What an accountant who runs his firm on AI told us about the next shock

On its own, AI makes a firm faster at whatever it already does, good or bad. The firms that do well with it start by understanding the tasks and processes that create value, then put the data, security, tech stack, compliance and governance under them in order. David Carr made the same case when we spoke to him last week.

David is the founder of Carr Accounting Studio. He trained at a Big Four firm and has since worked in fintech and accounting software, so he has seen finance from the practice side, the software side and the regulated payments side. This year he set up his own ICAEW chartered practice, and he has run it with AI from the first day.

AI makes a bad process worse, faster

David put the risk plainly. If you've got a bad process and you automate it or plug AI into it, "you just get more quickly to a bad answer. It literally makes things worse."

In a firm, a bad process shows up as jobs that keep running over. Every job has a time budget, and the fee is set from it. When a job runs over, the extra time is written off, so the firm bills less of the time it recorded and its recovery rate slips. Put AI on a task like that and it speeds up what is wrong with it, while the write-offs carry on.

So we start by breaking a firm's work down into the individual tasks and processes that create value for its clients: who does each one, how long it takes, what information it uses and who signs it off. Where a process is wrong, fixing it goes first on the roadmap. When AI then brings that job in inside its budget, less is written off, and the recovery rate, fee income and gross profit rise together, so the same team can take on more clients without adding to its headcount.

The data has to be ready before the next shock

A sound process still depends on what sits underneath it, and David thinks the firms that have that in order will cope best with what comes next. As he sees it, HMRC keeps moving the rules, US politics is unsettled, two of the frontier labs may be heading for public listings, and nobody yet knows what Europe and the UK will do about sovereign AI. A firm without its data in order, or a starting point for its compliance and information security, has to catch all of that up before the next big shock arrives. "No matter where you are," he said, "you've just got to take that next step."

Those are the layers every AI solution sits on: the tech stack and whether its software can connect to AI, the quality, completeness and security of the data, and the firm's compliance and governance. AI will work happily from incomplete or inconsistent data and give a confident, wrong answer, and the time spent putting that right is written off like any other overrun. So we check each layer before any solution is planned, and where one isn't ready, fixing it comes first on the roadmap.

To see where AI would free up your own team's time first, the quiz below takes a few minutes.

A firm should own its knowledge and rent the AI

Where the firm's knowledge lives is one of those layers, and David's own firm shows what good looks like. His workpapers, SOPs, checklists and call notes live in Obsidian, in plain files the firm owns, and his AI tools work from them. He has also tested running open source models on his own laptop, so that if the plug ever got pulled he knows he could keep operating.

Providers do switch things off. In June, Anthropic had to switch off two of its newest models, Fable 5 and Mythos 5, for every customer after a US government directive arrived with no warning (Anthropic). A firm whose know-how lived inside that one product's chats had nowhere to go, while a firm like David's could point a different AI at the same material and carry on. Prices can rise too, and a firm tied to one vendor has to pay them.

So every solution we implement is set up so a different model can take over, and where it runs follows from the data it touches. It uses the AI provider's own cloud only when nothing sensitive is involved. Client work runs on the firm's own server, where the provider never trains on what it sends, or on a machine in the office, where the data never leaves the building. That choice goes into the firm's AI policy before anything goes live.

Solutions have to be managed once they are live

Owning the setup also means looking after it. David was honest that his own isn't perfect: you have to manage it, the models change underneath you, and an accountant needs to be able to trace each number and know whether it can be relied on.

When a model changes, a solution that worked last month can drift, and if nobody is looking, its quality drops without anyone noticing. So every solution has a named owner who checks what it produces before anything reaches a client, and it is checked and tweaked each time the model underneath it changes. That keeps the standard of the work where it was, or raises it.

What firm leaders should do now

A lot of David's peers are unsure how to use AI at all, and he puts much of that down to the noise and fear on LinkedIn. Accountants, he said, are "quite sharp to that. They're not easily tricked, but they are easily overwhelmed." The way through the noise is a clear order of work, whoever does it.

Start with the firm's own work, broken down into the individual tasks and processes that create value for its clients.

Before any solution, check the layers those tasks sit on: the business and each department, the tech stack, the quality, completeness and security of the data, compliance and governance, and the people who will work alongside the AI. Where a layer isn't ready, fixing it comes first.

Then set out a roadmap of solutions, each fully costed with the hours it frees up, what it costs to implement, its first-year return and what it does to the recovery rate on the jobs it touches.

Implement the solutions one by one, in the roadmap's order. Each runs where its data allows, has its risks designed out, is covered by the AI policy and has a named owner who checks what it produces before it reaches a client. Once they are live, keep managing them as the models change, with the firm owning its knowledge and renting the AI.

Done in that order, the firm gets more capacity from the same team, hits its chargeable time targets and lifts its recovery rate and fee income, in a way that is compliant and keeps client data secure.

See where your firm stands

Our quiz asks how much of your team's week goes on repeat work, how often jobs run over budget, how your team uses AI, where client data goes and how you would know whether AI paid off. Your report shows where AI would free up your team's time first, the capacity it could give your firm in hours, and a 30-day plan for what to do.

If you want our support putting it in place, book a free call.