AI can draft the board pack commentary. Whether the board can rely on it depends on what your team does with that draft.
Data and AI Skills for Finance Teams
By James Cotton · Last updated · 8 min read
Part of our topic guides on AI Skills for Business and Data Literacy.
By James Cotton, Founder, iO-Sphere
Drafting the board pack with AI
Month end has closed, the management accounts are signed off, and someone now has to write the commentary that goes to the board: how the business performed, where cash stands, what has changed in the outlook. Given the accounts and last month's pack, an enterprise AI tool will produce a readable first draft quickly.
The tool is good at work that takes time without taking judgement. It turns tables into sentences and holds the house style from month to month. It can also move a decimal point, work a percentage on the wrong base, or supply a plausible cause that nobody in the business gave it.
So the analyst who owns the pack checks it before anyone else reads it. Every figure is ticked back to the management accounts. Every stated reason is confirmed with the person it came from, or cut. Anything the tool could not know, such as a contract that slipped last week, is written in by a person.
The check leaves something behind: the source the tool was given, the draft it produced, the corrected version and the sign-off, filed together. When the audit committee or the auditor asks how the commentary was produced, the answer is a folder, not a recollection.
One decision comes before all of this: which tool, and what goes into it. For a company with listed securities, Article 10 of the UK Market Abuse Regulation (UK MAR) treats disclosing inside information to any other person as unlawful "except where the disclosure is made in the normal exercise of an employment, a profession or duties".
Our own advice, separate from anything the regulation says, is that inside information goes only into tools the company has approved. Written into the team's staff rules for AI tools, that becomes something every person can follow without having to ask.
What the team needs to be able to do
The board pack shows most of what a finance team needs. The rest sits upstream, in the data the pack is built from.
| Ability | Where it shows in the board pack |
|---|---|
| Build the numbers | Ledger and operational data assembled into the figures the pack reports |
| Automate the repeat | The monthly extract lands in the template without retyping |
| Choose the tool and inputs | An approved enterprise tool; no client data or inside information in public ones |
| Trace every figure | Each number in the draft ticked back to the management accounts |
| Challenge the explanation | Each stated cause confirmed with the person it came from |
| Keep the record | Source, draft, corrections and sign-off filed together |
The first two rows are data skills, and they belong mainly to analysts and to whoever owns the routine. The last four are expected of anyone whose name goes on a number, whether or not they ever build a model.
The data work behind the pack
Before a word of commentary exists, someone has assembled the numbers. Data comes out of the ledger, is joined to sales, headcount or operational data, is reconciled, and is shaped into a forecast or a dashboard the business can read and question.
The national standard for data analysts at Level 4 describes an occupation "found in any employer in any sector that uses data to make business decisions". It names finance among the departments where analysts work and lists "Departmental Data Analyst" among its typical job titles.
Its duties include collecting, compiling and cleansing data and producing performance dashboards and reports, so a financial planning and analysis (FP&A) analyst who builds the pack's figures is already doing part of that job.
The ability that matters most here is being able to say where each figure came from. It is the same ability that makes an AI-drafted commentary checkable, which is why the data skills and the AI skills on this page are one subject.
Routine work to automate, and who builds it
Finance carries a great deal of work that repeats on a timetable. Ledger extracts are pulled into the same templates, schedules are rolled forward from last period, recurring journals are prepared, the same checks are run in the same order. Every one of these is a candidate for automation.
The people who own a routine task are the best placed to build the automation that runs it. The assistant accountant who prepares the month-end extract knows where the source system changes, which exceptions come back every month, and what a wrong output looks like at a glance. That knowledge is what makes an automation safe to rely on.
In the board pack, automation is the step before the draft. The figures arrive in the template the same way each month, nobody retypes them, and the analyst's time goes on checking the commentary and adding what no tool can know.
What the professional bodies and the audit regulator say
Two recent pieces of guidance are addressed to narrower groups: one to tax work, one to audit firms. In January 2026 the seven professional bodies that jointly prepare Professional Conduct in Relation to Taxation (AAT, ACCA, ATT, CIOT, ICAEW, ICAS and STEP) issued joint guidance on AI tools. It applies to their members, including "those working in business", who use AI "when advising on UK tax matters".
