The AI question your clients are asking is the wrong one

In practice, what holds a client back is almost never the absence of a clever tool. It is that their processes are not stable enough, or documented enough, or clean enough underneath, to be handed to anything, automated or not. Ask an owner to explain exactly how a quote becomes an invoice, or how a job gets costed, and you will often hear a description that only makes sense because they are the one doing it. The logic lives in their head. The data sits in three systems and agrees in none of them. That is a process you cannot safely automate, because you would be handing a machine a set of judgements that nobody has written down.

So, the useful question is not ‘Which tool?’ It is: ‘Which of our processes are stable, repeatable, rules-based and clean enough in the data to trust to a machine?’ Most owners cannot answer that. We usually can, or we can find out quickly, because we are already looking at the numbers those processes produce.

This is our conversation to lead

Accountants are unusually well placed here, and I think we underplay it. We see the ledger, the exceptions, the reconciliations that never quite tie out. We can tell which processes are genuinely repetitive and which ones carry judgement that should stay with a person. When a client is excited about AI, the most valuable thing we can do is slow them down for a fortnight and run a simple triage. What is high-volume and rule-based? What is low-volume and judgement-heavy? What data would an automated process rely on, and can we trust that data today?

There is a familiar lens for this. The discipline that makes a business ready for investment is the same discipline that makes it ready to benefit from AI. Clean, structured, consistent, well-governed numbers. A buyer will not pay a premium for a business whose performance cannot be explained. A machine cannot improve a process whose inputs cannot be relied on. It is the same requirement, arriving in different clothes.

Where to start

The path is unglamorous, which is why it works. Pick one or two processes that are genuinely repetitive and high-volume. Get the data underneath them clean and consistent. Write the process down so it no longer depends on one person remembering it. Then, and only then, automate. Then measure what it did to time and to margin, honestly, before moving to the next one.

Owners who follow that sequence tend to end up with something more useful than a new tool. They end up with a business that runs on documented, measured processes, whether or not a machine touches them. The AI becomes almost incidental. The discipline is the asset.

I am not arguing against AI. I am arguing against buying it as a substitute for work that has not been done. Our clients do not need us to be excited alongside them. They need us to ask the harder question first and to help them earn the right to a good answer.

Nitesh Roopa is a chartered accountant and founder of ProfitPulse.