The biggest story this week is the ERP vendors' simultaneous move to agentic — Xero's full platform repositioning as "Xero OS," NetSuite's Intelligent Close Manager, Business Central's autonomous AP workflows, and SAP Joule's plain-language journal entry drafting all shipped or were announced in the same seven days, joining Microsoft Copilot's agentic Excel launch to signal that AI-assisted close is a present-tense product decision, not a roadmap item. Alongside this, the MCP ecosystem for finance has reached a genuine inflection point: Digits launched a live accounting MCP server, Xero and Intuit are both building towards real-time MCP connections via Anthropic partnerships, and Goldman Sachs is running Claude agents on trade accounting and compliance. Set against this technology acceleration, the harder story is the adoption gap — Deloitte finds only 21% of organisations with deployed AI report tangible value, PwC finds the top 20% of performers capture 74% of all AI-driven gains, and Gartner confirms adoption has plateaued for a second year running. For finance teams, the question is no longer whether to use AI but whether you are in the minority extracting the returns, or the majority that isn't.
"The top 20% of AI performers capture 74% of all AI-driven value — and achieve a 7.2x performance boost over their peers."
Copilot in Excel has moved from passive assistant to active agent, taking multi-step actions — building formulas, creating pivot tables, generating charts, and editing data — without the user leaving the sheet. Available from 22 April to all Microsoft 365 Copilot and Premium subscribers; finance teams running complex models now have an agent that does the work rather than just suggests it.
Eight prebuilt AI agents handle the full lifecycle of vendor emails — payment status queries, bank detail change verification, duplicate detection, W-9 routing, vendor statement reconciliation, and remittance requests — end-to-end and with a visible audit trail. AP teams spend up to a week per month on vendor correspondence; the agents deploy in under 30 minutes with no IT involvement.
OpenAI acquired Hiro Finance on 13 April, a startup with math-heavy models built for balance sheet management, tax optimisation, and multi-step financial planning. The signal: OpenAI is building specialist financial reasoning directly into ChatGPT, pushing the general-purpose tools closer to dedicated finance AI.
From 16 April, Claude Enterprise seat fees no longer include any usage allowance — all token consumption is billed separately at API rates. Teams that previously stayed within their allowance face meaningful cost increases; worth factoring into any AI tool procurement decision now.
Announced 22 April, Xero OS repositions the platform from a ledger of record to an autonomous finance stack, with JAX (Just Ask Xero) acting as an always-on AI CFO: analysing incoming bills, setting up payments, and auto-reconciling transactions end-to-end. For finance teams on Xero, bank rec and AP matching can now happen continuously rather than as a closing sprint — the platform architecture has changed, not just the feature set.
NetSuite's 2026 Release 1 introduces an AI-driven close dashboard that monitors AR, AP, and accounting exceptions in real time, alongside AI agents for account reconciliation and generative AI bank transaction matching. Finance teams get a live view of close progress with exception alerts, rather than chasing status updates manually.
The Spring 2026 Wave embeds agentic AI into Business Central's finance module: autonomous AP workflows handle invoice processing, vendor matching, and approvals, while AI bookkeeping agents suggest journal entries from historical patterns. AP accruals and invoice coding become largely self-managing — available to Business Central Online customers from April.
SAP's Q1 2026 AI release extends Joule to draft journal entries from natural language instructions, summarise period-end closing status, and translate complex e-invoicing errors into plain English. For S/4HANA Cloud teams, recurring accruals and standard month-end journals can now be initiated by describing what you want rather than navigating transaction codes.
Digits launched its MCP Server on 21 April, giving accounting firms and finance teams direct real-time access to Digits financial data inside Claude, ChatGPT, and Cursor. Free on all plans, no developer experience required, and auto-categorises 95%+ of transactions — the first production-ready accounting MCP connector available to finance teams today.
Xero and Anthropic announced a multi-year deal to bring Claude directly into Xero's workflow via MCP, making Xero financial data available as a real-time data source inside Claude.ai. Live dates not yet confirmed, but when it ships, bank feeds, AP, and reconciliation data will be accessible to agentic workflows without permanent data ingestion.
Intuit and Anthropic signed a strategic partnership to surface QuickBooks, TurboTax, Credit Karma, and Mailchimp data inside Claude.ai via MCP, with spring 2026 rollout underway. QuickBooks and TurboTax are also now available as direct tool integrations inside Claude — finance teams can query Intuit data conversationally without switching applications.
Goldman Sachs is co-developing autonomous Claude agents with Anthropic engineers for transaction reconciliation, trade accounting, and client onboarding. In testing ahead of live launch; Goldman's CIO noted the team was "surprised" at Claude's capability outside of coding — and described it as combining "large data volumes with rules and judgment."
