Key takeaways
- AI compresses the production of deliverables. It does not produce the insight, and clients will get better at telling the difference.
- The most valuable thing to connect it to is your own body of past engagements, which nobody else has.
- The junior development pipeline is the real strategic risk, and almost nobody is planning for it.
- On a utilisation model, efficiency without a pricing answer reduces revenue. Decide the answer before rollout.
Consulting is unusually exposed to this technology, in a way that cuts both directions. A large part of what a firm produces is written analysis and structured argument, which is exactly what these tools do. That makes the efficiency gain larger than in most industries. It also means that any part of the deliverable which could have been written without meeting the client is now visibly cheap.
Firms that understand the second point tend to get more from the first.
Where the gain is real
The pattern that holds across consulting work is that the tool is strong wherever the firm has already gathered the material and weak wherever it is being asked to supply the substance.
| Task | Fit | Why |
|---|---|---|
| Synthesising twenty stakeholder interviews into themes | Strong | Enormous time sink, material is yours, themes are checkable |
| Turning a partner's verbal hypothesis into a structured storyline | Strong | All the thinking is supplied, the labour is structural |
| Drafting pages from your own analysis and exhibits | Strong | Writing time falls sharply, argument is unchanged |
| Assembling proposals and qualifications from past work | Strong | Highly repetitive, deadline-driven, low judgement |
| Searching your own past engagements for relevant precedent | Strong | The single most underused asset in most firms |
| Rapid orientation in an unfamiliar sector | Good | Useful scaffolding, but verify every figure before it is cited |
| Stress testing an argument before a steering committee | Good | Ask what a sceptical client would challenge, then prepare for it |
| Cleaning and reshaping data for analysis | Good | Effective, but check the transformations rather than trusting them |
| Market sizing and industry figures | Poor | Produces confident numbers with no source you can defend |
| The recommendation itself | Unsuitable | It is what the client is paying for and it needs to be defensible under challenge |
The fifth row is worth dwelling on. Most consulting firms hold years of engagement material that is functionally inaccessible, sitting in folders organised by client and date, findable only by asking whoever happens to remember. Making that searchable in plain language is often the highest value application available to a mid-size firm, and it is one that no competitor can replicate because the material is yours.
Before doing that, resolve the confidentiality question properly. Past engagement material belongs to clients under agreements that usually restrict reuse. A searchable internal knowledge base built on it needs the material to be anonymised, aggregated or covered by the engagement terms. This is a contract review, not a technology decision, and it should happen first.
The commoditisation problem, stated plainly
Take a recent deliverable and ask how much of it could have been produced by a capable person who had never met the client. In most firms the honest answer is somewhere between a quarter and a half: the market context, the framework explanation, the general good practice, the benchmark section, the standard recommendations that appear in every version of this engagement.
That content was always the weakest part of the product. It was tolerated because producing it took real hours and clients understood they were paying for those hours. Once a client can generate a competent version of it in a minute, its presence in a deliverable stops reading as thoroughness and starts reading as padding.
The response is not to hide the AI use. It is to change the mix. Less general context, more client-specific evidence. Fewer frameworks explained, more findings from the actual interviews. Less benchmark, more of what your consultants observed in that particular operation that nobody else would have seen.
Firms that make that shift find their deliverables get shorter, sharper and better received. The efficiency gain funds the change rather than being extracted as margin, which is the trade most partners eventually conclude is correct.
The junior pipeline problem
This is the issue that receives the least attention and has the longest consequences.
A consulting firm develops judgement through an apprenticeship. An analyst builds the model, does the research, drafts the pages and gets corrected. The corrections are the education. Over three or four years of being wrong about things and being told why, a person acquires the pattern recognition that lets them run an engagement.
The work in that apprenticeship is precisely the work AI does well. A firm that automates it fully has removed the mechanism by which its future managers were produced, and will not notice for about five years, at which point the problem is unfixable in the short term.
The firms thinking about this are handling it in a few ways. Some require juniors to produce the first version themselves and use AI only to critique it, which preserves the struggle that generates the learning. Some keep certain engagement types deliberately manual as training grounds. Some have shifted the apprenticeship toward client interaction and interviewing earlier, on the reasoning that the analytical labour is going away but the judgement about people and organisations is not.
All three are reasonable. Doing nothing is not, and doing nothing is the default.
A concrete version worth adopting: for the first year, juniors draft without AI and then use it as a critic, asking what is unclear, what is unsupported and what a client would push back on. They get the speed benefit as a reviewer rather than as an author, and the learning survives.
Utilisation, and what happens to the model
A firm billing by time has the same arithmetic problem as a law firm, and consulting has less protection because clients are more sophisticated buyers and more accustomed to negotiating scope.
The honest positions are the same three. Take the capacity gain and sell more work, which requires demand to exist. Move to fixed fee or outcome pricing on the engagement types where the scope is predictable, which is where the market has been drifting anyway. Or reinvest the time in depth, which improves the work and is the hardest to defend to a finance partner.
What does not work is the passive option, where nothing is decided and consultants quietly work out that using the tool reduces their recorded hours. The result is a firm that has bought licences, run training, and observes almost no usage among the people whose time is most expensive. Partners then conclude the technology is overrated, which is not what happened.
What to configure
- One firm account on a business tier, so client material stops passing through personal logins.
- Training on inputs disabled and retention configured, with evidence kept for client security questionnaires, which in consulting arrive constantly.
- A client permission review, because engagement letters and outside adviser agreements increasingly speak to this.
- A hard spending ceiling, since consultants generate volume and long documents are where usage costs accumulate.
- Task guidance by grade, since what a partner, a manager and an analyst should use it for are three genuinely different lists.
- A written position on the pipeline question, even if it is provisional.
Frequently asked questions
What should a consulting firm use AI for first?
Interview and document synthesis, and proposal assembly. Both are large, repetitive time sinks where the firm already holds the material, errors surface immediately, and the judgement stays with the consultant. Making your own past engagements searchable is often the highest value application, subject to a contract review first.
Does this commoditise our deliverables?
Only the parts that could have been written without the engagement, which was always the weakest content. The response is to reduce generic context and increase client-specific evidence. Firms that do this end up with shorter, sharper deliverables that are better received, funded by the efficiency gain.
What happens to our analysts?
The work they learned from is the work AI does well, which puts the development pipeline at risk in a way that is invisible for several years. Practical responses include having juniors draft first and use AI as a critic, keeping some engagements deliberately manual, or moving client interaction earlier in the apprenticeship.
How do we handle client confidentiality?
A firm-controlled business account with training on inputs disabled and retention configured, plus a review of what your engagement letters and clients' adviser agreements actually permit. Keep the settings evidence to hand, because client security questionnaires in consulting arrive constantly and a screenshot answers most of them.
Get your firm set up without hollowing out the bench
We choose the provider, configure the confidentiality settings, cap the spend, and write task guidance separately for partners, managers and analysts, including how to use it as a critic rather than an author where that matters. Fixed price, live in 30 days or less.