Does AI touch it, and does the value last?

Most companies start their AI work in one of two places: with a list of use cases, or with a governance model. Both are useful. Both skip the question that comes before them: where does the value of the company come from, and what does AI do to it?

Take a company, or the part of it that serves customers, and write down where its value comes from. Then ask two questions of each item. Does AI touch it? And if it does, does the gain stay with the company, or does it fade or leak away?

The grid below places the answers. What sits in the top half keeps its value.

Baseline — where every plan starts

Efficiency, speed, throughput and reach: the work gets done cheaper and faster, at higher volume, or for customer segments that were never economic to serve before. This is where AI changes the most, and where every plan starts, because the metrics are familiar and a pilot shows an effect within weeks.

It is also the box where the gain does not stay. Every competitor can buy the same AI models and platforms, so within a year or two a faster, cheaper service is simply what the market expects. A company that sells this work to its customers by the hour loses the gain even earlier: a faster hour is a cheaper invoice. And where the work runs on a platform that charges per use, part of the saving goes to the platform.

The large offshore IT services companies now expect their prices to fall three to five per cent a year, and their own CFOs say the productivity gains will go to the client. The work got better. The customer got the benefit.

Premium — what you can price

Insight, risk and control, and orchestration all get stronger with AI. But the premium sits in their combination: a company that delivers an outcome to its customer, built on insight only it has, through complex work across systems, departments and partners that it orchestrates, in a controlled way it stands behind.

The clearest example is the new generation of accountancies. They take the client’s books off their hands and run them end to end — bank feeds, invoices, payroll, tax — from the client’s own data, with the firm checking every file and taking responsibility for the result. They charge per finished file, not per hour, and nobody asks whether they used AI. They are often far cheaper than the firms they replace, and still keep the gain, because they own the whole job. The incumbent that adopts the same tools and keeps billing by the hour gets cheaper too, and loses it. A general model can produce the entries; it cannot stand behind the file.

Take one of the three away and the premium goes with it. An online student help desk held only the insight — ten years of expert answers to students’ questions. When general AI models could produce the same answers, nothing else was holding, and it lost more than forty per cent of its revenue.

Anchors — what carries the value

This is the part of the company that AI touches least: the reason a customer chooses this company over another, the relationships that get it the meeting, its licence to operate and the liability that comes with it, its proven ability to change, and its people on site.

AI does reach the surface. A growing share of business buyers now start their shortlist in an AI assistant, and two in three say they would prefer to buy without a salesperson. But the same buyers still check what the assistant tells them with a person, and the proposal — the commercial argument — still decides. The licence and the liability do not move to the AI either. When a result produced by AI is wrong, the company that used the AI is still the one responsible — European law has kept it that way — so the right to operate keeps its value, and so does the responsibility that comes with it.

One thing an anchor cannot do is protect value that belongs in another box. The largest listed IT services companies are now valued at far lower multiples than five years ago. Their relationships did not break — those held, as anchors do. But the revenue those relationships carried was billable hours, and hours sit in the baseline. The relationship kept its value; the hours behind it leaked away.

Quiet erosion — looks untouched, isn’t

Switching costs and brand as a source of traffic both look as if AI leaves them alone.

For a service company, switching costs are the client’s dependency on it — the people who know the account, the processes that exist only in their heads, the months it takes to bring a new provider up to speed — and AI is lowering each of them. A new provider now learns the environment in weeks. Clients know it: contracts are being reopened long before they end, and the price cuts demanded at renewal have roughly doubled. Brand still earns trust, and that trust carries over into AI: the brands mentioned most across the web are the ones AI assistants cite. But the route a brand used to own — the search, the shortlist, the first click — is changing hands. When an AI summary sits above the search results, clicks halve.

The pattern

Look at what holds in each box. It is never the work; AI does the work. What holds is the part that carries a consequence: an outcome someone owns, a result someone stands behind, a customer who knows whom to call when it goes wrong.

Does AI touch it, and does the value last? The value lasts where the company stays responsible for the result. AI can do the work. It cannot take the responsibility.

Where does your company sit?