KPMG's "Ignite Financial Close Companion" automates routine close tasks — journal entry execution, reconciliation, anomaly flagging — via natural language commands within Workday, powered by Gemini Enterprise. For Finance Directors on Workday, this is a deployed commercial product, not a prototype, and shortens the reporting cycle while reducing human error risk.
Forrester identifies the AP automation priorities: invoice capture, exception handling, three-way matching, supplier communications, on-demand reporting, and continuous fraud detection — shifting from rule-based bots to context-aware agents that handle ambiguous cases. Finance teams can act now by mapping which of the six they still handle manually and targeting the highest-volume exception category first.
Trintech's CFO argues AI outputs must withstand auditor scrutiny before going live, and internal champions must be built rather than adoption mandated. Trintech's Beacon AI uses a human-in-the-loop model — AI surfaces the finding, a human approves before posting — the framework most likely to satisfy any audit committee.
Deloitte's State of AI 2026 report finds 87% of CFOs believe AI will be extremely important — but only 21% of organisations with fully deployed AI say it has delivered measurable value, and just 14% have fully integrated AI agents. The gap between "we use AI" and "AI is working" is the defining business problem for the sector right now.
Shadow AI accounted for one in five AI-related security incidents in financial services last year, with 97% of affected organisations lacking adequate AI access controls. Average breach cost reached $5.56 million; the entry point is typically analysts pasting GL data and forecasts into unsanctioned consumer AI tools — not a sophisticated external attack.
The Charity Excellence Future Charity Report (April 2026) finds fewer than 25% of charities have approved AI tools, policies, or training — even as two-thirds describe themselves as "experimenting." All three trustee-level governance controls are rated red. Finance Directors whose teams are already using AI tools are exposed without a documented governance framework to point to.
Charity Digital's April guidance recommends FDs start with low-risk automation and anomaly detection, insist on data anonymisation from vendors, and embed governance policies before scaling. 94% of finance professionals regretted rushing software decisions — a direct warning for FDs under pressure to "do something" with AI.
AICPA and CIMA have upgraded the CGMA Professional Qualification to embed generative AI across budgeting, forecasting, and reporting — examinable from May 2026 case study sittings. AI fluency is now a core, non-optional part of the management accounting qualification from this exam cycle.
More than half of UK accounting professionals now use AI regularly (36% weekly, 24% daily), while the share describing themselves as "fearful" has dropped from 20% to just 8% year-on-year. The challenge for 2026 is governance — accuracy and hallucinations remain the top concern, and most practices are still managing AI ad hoc.
Full obligations for high-risk AI systems apply from 2 August 2026. Organisations must complete conformity assessments, finalise technical documentation, and register systems in the EU database. A European Commission simplification proposal is in train but should not be treated as a delay — the deadline remains binding.
The ICO's Tech Futures: Agentic AI paper flags that controller/processor accountability across multi-vendor agentic chains is legally unresolved. Finance teams should document data flows, define responsibilities with vendors in writing, and ensure automated decisions allow for human intervention — before regulators formalise the requirements.
| Finding | Source | Date |
|---|---|---|
| 87% of CFOs expect AI to be extremely or very important to finance operations in 2026 | Deloitte Q4 2025 CFO Signals | Q4 2025 |
| Only 21% of organisations with fully deployed AI report tangible value; just 14% have fully integrated AI agents | Deloitte State of AI 2026 | 2026 |
| Finance AI adoption 59% in 2025 vs 58% in 2024; 60% of projects fail due to organisational readiness | Gartner CFO Survey (n=183) | Nov 2025 |
| Top 20% of AI performers capture 74% of value; 7.2x performance boost over peers | PwC 2026 AI Performance Study | Apr 2026 |
| Finance AI adoption 56% by 2026 (from 17% in 2023); 45% still in limited pilot; only 17% use AI in core workflows | CFO Connect State of AI in Finance 2026 | 2026 |
| A task taking 3 hours now takes 15 minutes in AP/AR automation — 12x faster | L.E.K. Consulting 2025 CFO Survey | 2025 |
MIT's McAfee argues accessible AI will increase the gap between leaders and laggards — success comes from organisational commitment and rapid diffusion, not from access to the same tools. The window for first-mover advantage is narrower than it looks.
71% of CIOs say AI budgets will be frozen or cut if value cannot be demonstrated within two years. One of the seven key factors: involving finance in certifying AI value — positioning the FD as the organisation's strategic proof-point generator.
"Vibe coding" — expressing intent in plain language and letting AI translate it into functional outputs — is moving into the finance stack. Platforms like Datarails give CFOs a natural-language interface to their own data without touching a formula.
Genpact argues autonomous finance has moved beyond proof-of-concept. The central tension: 96% of CFOs say human oversight remains critical, yet without genuine autonomy, agentic AI cannot deliver the step-change capacity that finance functions are betting on.