AI and Accountants: What's Actually Happening and What to Do

The honest assessment

Let me be blunt. If your day involves categorising transactions, reconciling accounts, preparing standard tax returns, or producing management accounts from a template... AI can already do most of that. Not theoretically. Right now. Xero and QuickBooks have had auto-categorisation for years, but the new wave of tools built on large language models is a different beast entirely. Microsoft Copilot sitting inside Excel can write formulas, spot anomalies in datasets, and summarise financial statements in seconds. ChatGPT can draft client-facing tax summaries that would take a junior an afternoon.

What AI can't do yet is the stuff that keeps senior accountants up at night. Complex tax planning across multiple jurisdictions. Reading between the lines of a client conversation to figure out what they're actually asking. Sitting in front of HMRC during an enquiry and making a judgement call about what to disclose and when. The professional scepticism that comes from twenty years of seeing creative accounting... that's not in the training data.

But here's the thing that should worry you. The "can't do yet" list is getting shorter every quarter. Two years ago, AI couldn't reliably interpret UK tax legislation. Now Claude can parse the Income Tax Act and give you a reasonable first-pass analysis of a scenario. It's not perfect. It hallucinates. But it's getting better at a pace that should make anyone in compliance-heavy work pay attention.

The parts that are actively changing right now: audit sampling, variance analysis, client correspondence, and anything involving pattern recognition across large datasets. The Big Four are pouring billions into this. Deloitte's AI tools are already doing work that graduate trainees used to do. PwC has committed over a billion dollars to AI deployment. If you're at a smaller firm and think this doesn't affect you... it does. It just hasn't hit you yet.

Your exposure level: High

I'm not going to sugarcoat this. Accounting has one of the highest exposure levels to AI disruption of any profession. A Goldman Sachs report estimated that 46% of tasks in finance and accounting could be automated by current AI technology. Not future technology. Current.

The reason is structural. Most accounting work involves taking information from one format, applying rules to it, and outputting it in another format. That is precisely what large language models and automation tools are good at. Bookkeeping is already being hollowed out. Tax preparation for straightforward returns is heading the same way. Even advisory work is getting squeezed as AI gets better at scenario modelling.

That said... "high exposure" doesn't mean "everyone gets sacked next Tuesday." What it means is that the ratio of accountants to clients is going to shift. One accountant with good AI tools will be able to handle the workload that currently takes three. Firms will need fewer people. The ones who stay will be the ones who learned the tools early and can do more with less. i've sat in enough restructuring meetings to know how this plays out. The people who survive aren't always the most skilled. They're the ones who made themselves look indispensable before the spreadsheet came out.

The 90-day action plan

1. This week: test AI on your actual work. Take your last three sets of management accounts. Paste the P&L into ChatGPT or Claude and ask it to write a commentary. Compare it to what you wrote. Notice what it gets right and where it falls flat. That gap is your value. Learn it.

2. Week two: automate one recurring task. Pick the thing you do every month that makes you want to scream. Bank reconciliation formatting. VAT return data prep. Whatever it is, spend two hours figuring out if Microsoft Copilot or a GPT can do it. Even a 50% time saving on one task compounds fast.

3. By day 30: learn to prompt for tax research. Open Claude and give it a specific client scenario. "UK-based sole trader, turnover of £90k, working from home, wants to know about capital allowances on a new laptop and office furniture." See what it gives you. Learn what follow-up questions make the output better. This is a skill, and it's learnable.

4. By day 45: build a client communication template system. Use ChatGPT to draft five standard client emails you send regularly. Year-end reminders, information requests, fee quotes. Refine them until they sound like you. Save them. You've just bought yourself hours every month.

5. By day 60: get visible. Present something at a team meeting. Doesn't have to be fancy. "I tested Copilot on our monthly reporting process and here's what happened." Show the before and after. Partners notice this stuff.

6. By day 75: pick an advisory niche. The accountants who will thrive are the ones doing work AI can't easily replicate. Cash flow forecasting with strategic context. Sector-specific tax planning. Business acquisition due diligence. Pick one area and start going deeper.

7. By day 90: audit your own role. Write down every task you do in a typical month. Put them in three columns: "AI can do this now," "AI can help with this," and "AI can't touch this." If that third column is thin... you know what you need to work on.

AI tools you should be using this week

• Microsoft Copilot for Work — Lives inside Excel and Outlook, which is where you already are. Use it for formula generation, data summarisation, and drafting client emails directly from financial data. The Excel integration alone is worth the subscription for accountants.

• ChatGPT for Work — Best for drafting management account commentaries, summarising long HMRC guidance documents, and generating first-pass tax research. The GPT-4 model handles UK tax scenarios surprisingly well if you prompt it properly.

• Claude for Work — Handles longer documents better than ChatGPT. Excellent for pasting in entire sets of accounts or lengthy tax legislation and asking specific questions. i use it for parsing Companies House filings and extracting key data points.

• Perplexity for Research — When a client asks you something obscure about a tax relief or filing deadline, Perplexity searches current sources and gives you a cited answer. Faster than trawling through GOV.UK, and it actually finds the relevant page.

What to say in meetings

Next time your firm has a partners' meeting about "digital transformation" or "AI strategy," don't sit there silently. Say: "I've been using Copilot to cut our monthly reporting prep time by about 30%. Happy to show the team how it works." That's it. You've just positioned yourself as the person who's already doing it rather than talking about it.

If a client asks whether AI will replace their accountant, try: "It's going to replace the boring bits, which means we can spend more time on the parts that actually save you money." Clients like that. It's honest and it reframes the conversation around value.

And if leadership starts talking about headcount in the context of AI, make sure you're the one saying: "I've been thinking about which of our processes could be streamlined. Want me to put together a short proposal?" Better to be driving the change than being driven by it.

If the worst happens

If you're made redundant from an accounting role, your skills transfer more broadly than you might think. Financial analysis, management reporting, and an understanding of business performance are valuable in operations, consulting, and FP&A roles across every industry. Companies always need people who can look at numbers and tell them what they mean.

The adjacent roles to look at: financial analyst, business analyst, management consultant, and internal audit. All of these use the same core skill set but are positioned slightly differently in the market. If you have client-facing experience, fractional CFO and finance director work is growing... small businesses need financial leadership but can't afford someone full-time. AI tools actually make this more viable because one person can manage more clients.

The honest truth about the market: there are fewer pure bookkeeping and compliance roles every year, and that trend is accelerating. But there's growing demand for accountants who understand both the numbers and the technology. If you've spent your 90 days building real AI skills with real tools, you're not just another redundant accountant. You're someone who knows how modern finance operations actually work. That's a different conversation entirely.