If Issue 003 belonged to Anthropic, this fortnight belongs to the ERPs. In barely two weeks, SAP (200+ agents and an Autonomous Close Assistant going GA in Q2), Xero (XeroForce, a plain-language agent builder), Sage Intacct (a Finance Intelligence Agent and an open AI Gateway), NetSuite (Autonomous Close agents) and OneStream (a governed agentic finance layer) all shipped tools that promise to run the close itself — not assist with it. The common thread underneath them is Anthropic: Claude is now the embedded reasoning engine inside SAP, Xero and Intuit, and live in production at FIS and Goldman. The demand signal is real — KPMG's survey of 1,013 finance leaders found active AI use has more than doubled to 75% in a year — but so is the governance gap. Shadow-AI compliance deadlines are now live, the EU AI Act's high-risk enforcement date is ten weeks away, and the accountancy bodies (CCAB) are formalising that your professional obligations now extend to the AI tools your team uses. For charity finance teams the gap is starkest of all: 88% of charities now use AI day-to-day, yet just 3% of trustees know it.
"Active AI use in finance has more than doubled — from 30% to 75% — in twelve months. Assurance-ready organisations report error reduction 3–6× higher than their peers."
SAP's 12–14 May keynote introduced more than 200 specialised agents across finance, procurement, HR and supply chain. The headline for finance teams is the Autonomous Close Assistant — GA in Q2 2026 — which automates journal entries, reconciliations and error resolution to compress the close "from weeks to days," with Anthropic's Claude confirmed as the primary embedded reasoning engine.
Announced 13 May, XeroForce lets accountants and finance teams create custom AI agents over Xero and connected apps using natural-language prompts — no code — built on Xero's JAX superagent architecture with a live Claude integration. For smaller finance teams, it lowers the barrier to building bespoke month-end workflows from "engage a developer" to "describe what you want."
Released 8 May, Sage Intacct's R2 introduces a Finance Intelligence Agent for plain-language data querying with visible reasoning, and — more strategically — an AI Gateway that opens Intacct to external agents including Microsoft Copilot and Claude through standard APIs. It signals that the major mid-market ERPs now expect to be queried and actioned by AI tools they don't own.
Microsoft's Dynamics 365 Business Central Expense Agent (May 2026 Wave 1) captures, validates and submits employee expenses without manual intervention — part of a broader Wave 1 shift from assistive Copilot to autonomous agents that also covers automated cash application and overdue-invoice follow-up. For teams already on Business Central, these arrive as platform features, not separate purchases.
On 19 May, OneStream made its agentic AI for corporate finance generally available, including a Finance Agentic Layer that lets its governed data and agents interoperate with external AI tools — Claude, ChatGPT and Microsoft Copilot. The design point matters: it keeps governance and the single source of truth inside OneStream while letting finance teams act on it from whichever assistant they already use.
Announced 13 May, Vena's MCP Server lets finance teams trigger governed planning workflows directly from AI tools, alongside a new Excel-native Financial Consolidation handling intercompany eliminations, FX translation and ownership structures inside the close. It is one of the first FP&A platforms to expose a production MCP server rather than a roadmap promise.
NetSuite is broadening its AI Connector Service and MCP support so customers can query NetSuite data through assistants such as Claude and ChatGPT in a controlled, permissioned way. Combined with its Autonomous Close agents, it points to a model where the ledger is both run by background agents and queryable in natural language by the finance team.
Three more production-ready MCP servers: Zoho Books (read/write accounting with admin-controlled permissions), Norman Finance (UK invoicing, VAT and transaction categorisation) and LSEG (ten market-data tools spanning bonds, FX, yield curves and equities). The UK angle is notable — Norman Finance brings VAT-aware bookkeeping into Claude for British small organisations.
Copilot's 2026 Wave 1 ships a Payflow Agent that autonomously executes vendor payment runs inside Dynamics 365, and a Financial Reconciliation Agent in Excel that detects structural mismatches, drafts reconciliation reports and suggests fixes without leaving the spreadsheet. A new Plan Mode previews every workbook change step-by-step before it is applied — a useful control for audit-conscious teams.
VicInbox, now on the Outlook Add-in marketplace, reads invoices, payment queries, duplicates and statements straight from the inbox, routes them into Vic.ai and auto-drafts vendor responses. It is a concrete example of agents moving upstream — handling the messy, email-driven front end of accounts payable that has resisted automation.
OpenAI made GPT-5.5 Instant the new default model on 5 May, citing a 52.5% reduction in hallucinated claims on high-stakes finance, law and medicine prompts. For finance teams using ChatGPT for analysis or drafting, accuracy on numbers and rules is the differentiator that matters most — and the gap is closing.
KPMG's month-end AI assistant follows the finance team's own close checklist, automates repetitive posting tasks, interprets data and flags discrepancies — all inside Workday. Crucially it works to the team's existing sequence rather than imposing a vendor-defined one, which is the difference between an automation a team adopts and one it quietly abandons.
Gartner projects organisations on AI-enabled cloud ERPs will close 30% faster by 2028, with AI-enabled ERP spending rising from 14% of the market in 2024 to 62% by 2027. The implication for any finance team weighing a multi-year system contract: staying on a platform without embedded AI becomes a structural close-speed disadvantage within two reporting cycles.
Nearly 80% of finance professionals say month-end delays come from waiting for data from other systems — not a lack of AI tooling (CFO Dive). The practical lesson before buying anything: automating journals and reconciliations on top of a fragmented data environment just accelerates the wrong workflows. Sort the upstream pipeline first.
