Why has AI turned up in the pitch at all?

Three forces put it there. Procurement teams have added questions about it to their standard question sets, so it arrives in the pack whether or not anyone on the panel has thought hard about it. Buyers read the same trade press you do and want to know whether they are still paying senior rates for work that now takes minutes. And rival firms have started volunteering it, usually under innovation or added value, which creates pressure to say something back.

So the real question is not whether AI comes up. It is whether you control how it comes up, or whether a partner improvises an answer at interview that nobody in the firm can stand behind afterwards.

What is the buyer actually asking?

One question on the page is usually three questions underneath it.

  • Am I paying for hours that no longer exist? A finance director who hears that your reporting is now automated will reasonably ask why the fee looks the same as last year. That is a pricing conversation, and it is worth having deliberately rather than by accident. We set out the thinking in pricing work that AI makes faster.
  • Where does my material go, and who sees it? This is the one in-house counsel and compliance leads care about. They want to know what happens to confidential documents, what is retained, and whether anything is used beyond their engagement.
  • Who is accountable if it gets something wrong? Panels are not frightened of a drafting system. They are frightened of a report nobody read before it went out.

Answer those three and the AI question is finished. Answer none of them and you have given the panel an excuse to score you down on a section you thought was decoration.

Why does a vague AI claim cost you the pitch?

A line like "we use AI across the practice to deliver faster" sounds safe. It is not. It is an open invitation to a follow-up: which part of the work, on what, who checks it, and what happened the last time it produced something wrong. If the answer collapses within two follow-up questions, you have spent your credibility on a sentence that was never going to win anything.

Overclaiming also creates obligations you did not price. Promise same-day turnaround because a system drafts the first version, and you have just written a service level into the bid. Promise capability you intend to build, and you have made a commitment that falls due in month two of the engagement.

What can you say that is true, specific and safe?

Describe the process, not the technology. A claim that survives scrutiny has four parts: the job, what the system does, what a person does, and what the client receives.

For example: monthly reporting is produced from the records the team already keeps, in the firm's own format, so a draft exists before anyone sits down to write it. The responsible adviser checks it against the file, corrects anything that needs judgement, and signs it off. The client receives the report in the first week of the month rather than the third. That is a specific, checkable statement about client reporting and it says nothing a panel can knock over.

The sentence that does most of the work in any bid is the review sentence. A named person reviews and approves anything that goes to a client. AI drafts; people approve. If your firm has written that rule down and can describe how it is enforced, say so plainly. The guide to reviewing AI output on client work covers what a real review step looks like, as opposed to a tick box.

What should stay out of the bid?

  • Numbers you cannot evidence. Time saved, error rates, productivity gains. If you did not measure it on your own work, do not print it.
  • Tool names used as a badge. Listing a licence tells the buyer what you bought, not what changed. Plenty of them have bought the same licence and watched nothing change, which is the point of a tool is not a process.
  • Future capability described in the present tense. "We are rolling out" and "we are exploring" belong in an internal paper, not a tender response.
  • Anything that contradicts your own policy. If the bid says one thing and your AI policy says another, you have handed the panel a governance problem.
  • Anything your insurer has not heard. Bid text is a representation about how you work. It is worth checking that it matches what you told your broker, which is the argument in telling your PI insurer about AI.

What if the firm does not really use AI yet?

Say so, and then answer the question behind the question. The buyer wants speed, consistency and accountability. You can describe how you get those today: standard formats, a review rule, a named point of contact, a turnaround you commit to. That is a stronger answer than a borrowed claim.

What you should not do is write a roadmap into a bid to fill the box. If you genuinely intend to rebuild a process, pick one, do it, and write about it next time with facts attached. One rebuilt process you can describe in detail beats a page of intent.

How do you make the answer survive the interview?

Write it once and keep it in the bid library, so the same words go out every time and the partner at interview is repeating something approved rather than inventing. One page per rebuilt process: what it does, who reviews, how material is handled, what the client sees. Keep it next to your standard quality and data sections so it is pulled in automatically during proposal and bid production, rather than rewritten by whoever is drafting at eleven at night.

Public sector work and larger private frameworks will ask in a more formal way, sometimes alongside questions on data handling and subprocessing. Keep the approved wording aligned with how you answer those, so your tender responses do not say two different things in two different sections. The guide to data protection in AI client work covers the ground those questions sit on.

Then rehearse the follow-ups. Which process. Who signs. What happens when it is wrong. A partner who can answer those three calmly turns a scoring risk into a point of difference.

So what is the honest position?

Mention it when it is true, specific and already running. Stay quiet when it is not. The firms that answer this question well are not the ones with the most licences, they are the ones that have rebuilt a job and can describe it in a paragraph without flinching. That also makes the conversation with existing clients easier, which is the subject of telling clients you use AI.

If you want something concrete to say in the next bid, start by finding the process worth rebuilding first. The audit takes about three minutes and names one.