Agentic AI for UK finance and operations in 2026.

Most UK finance and operations directors are still reading about AI in the abstract. The work has moved on. Nine specific workflows in finance and ops are now running with agents in production inside mid-market businesses. Most of the rest is still theory. The gap between what is deployed and what is discussed is wider than it has ever been.

Where agentic AI is actually working in finance.

Five finance workflows. Month-end reconciliations, accounts payable, group consolidation for multi-entity businesses, management reporting, and cash forecasting. These are the deployments that show up in Deloitte's State of Generative AI in the Enterprise 2026, McKinsey's State of AI 2026, and Stanford Digital Economy Lab's Enterprise AI Playbook when researchers go looking for production agentic workloads. They share a profile: defined inputs and outputs, controls expressible in rules, audit trails the FD already needs for other reasons, and enough monthly volume that the build pays back inside one year.

The pattern across all five is the same. The agent does the mechanical run, a person handles exceptions and signs off. Reconciliations clear at 70%+ on the first pass and a worklist of fuzzy matches gets surfaced for batch approval. AP runs invoice extraction, GL coding against historical patterns, and routing under existing delegation rules. Consolidation rebuilds the group P&L on demand instead of monthly. Reporting packs assemble overnight with variance commentary written against rules the finance director defined. Cash forecasting pulls daily off the ledgers and ages thirteen weeks. None of these are speculative. McKinsey reports finance as one of the three functions most likely to have at least one scaled agentic workflow in 2026. Payback by workflow type sits between two and fourteen months.

Where agentic AI is working in operations.

Four operational workflows. Customer service triage, supplier onboarding, returns reconciliation, and stock allocation. Deloitte's 2026 enterprise research puts customer operations in the top two functions for production agentic deployment alongside IT. The rest follow the same pattern: high-volume work, well-defined decision trees, bounded cost of error, and enough repetition that consistent execution is worth more than occasional brilliance.

Tier-one support agents read the ticket, pull the customer record, resolve the simple cases against a defined playbook and escalate the rest with the context pre-loaded. Supplier onboarding runs KYC against company registers, document validation, ledger setup, and the email chain that used to take a finance assistant three days. Returns reconciliation does the cross-system match across the CS platform, the ecommerce store, the warehouse, and the payment processor. Stock allocation runs daily across sales channels with conflicts routed to a person to resolve. The map below shows all nine alongside what is emerging and what is not yet reliable.

What "working" actually means in 2026.

The workflows above sit inside the 23% that have scaled. The 65-point gap to the 88% using AI in any form is where most UK mid-market businesses currently live. A pilot in slide 14 is not a production workflow.

Working means in production with audit trails. Not a pilot. Not a copilot. Not ChatGPT in a browser tab.

"Working" here means something specific. Named ownership, full audit logging, exception routing to a human, monitored failure rates, and an output the business actually uses. Anything short of that is preparation, not delivery.

What is not yet working reliably.

Strategic FP&A, scenario modelling for capital decisions, M&A screening with commercial judgement, contract negotiation, and anything where the right answer depends on a political read of stakeholders. These are still firmly people work in 2026. The adjacent emerging cases (anomaly detection, forecasting support, compliance monitoring) are not yet at the reliability bar where the workflow can run unattended.

UK-specific considerations.

The ICO's Tech Futures: Agentic AI paper (January 2026) is the regulatory anchor for UK finance teams. It confirms agentic AI is workable under UK GDPR, with attention to lawful basis, transparency, and the Article 22 right to human review for any decision producing legal or significant effect on individuals. Three practical implications: customer-facing automated decisions need a documented human-in-the-loop pathway, the full chain of an agent's actions has to be logged so an auditor can reconstruct it, and each deployment needs a named accountability owner. The production cases cited by McKinsey, Deloitte, and Stanford include this control layer as standard. In UK regulated finance the governance work is the gating item, not the agent.

Frequently asked questions