AI Skills for Accountants: What to Actually Learn Right Now

So you've read the doom articles. You've probably read the one about whether AI will replace accountants. And you're still here. Good. That means you're past the fear stage and into the "right, what do I actually do about this" stage.

That's the stage where things get better. i promise.

I was made redundant from a data science role. Spent about three weeks refreshing job boards and feeling sorry for myself before i realised the only useful thing i could do was learn the tools that replaced me and figure out how to work alongside them. Turns out that's quite a marketable skill.

Here's what i'd focus on if i were an accountant right now.

The skills that actually matter

Not "learn to code" or "understand machine learning." Specific, practical skills that will make you more valuable next quarter.

1. Prompt engineering for financial analysis. This sounds wanky but it's genuinely the most important skill. Knowing how to ask an LLM to analyse a set of management accounts, flag anomalies, and produce a narrative summary... that turns a two-hour job into a fifteen-minute job. The accountant who knows how to prompt well produces more than the one who doesn't. It's that simple.

2. AI-assisted audit and reconciliation. Understanding how to set up AI-powered rules for transaction matching, exception handling, and audit sampling. Not just clicking buttons in software. Understanding what the tool is doing so you can spot when it gets it wrong. Because it will get it wrong.

3. Automated reporting workflows. Building chains where data flows from your accounting system through an AI summarisation layer into client-ready reports. The accountant who can set this up for a practice saves everyone hours. The one who can't is the one whose tasks get redistributed.

4. Data cleaning and preparation. Half of accounting AI tools fall over because the data going in is messy. If you can clean, structure, and prepare data for AI processing, you become the person who makes everything else work. It's not glamorous. It's essential.

5. AI output validation. This is the one most people skip. Being able to systematically check what AI has produced against source data, regulatory requirements, and professional standards. Your professional judgement is the quality control layer. Make sure everyone knows it.

Tools to learn first

ChatGPT or Claude for financial narrative. Pick one. Learn to use it for drafting management account commentaries, explaining variances, and producing client-facing summaries from raw numbers. Start with your own data. The key skill is knowing what to include in the prompt... give it the numbers, the context, and the format you want. Then edit the output with your professional brain.

Dext or Datamolino for document processing. These tools extract data from invoices, receipts, and bank statements using AI. If you're in practice, knowing how to set up and optimise one of these properly will save you hundreds of hours a year. The trick is in the categorisation rules. Spend time getting those right.

Power Automate or Zapier for workflow automation. Not strictly AI, but the glue that connects AI tools together. Set up a flow where client documents arrive, get processed by an AI extraction tool, land in your accounting software, and flag anything unusual for your review. This is where the real time savings live.

How to demonstrate these skills

Nobody cares about your AI certificate. Sorry. I know you probably just bought one. Here's what actually matters.

Build a before-and-after case study. Take a process you currently do manually. Time it. Then do it with AI tools. Time that. Document the difference. Show it to your manager. This is worth more than any qualification because it proves you can apply AI to real work.

Volunteer to be the AI pilot person. Every practice, every finance department is thinking about AI adoption. Most are paralysed by indecision. Be the one who says "I'll test it on my workload and report back." This puts you at the centre of the firm's AI strategy without anyone having to make a big decision.

Document your professional judgement calls. Every time you override an AI suggestion or catch something it missed, write it down. Not in a smug way. In a "here's where human expertise added value" way. You're building evidence that the human layer matters.

Share what you learn with colleagues. Run a lunch-and-learn. Write an internal note. The person who helps others adapt becomes indispensable in a different way.

The 1-hour weekend project

Here's something concrete you can do this Saturday.

Take last month's management accounts for any client or your own department. Paste the key numbers into ChatGPT or Claude. Ask it to: identify the three most significant variances from the prior period, suggest possible explanations, and draft a two-paragraph commentary suitable for a board report.

Compare what it produces with what you or your team actually wrote.

You'll learn three things. Where AI is surprisingly good (usually the variance identification). Where it's mediocre (the explanations often lack context). And where you add value (the professional judgement about what actually matters and what's just noise).

One hour. No cost. And you'll understand the technology better than 90% of your peers. That's not a bad return on a Saturday morning.

What to do right now

Pick one AI tool from the list above. Not all of them. One. Spend a proper afternoon with it this week. Use your actual work data, not tutorial examples. Get it wrong a few times. That's how you learn what it can and can't do.

The accountants who'll thrive aren't the ones with the most certifications. They're the ones who actually use these tools every day and know where the edges are. Be that person.

If you want to understand the bigger picture of what's happening to accounting roles, that's worth a read too. But the doing matters more than the reading right now.