The Financial Reporting Council (FRC), after its first guidance on AI in audit in June 2025, issued Generative and Agentic AI Guidance in March 2026 for audit firms. Neither governs a management board pack. A finance team can still take its practice from both, because both describe the review an AI draft needs.
The tax guidance says AI output "should also be regarded as if it were prepared by a less experienced junior colleague and reviewed with appropriate scepticism". Members "are ultimately responsible for any work they produce", and the principle of professional competence and due care "requires members to ensure that staff receive appropriate training for any AI tool used".
The junior-colleague test transfers well to the board pack. A first-year analyst's draft would never reach the board unchecked, and a tool's draft earns the same treatment.
The FRC's announcement states that "the human auditor is always accountable". Two of the three risks its factsheet names are "misuse of output", where the user misreads what the tool produced, and "deficient output", where flawed inputs or design produce something wrong that is then relied on. Among its mitigations is "staff education and governance".
A finance team inside a bank or insurer works under a further layer of rules, which our page on AI governance in regulated sectors deals with.
What the surveys show
The most detailed UK evidence on AI in reporting comes from research the Financial Reporting Council released on 8 July 2026: 39 interviews and an online survey with 103 responses, run for the FRC by Lancaster University with Loughborough University academics and open from 17 February to 21 April 2026. The respondents prepare corporate reports and lean towards listed and financial companies; fewer than half (43%) work in finance.
Of those who gave a definite answer, 39% were using generative AI, 31% piloting it and 18% considering it within a year. Among the 70 piloting or using it, the survey results put drafting narrative sections first, at 61%, and copyediting narrative second, at 57%. Board pack commentary is the same kind of task: narrative drafted from figures.
Among the same 70, 91% used enterprise tools and 24% also used public ones. Fewer than half (47%) had mandatory human review of AI outputs in place, and interviewees told the FRC that training was "patchy" and "primarily focused on risk mitigation".
Asked where their organisation uses automated technologies such as robotic process automation (RPA) or machine learning, 68% of the 50 respondents to that question named financial data extraction and 34% journal entry preparation.
Surveys by finance professionals' own bodies report a skills gap. In ACCA and CA ANZ's global survey, published July 2026, 72% reported only basic or no generative AI skills, and 93% had at least moderate concern about relying on AI-generated insights they cannot independently verify. In the AICPA and CIMA global survey of finance and accounting leaders and managers (2025), generative AI was the most-cited skills gap, at 46%.
One qualification has moved with it. CIMA's CGMA Professional Qualification includes generative AI questions in the E1, P1 and E3 objective test exams from January 2026, and in case study exams for all pathways from May 2026.
Three routes into the work
Analysts who build the numbers can take Advanced Data & AI (Level 4), which is laid over the finance role they already hold; the Level 4 data analyst standard has its own guide.
The people who own routine work can learn to automate it on the AI Automation short course, five evening sessions at £1,195 + VAT, or go further on the AI Transformation apprenticeship (Level 4).
Everyone else needs the foundations: choosing the tool, checking the output, keeping the record. Our foundations training works with a whole team, and custom programmes are priced per engagement. How each route can be paid for is in our guide to funding and scoping team training.
iO-Sphere has trained teams at PwC, BAE Systems, Dunelm and Wunderman Thompson.
Frequently asked questions
Can a finance analyst do a data apprenticeship and stay in finance?
Yes. Advanced Data & AI is laid over the job the person already holds, and the project it ends with is a real analysis task from that job, so the work stays in finance.
Should finance staff use ChatGPT for reporting?
Only the version the company has approved, with data the team is allowed to put into it. In a 2024 Chartered Accountants Worldwide survey, 70% of the chartered accountants surveyed who used AI at least monthly used publicly available chatbots such as ChatGPT or Gemini at least once or twice a month; the survey says its results are not intended to be representative of the profession.
What should an auditor expect to see?
The folder: the source the tool was given, the draft, the corrections and the sign-off. Raise it early, too. Among the 70 FRC respondents piloting or using generative AI, 21% had discussed its use with their external auditors and 17% had given auditors access to AI tools, logs or outputs.
Upskilling a whole team?
Tailored data and AI training for organisations, from data literacy to technical upskilling, built around your team's real work.