AI in Finance Function: Notes from the Room 

On 13th May, FD Recruit hosted a virtual boardroom for UK Finance Directors and CFOs who are already running AI processes and workflows inside their finance functions. We were joined by Adam Wilkinson, CEO at Meritec, Alex Abbey, Head of Sales & Marketing at Meritec and Gary Hodgson, Digital Sales Consultant at Meritec who brought a workflow automation lens to the compliance side of the discussion. 

Here is what came up: 

The Bright but Clunky Junior 

The dominant metaphor in the room was that today’s AI tools behave like a bright but clunky junior. Strong at language, drafting and structuring data, but weak at basic arithmetic and at holding logical consistency across a model. 

One CFO described building a ten-year financial model with AI for a venture pitch. The structure came back looking professional, yet the balance sheet didn’t balance. That gap, between something that looks right and something that is right, is the boundary every FD in the room had bumped into. 

The implication isn’t to put the tools down. It’s that accountability for the output stays with the human, not the model. 

“You’ve got to own the output. It’s like having a bright but clunky junior and you’ve got to own the output”, claimed one CFO 

Where AI is Actually Earning its Keep 

Across the room, the practical wins clustered in a few areas: 

• Executive reporting: Turning dense spreadsheets into board-ready packs such as traffic lights, executive summaries and variance commentary. So management attention lands on risks and movement rather than raw data. 

• Bridge code for legacy systems: Particularly in manufacturing, where AI is being used to write the connective tissue between archaic ERPs and modern tools. Holiday request flows and internal reporting jobs that third-party developers would quote thousands for. 

• Scenario planning without the ERP overhead: Some leaders are running multi-scenario budgets directly over large Google Sheets, bypassing traditional FP&A software entirely. The trade-off is governance for agility. 

• Onboarding and client outreach: AI agents scraping, enriching and triaging data that used to sit in a queue for an admin team. 

• Local model deployment: Where token costs on frontier models stack up for high-volume, repetitive work, some firms are hosting smaller open models in-house. 

The headcount question came up, candidly. One participant described an administrative function shrinking from eight people to two – not because AI replaced the team, but because the work itself changed. The remaining roles look more like exception-handling and oversight than process execution. 

AI is changing what finance people are paid to do, not removing them.

The Governance Problem Nobody has Solved 

Compliance and audit were where the conversation got harder. The same CFOs who are happy to let AI draft a board paper aren’t yet willing to let it near fraud risk, AML, or anything that needs an evidence trail. 

The reasons are practical not philosophical: 

• Audit logs: If a regulator or auditor asks why a specific risk was accepted, “the AI did it” is not a defensible answer. Most off-the-shelf AI tools don’t produce the audit trail a controls environment requires. 

• Data residency: It’s not simply that frontier models are hosted overseas, as most providers now offer UK or EU regions through the hyperscalers. The real question is whether your finance team knows which deployment route they’re on and whether that’s documented. Many don’t. 

• Workflow versus chat: Adam Wilkinson at Meritec made the point that the compliance value isn’t in the chatbot, it’s in the workflow underneath. Connecting data, approvals and evidence into one chain gives leaders visibility over risk control and meets ECCTA obligations. Several FDs recognised the gap. They had AI doing analysis, but no workflow infrastructure to make the resulting decisions auditable. 

The honest summary was that most UK finance functions can show you AI in use. Far fewer can show you AI under control. 

What they would do Differently 

When asked what they’d change with hindsight, three themes recurred: 

• Start with reporting and drafting, not with the GL. The wins are faster and the downside is contained. 

• Don’t outsource prompt design. The FDs getting the most out of AI are treating prompt and context engineering as a finance skill, not an IT one. 

• Budget for the boring infrastructure, such as logging, version control and access management, before scaling use. Retrofitting governance is harder than building it in. 

What this means for the Finance Leadership Job 

Two shifts came through clearly: 

Firstly, the skill mix on finance teams is changing. There is growing demand for hybrid profiles. People who understand both accounting principles and how to assemble an AI workflow. Not data scientists, not pure FP&A analysts. Something in between. 

Secondly, the FD and CFO role itself is being pulled further toward judgement and away from production. If AI can produce a passable first draft of almost anything in finance, the differentiator becomes knowing what to ask of it, what to check and what to escalate. 

“My concern is whether AI is yet at the level of confidence an enterprise actually needs”, stated one participant 

For now, the answer in the room was not quite, but closer than last year and closing fast. 

Join the Next Conversation 

If you missed our roundtable, or want to pick up where it left off, our next peer-to-peer session on AI in the Finance Function is now open for registration: Book here

With special thanks to Meritec for joining our session.