KPMG's survey of 1,013 senior finance leaders found active AI use jumped from 30% to 75% in a year. The differentiator is assurance: organisations with mature AI governance report error-reduction rates 3–6× higher than peers (33% vs 6%) and are three times more confident about scaling. Adoption is no longer the hard part — trustworthy adoption is.
Active agents in Microsoft 365 grew 18-fold among large enterprises year-on-year, yet tying that activity to margin, cost reduction or payback remains unsolved for most finance functions. The gap between productivity signals and provable financial return is now the central CFO question — and a reason to define success metrics before you deploy.
After the number of AI systems needing oversight grew 60% year-on-year, Mastercard's CFO-backed framework embedded scorecards into product development and turned its audit trail into a revenue-retention tool when bank clients demanded assurance on model fairness. A useful template for finance leaders who need governance that scales without a new team.
Charity Excellence research finds fewer than one in four UK charities have approved AI tools, written policies and staff training in place — yet 88% report AI is used day-to-day, and only 3% of trustees are aware their charity uses AI at all. For charity Finance Directors this is now a board-accountability question: trustees carry the legal risk for a tool they don't know exists.
In partnership with UK Community Foundations and CAST, the £3m programme (14 May) funds civil society to shape how AI is used in their communities, including a UK-wide "AI pulse network" of 50 community organisations. It is a rare instance of sector infrastructure being built to help charities adopt AI on their own terms rather than a vendor's.
EY's AI Sentiment Index (1,000 UK respondents) finds widespread use but narrow trust: just 14% are comfortable with fully autonomous, agent-led systems and only 41% trust government to manage AI data responsibly. For charities, it underlines why transparent human-oversight frameworks matter to donors and beneficiaries, not just regulators.
A Fortune investigation (BambooHR data, 12 May) finds senior finance staff increasingly use AI to do work once handed to juniors, producing a roughly 3:1 senior-to-entry-level hiring ratio — while one in three new accounting and finance hires leave within their first year. The talent pipeline question is becoming as urgent as the technology one.
ICAEW's April guidance names the priority capabilities — Python, SQL, and the ability to critically question AI outputs — and warns that the deficit itself is now a risk. Early-career professionals are already spending less time on data entry and more reviewing AI-generated work, which changes what "training a junior" means.
ICAEW analysis points to three emerging job titles: finance prompt engineer, finance storyteller, and ESG Finance Director — alongside up to eight hours of free GenAI training now available to members through a Microsoft partnership. The signal for finance leaders: build AI literacy into development plans now, before these roles become hires you're competing for.
The Consultative Committee of Accountancy Bodies (ICAEW, ACCA, ICAS, AAT and others) is preparing a formal Statement to the Profession covering hallucination risk, data privacy and bias. The principle is significant: relying on AI outputs without understanding their basis may breach the fundamental principles of objectivity and professional competence — your ethical duties now follow the tool.
Following the Treasury Select Committee's findings, the Bank of England confirmed plans to stress-test AI agents in trading markets and the FCA committed to publishing practice examples for aligning AI with conduct rules. Major AI and cloud providers may be designated Critical Third Parties by the end of 2026 — pulling the AI supply chain itself into the regulatory perimeter.
The PRA's Sam Woods told a UK Finance event (12 May) that frontier models pose a material operational-resilience risk — specifically through their ability to find system vulnerabilities faster than firms can patch them. A reminder that the governance conversation is no longer only about what AI gets wrong, but about the pace it sets for everything around it.
| Finding | Source | Date |
|---|---|---|
| Active AI use in finance more than doubled — from 30% to 75% in a year; 71% say AI meets or exceeds ROI expectations | KPMG Global AI in Finance Report | May 2026 |
| 74% of AI's economic value captured by just 20% of organisations; leaders see 25–40% productivity gains vs 3–7% for the median firm | PwC 2026 AI Performance Study | Apr 2026 |
| CFOs deploying strategic AI will add 10 margin points by 2029; AI-enabled cloud ERP delivers a 30% faster close by 2028 | Gartner | Apr 2026 |
| Active AI agents in Microsoft 365 grew 18-fold year-on-year; only 26% of AI users say leadership is clearly aligned on strategy | Microsoft Research (via Fortune) | May 2026 |
| 81% of financial services firms are adopting AI, but over 50% remain pre-production | Cambridge CCAF / IMF / BIS | Apr 2026 |
| 88% of UK charities use AI day-to-day, yet only 3% of trustees are aware of it; fewer than 1 in 4 have a policy | Charity Excellence (via Civil Society) | May 2026 |
A survey of 1,800 global finance leaders finds AI still hasn't liberated teams from transactional work — and that CFOs are increasingly hiring for creativity and empathy alongside data skills. The tension worth sitting with: the leaders who believe most in AI are also the most worried about employee resistance.
Hg Capital, a major private-equity backer of accounting software, has marked down almost two-thirds of its 20 largest portfolio companies as AI disruption fears intensify. The signal for any finance team mid-way through evaluating a multi-year ERP contract: the market no longer treats incumbent software as protected territory.
Lloyds Banking Group became the first FTSE 100 firm to deploy AI in the boardroom — a "board bot" that helps directors review confidential papers, prepare for meetings and flag decision-making bias, with the bank estimating £100m in AI-driven value in 2026. It sets a reference case chairs and trustees elsewhere will increasingly be asked to match.
Controllers and FP&A leads are building working dashboards with LLMs, often the first time AI has felt genuinely usable to them. The corrective is the useful part: the gap between a prototype that impresses in a demo and a system that withstands audit, controls and integration is exactly where most finance AI projects stall.