The most successful AI project a consultancy delivers this year could be the one that costs it most. If the firm builds its judgement into a system the client owns, on the usual terms, it has been paid once for something the client will use for years without it. The better the system works, the less reason the client has to call.

Expertise is becoming something that runs

For most of consulting's history, expertise left the firm as advice: a report, a recommendation, a model handed over at the end. The client applied it occasionally and called again when circumstances changed. Each call was the next fee.

AI changes that. Judgement can now be built into a system that applies it every day.

• An actuarial firm's view of risk can sit inside an insurer's quoting engine, pricing every policy.
• A reward consultancy's job architecture can set the pay range for each new hire without anyone consulting the consultants.
• A supply-chain adviser's planning rules can run the replenishment decisions it used to review once a year.
• A customer-analytics firm's targeting logic can choose which customers receive which offer, every week.

In each case, the expertise that used to justify the next assignment is now working without the firm.

The value goes to whoever owns the running system

Once expertise runs as a system, it can belong to the client, a technology partner or the consultancy. Most firms have not decided which they intend to be, and the default does not favour them.

The technology companies have decided. OpenAI and Anthropic have each set up businesses, with heavyweight investors behind them, that build AI systems directly inside client organisations. Several of the largest consultancies, McKinsey and Bain among them, have joined as partners rather than compete. Mid-sized and specialist firms face the same choice: build and operate the capability, partner with someone who will, or supply the expertise and accept that someone else keeps the system.

Any of these can be a good business. They are worth very different amounts.

The contract is now the asset

What a firm keeps is decided less by the quality of its work than by a handful of terms, often settled from a client's standard template:

• who owns the decision logic once it is built
• whether the firm may reuse what it learned, suitably anonymised, with the next client
• whether it can see the outcomes: which prices held, which hires stayed, which offers converted
• who is allowed to change the logic when conditions move.

Sight of outcomes matters most and is most often missing. A firm that never learns whether its system's decisions worked has nothing to improve, so each engagement starts where the last one did. A folder of confidential client work is not an engine that gets better with use.

Most consulting contracts let the firm keep the methods it brought to the work. The trouble is that the logic built into a client's system is new work, created on the project, so it usually belongs to the client, even though it embodies the firm's method. Few contracts say anything about outcome data. Signed unchanged on an AI engagement, they hand the firm's method to the client in working form.

What survives is the part someone is accountable for

Even well-protected logic has a limited life. Expertise that can be written down will, in time, be absorbed into general AI models, and a method that is distinctive this year may be reproducible in two. What resists this is judgement that someone answers for: the actuary who signs the valuation, the adviser who stands behind a recommendation to a board and is accountable when it goes wrong. Clients will keep paying for a name on the decision long after they stop paying for the analysis behind it. That kind of judgement has to be grown in people, and AI is making that harder, as we argue in "AI is removing the work that built your firm’s judgement".

The questions we would ask

A consulting firm's AI capability is worth what the firm can keep. Before backing one, we would read its client contracts before its case studies, and want answers to five questions:

• Who owns the logic in the systems the firm has built, and who can change it?
• Can the firm see whether those systems' decisions worked?
• What can it lawfully take into its next engagement?
• Which revenue ends with the project, and which continues because the firm still runs or improves the system?
• Which part of its expertise carries accountability that a model cannot replace?

A consultancy with AI on its slides is still a consultancy, and should not be priced like a software company unless its answers justify it.

For leaders, the decision comes before the sale. Choose the role the firm intends to play, then negotiate the terms that support it: rights to reuse, sight of outcomes, a continuing part in updating the logic. A good partnership may create more value than an expensive attempt to become a software company. Expect resistance. Clients' procurement and legal teams increasingly insist that nothing from an engagement is reused or used to train anything. Firms that simply ask for rights will be refused. Firms that trade for them will often succeed: a lower fee in exchange for anonymised reuse or sight of outcomes, or an offer to run the system for the client rather than hand it over.