2026 | source | |
| ServiceNow | Software | Central treasury data layer (10+ source systems), then LLM chatbot, ML AP cash forecasting, automated FX hedging (40+ currency pairs), AI counterparty exposure monitoring | In-house (with Tradeweb ICD Portal and Clearwater data) | Previously 40-60 hrs/week of manual data collection, much of it now replaced (no figure given); AFP 2026 Pinnacle Awards finalist (winner to be announced 8-11 Nov) | 2026 | source |
| HelloFresh | Consumer goods/Meal kits | Agentic AI to find hard-to-reconcile payments and answer TMS documentation questions | Kyriba (TAI) | Vendor-reported: hard-to-reconcile payments surfaced within seconds; documentation questions that took hours now answered via TAI | 2026 | source |
| Airswift | Workforce solutions | Agentic AI for reconciliation, cash optimisation and risk monitoring | Bond | Selected Bond (vendor announcement, 22 Oct 2026); no results published yet | 2026-10 | source |
| UNHCR | Humanitarian/Non-profit | AI cash forecasting: automated data collection and normalisation, forecast model and dashboard; run in parallel with manual forecast before adoption | In-house with UNICC | Estimated 40-60% less data collection and consolidation work once the model is fully stabilised (estimate, EuroFinance 2026) | 2026 | source |