AI for accountants: what accounting firms actually use

Start with the AI you already pay for

Your ledger, receipt capture and practice management software probably already include AI features. Xero has its JAX assistant and reconciliation suggestions. QuickBooks has Intuit Assist. Receipt tools read supplier, date and amount from a photo or PDF. Many practice management systems now offer email summaries and draft replies.

These features often need setting up and some patience. Suggested bank matches improve as staff correct them, and a feature switched on for one client may not suit another. Before trialling a new product, ask the person who uses each system day to day what is already available and what they have tried.

Write down what you find in a simple list: the tool, the feature, who uses it and whether it helps. Firms are often surprised by how much is already there and not being used.

General assistants: drafting, explaining and summarising

Staff use general assistants for writing a first draft of a client email, explaining a tax rule in plain words, turning meeting notes into actions or summarising a long letter from HMRC. Used this way, the assistant saves typing and thinking time, and a qualified person still decides what is sent.

Agree simple rules before people use these tools with client work. Which tools are approved, on which plan, with which settings? What client information may go in, and what must never go in? Business plans from the major providers give more control over how data is stored and used than free personal accounts. Check your professional body guidance and your engagement letters as well.

Many firms find staff are already using personal accounts quietly. A short written policy and an approved tool are safer than a ban that people ignore.

Specialist tools for audit, tax and review

Larger firms are trialling AI tools that read audit documents, search tax sources or check workpapers against a standard. These can be useful where a firm does a lot of the same specialist work and has someone to own the trial.

For a smaller practice, a specialist tool usually only makes sense once you know exactly which recurring job it would take on and how much time that job takes today. Ask the supplier to demonstrate it on a made-up example that looks like your work, including the awkward cases.

What AI gets wrong in a practice

AI tools can be confidently wrong. They may quote a rule that has changed, invent a reference or misread a figure on a poor scan. In accountancy those errors matter, so every AI output that reaches a client or a return needs a person who checks it against the source.

Build the check into the job rather than relying on people to remember. For example, an AI-drafted reply to a client query should show which document or rule it relied on, so the reviewer can confirm it quickly. If checking takes longer than doing the work by hand, that job is not a good first choice.

Before you hire an AI person

Some firms respond to AI by advertising for an AI or automation analyst. That can work well, but the new person will spend their first months finding out which jobs matter, who does them and what the partners want. If nobody has agreed that list, the role can drift into tool demos.

A short piece of mapping first gives a new hire, or your existing team, a clear starting point: the few jobs worth changing, in order, with an owner for each.

How to choose your first job for AI

Pick one recurring job and score it against four questions. How often does it happen? How much hands-on time does it take? Can a person check the result quickly? Are you allowed to use the information it involves in the tool you have in mind?

Good first jobs tend to be frequent, text-heavy and easy to check: drafting replies to common client questions, preparing a records request from a checklist, or summarising a new client’s documents for the person onboarding them. Poor first jobs involve judgement on tax positions or anything where a mistake would reach HMRC unchecked.

Illustrative example, not a client case: a practice lists three candidates. Drafting replies to common payroll questions happens daily, takes about half an hour a day and is quick to check, so it scores well. Reviewing year-end accounts is high value but hard to check quickly, so it waits. Chasing missing records happens weekly and depends more on a clear checklist than on AI, so the practice fixes the checklist first.

Common questions