How do you win over a partner who is against AI?

Why is the sceptic usually worth listening to?

Every firm of 20 to 200 people has one. A partner, often a good one, often the one with the longest client list, who will not have it. The rest of the partnership reaches for the obvious reading: they are behind the times, they will come round, carry on without them. That reading is usually wrong, and acting on it costs the firm a year.

The sceptic is rarely objecting to technology in the abstract. They are objecting to something concrete that nobody has answered. Their name goes on the advice. Their relationship absorbs the damage if a figure is wrong. They have watched the firm buy things before: the intranet, the client portal nobody logged into, the module that was going to end double entry. They carry a working memory of promises that did not land, and that memory is an asset.

Treat the objection as information. In most cases it is a design requirement nobody has written down yet.

What are they actually objecting to?

Push past the first answer and the objections get specific. They are almost always one of these.

• My name is on it. A draft that looks right but is wrong is more dangerous than a blank page, because it invites a skim rather than a read. This is the serious one.
• Client information leaving the firm. Where does the material sit, who can see it, what is kept.
• The juniors will never learn. If the drafting goes, what does a second-year do, and how do they become the person who spots the error in year six.
• We tried this already. The firm bought licences, ran a session, nothing changed. To the sceptic that is evidence. To everyone else it is an inconvenience to skip past.
• It is a distraction. A project that eats management attention for a year is no answer to a capacity problem this quarter.

Notice what is missing. Almost nobody says they prefer typing. The objection is about control, exposure and wasted effort.

Which arguments make it worse?

Three approaches reliably harden the position.

The first is the competitive threat. Tell a partner with thirty years of relationships that firms who do not adopt AI will not exist in five years and you invite them to take the other side of the bet. They will. Nobody changes a working practice because of a prediction.

The second is the demonstration. A polished example built on somebody else's documents proves nothing to a person whose entire doubt concerns their documents, their clients and their standards.

The third is going around them. A quiet project in another team, presented afterwards as settled, turns a technical disagreement into a governance one, which is a far harder argument to win. It is also a different problem from getting the partners to agree.

How do you turn an objection into a design requirement?

This is the whole move. Write each objection down as a condition the rebuilt process must meet, then satisfy it in the build rather than in conversation.

My name is on it becomes: nothing reaches a client without a named person reviewing and signing it off. AI drafts, people approve. It also becomes a question about what the reviewer is handed. A draft that cites its sources, flags what it could not find and leaves gaps visible can be reviewed. A fluent draft that hides its assumptions is exactly the thing the partner fears, and they are right to. The guide to reviewing AI output on client work sets out what a real review looks like.

Client information becomes a written answer on where material sits, what is kept and who can see it, in the same terms the firm already uses with its own suppliers. Data protection in AI client work covers it. If the firm cannot answer that for a given process, start elsewhere.

The juniors becomes an explicit decision about what a junior does instead and who reviews what. We argue that case in what changes for junior staff. It is a staffing question with an answer, not a reason to stop.

We tried this already is your strongest point, if you use it. The firm bought a tool and expected a process. A licence does not change how a job is done, and more training does not fix it either; the job itself has to be rebuilt. That is the case in a tool is not a process, and it is the one a sceptic accepts most readily, because it confirms what they watched happen.

The distraction becomes a boundary: one process, a fixed scope, a date it goes live, a fee agreed in writing first. If nobody can say how many hours this will take from their team, refusing is the correct response.

What should you put in front of them?

Not a deck. One process, rebuilt on the firm's own templates and language, running on live work, with a before and after anybody can check.

Choose it carefully. The best candidate is one the sceptic does not own but can see clearly: something repetitive and visibly painful, such as client reporting or meeting notes and follow-ups. Stay away from the judgement they are known for. Pick that first and you have made the test about their expertise, and you will lose it. Agreeing openly on what should never be automated buys more trust than any amount of reassurance.

Bring figures they can audit: the firm's own hours, fee earners affected, charge-out rate and 46 working weeks. The unbillable hours calculator does the arithmetic and shows every assumption, which matters, because a sceptic goes straight to the assumptions.

What if they still say no?

Some will, and that is a legitimate outcome. A partner who does not want the rebuilt process in their own team can sit it out, provided the firm is clear about who owns the decision and the people using it are not made to run two versions of the same job.

What you cannot allow is a refusal that drives the behaviour underground. If the process is blocked officially while the team quietly pastes client material into whatever they can find, the firm has no control and no benefit. That is the pattern described in shadow AI, and it appears where the formal answer has been no for too long.

Positions usually move, but not because of an argument. They move when a colleague's work starts leaving the building faster, with the same signature on it and no complaints. Evidence from inside the firm beats everything else. The quickest route there is to rebuild the one process costing the most. The audit will tell you which one that is, and it gives the sceptic something they can mark for themselves.