For a while the AI conversation was about adoption. Should we experiment, where might it apply, which tools should we back. That phase closed some time in the last year.
The question now is simpler and considerably more dangerous: where is the return, and who is accountable for it?
The research points the same way from both directions. Vendors report that a growing share of organisations have realised a return on AI-driven productivity, and that many more expect to inside a year. The same reports then explain that value shows up when organisations redesign operating models, rework value streams, introduce governance and realign teams around the technology.
Those two claims sit awkwardly together. If the value were inherent, none of that redesign would be needed.
You would deploy, bank the gain, and move on. The fact that so much structural change is required tells you the value is not in the tool. It is in what the business does differently once the tool is there. That is a commercial architecture question, and it falls well outside the remit of whoever ran the pilot.
Visible productivity invites scrutiny
Productivity is the proof point everyone reaches for, and for good reason: it is one of the few outcomes that can be pointed at across functions. Time saved. Faster processing. Fewer manual steps. Higher throughput.
The trouble starts when those gains become measurable. Once a saving is visible, it stops being an abstract benefit and becomes an economic signal, and economic signals invite questions that were previously easy to avoid.
If time is saved, where does it go. If cost is reduced, who captures the benefit. If the work is faster, why is the price the same. AI did not create those questions. It made them unavoidable.
The risk has stopped being one-sided
Software and services were sold on a quiet assumption: the customer carried the risk. You paid for access, you adopted the tool, and if the value stayed theoretical that was framed as an adoption problem, a change management problem, or a mismatch of expectations.
When a vendor claims AI will materially improve productivity or decision quality, and those outcomes are measurable, the buyer asks who stands behind them. That is a reasonable question and most commercial models have no answer prepared.
Watch where it surfaces. Security reviews run longer, because the buyer is delegating action to a system rather than storing data in one. Pilots multiply, because stakeholders want evidence the result repeats. Procurement introduces prove-it clauses and service credits. Expansion stalls, because the first deployment never established a baseline and nobody can defend the next budget request.
Why the pricing model starts to fracture
Subscription pricing assumes predictable margin and low marginal cost. Once the software exists, selling more of it costs little more to deliver.
Inference, orchestration, data access and continuous agentic workflows break that. Cost now scales with usage rather than with seats, so the more value a customer draws, the more it costs to serve them. Hold the price fixed while the cost floats and margin erodes. Float the price and the customer wants transparency and proof.
Subscriptions also assume value is durable even when it is loosely measured. AI removes that comfort by making value easier to observe and easier to compare. Buyers are sceptical because they want evidence before they pay for potential, which is a shift already visible in how buyers negotiate.
What a value economy actually demands
The direction of travel is toward an economy where value has to be stated rather than implied. Four pressures arrive together, and AI accelerates all four.
1. Identifiable
Named, specific, and attached to a business outcome somebody owns. If productivity, cost, revenue and experience are all being claimed at once, that is a narrative rather than a value definition. Pick one primary outcome per initiative and write down what changes, for whom, by how much, and over what period.
2. Measurable
Baselines locked before anything scales, attribution rules agreed, and a shared view of what counts as success against noise. Reporting a gain without a baseline works once. It survives neither a CFO nor a procurement review.
3. Visible to the buyer
Value the customer can see for themselves, in their own numbers, rather than value asserted in a quarterly review. If you cannot show it in their language, you cannot price against it with any confidence.
4. Priced with accountability
Someone owns the economic result end to end. Not adoption, not enablement, not experimentation. The number itself.
The choice in front of leaders
The decision is not which tools to adopt. It is whether the business is willing to operate where what it charges for is stated plainly, and that requires uncomfortable changes.
Measurement stops being a reporting exercise and becomes a commercial capability. Finance moves from cost control toward unit economics. Sales moves from persuasion toward proof. Customer success becomes accountable for outcomes rather than activity. And leadership has to decide who owns value delivery from one end to the other.
Stress-test the model honestly. What happens when a customer asks for proof before renewal. What happens when productivity gains are benchmarked across vendors. What happens if usage grows faster than revenue. If the answer only works while value stays fuzzy, the model is already fragile.
The race is on, but it is not a race to deploy. It is a race to credible value creation. AI makes value visible, visibility forces accountability, and accountability reshapes pricing, contracts and operating models in that order.