What is a client actually judging?
Very few clients can assess the technical quality of what you do. They are not qualified to, which is why they hired you. So they judge the things they can see: whether you came back when you said you would, whether the document made sense, whether they felt understood in the meeting, and whether anything landed on them as a surprise.
That is the honest starting point for this question. AI does not change what clients value. It changes how fast certain things reach them, and how much of the firm's attention sits behind each one. Those two effects pull in opposite directions, which is why the same firm can make its service noticeably better and noticeably worse in the same quarter.
Where does speed genuinely improve the service?
Wherever the client is waiting, and especially where they are waiting to make a decision of their own. A buyer who cannot exchange until your report arrives. A finance director who cannot close the month until your figures land. A board that has deferred an item because the paper was not ready.
In those situations speed is not a nicety. It is the service. The client is carrying risk, cost or irritation for every day the thing is outstanding, and no amount of polish in the eventual document buys that time back. The cost of a late report is mostly borne by the client, which is exactly why they remember it.
The same applies to the small, dull moments that shape how a relationship feels. Confirmation of what was agreed arriving the same afternoon rather than next week. An onboarding pack that does not ask for the same information twice. A straightforward factual question answered in an hour instead of sitting in an inbox. Each is minor in isolation. Together they are what clients mean when they say a firm is easy to deal with. The processes behind them, meeting notes and follow-ups, client onboarding and email triage, are unglamorous and almost invisible when they work.
Where does it start to feel impersonal?
At the point where output grows faster than thought. The warning signs are consistent across firms.
- Length without a view. A twelve page report that sets out the position fully and never says what the client should do. The client wanted the last paragraph and had to pay attention through the other eleven pages to find out it was not there.
- Bad news in a standard wrapper. A delay, a cost increase or an unfavourable finding delivered in the same neutral register as a routine update. Clients forgive problems. They do not forgive being handled.
- Volume for its own sake. More updates, more dashboards, more attachments, because producing them became cheap. The client now has to work out which of your emails matters.
- Answers that arrive before the question was understood. A fast reply to the thing that was asked, when the client was circling something they had not yet articulated.
None of this is caused by the technology. It is caused by treating faster production as the whole objective. If you reduce the time it takes to assemble a document and then add nothing to the time spent deciding what the document should say, the client experiences the change as a drop in attention.
Is it the AI they object to, or the genericness?
In most cases, the genericness. Clients have been receiving template-shaped correspondence from professional firms for decades, and they have always been able to tell. Templates rot precisely because they drift towards the generic as people stop editing them.
What is different now is that producing plausible generic content costs almost nothing, so there is no natural brake. A partner who would not have had time to write four pages of background will happily approve four pages that appeared in thirty seconds. The volume rises, the signal of effort falls, and the client reads it as distance.
This is also why hiding the method tends to backfire. If a firm is confident in how it reviews work before it goes out, saying so is straightforward, and telling clients you use AI is easier than being asked about it later.
How do you design the split?
The workable rule is to speed up the assembly and protect the judgement and the delivery. Assembly is gathering the facts, pulling the figures, putting them into the firm's format, checking the file references and getting a complete draft in front of the right person. That work is necessary, repetitive and worth almost nothing to the client in itself.
Judgement is what the facts mean for this client, what you recommend and what you would do if it were your money. Delivery is how it is said, how bad news is framed, and the phone call that should precede the email. Both of those are the reason the client pays you, and the work that stays human sets out where the line sits.
In practice that means a named person reads and signs off anything that leaves the firm. Not a glance at the formatting: a read against what they know about the matter. Reviewing AI output on client work covers how to make that review real rather than ceremonial, and the risk in copy and edit explains what happens when it is not.
What should you measure to know which way it has gone?
Satisfaction scores arrive too late and tell you too little. Look at operational signals instead, and look at them monthly.
- Days from the triggering event to the client receiving the output. Meeting to confirmation, instruction to first draft, period end to report.
- Chaser emails received. If clients are asking where something is, you already know the answer.
- Work reissued after an error or an omission, which is where speed turns expensive. The cost of rework covers how to count it.
- Time senior people spend on client-facing thought rather than assembly. If the admin fell and that figure did not rise, the saving went somewhere unhelpful.
That last one decides the answer to this whole question. Taking four hours out of a reporting cycle improves client service only if some of those hours come back as attention. The numbers worth a monthly look puts this alongside the rest.
Where should a firm start?
Start where clients are waiting longest, not where the technology looks most interesting. For most firms that is one of a small number of jobs: the periodic update in client reporting, the note after a meeting, or the first two weeks of a new engagement. Fix the wait, keep the judgement in human hands, and the client experiences both halves of the improvement.
Firms that get this wrong usually did it the other way round: they made the clever part faster and left the client waiting in exactly the same place. The audit will tell you which process is costing your clients the most patience, and your fee earners the most unbillable time.