Paying for AI by the Result in Pharma and Life Sciences

The approver keeps the decision

Most AI pricing conversations in pharma stall on one objection: a person has to sign off anyway, so what exactly are we paying the software for? It is a fair question. Promotional material still goes through medical, legal and regulatory review, and a deviation report still needs a named owner in quality. No contract should pretend otherwise.

Per-result pricing works here once you move the result one step earlier. The result is the work handed to the approver, and it only counts when the approver accepts it without rework. The decision stays with the person. What the client pays for is a draft good enough that the reviewer spends their time on judgment instead of assembly.

What a result looks like in practice

For promotional review, one result is an asset with every claim matched to an approved reference in the claims library, or flagged with the reason it could not be matched, and with the required safety information checked. In quality, it might be a deviation report drafted from the batch record with the evidence attached and the root cause section left for the investigator.

Each definition also names what is out of scope. The final sign-off is always out of scope. So is a new product launch whose claims library is still being written, because the agent has nothing approved to check against yet.

Checks a reviewer can trust quickly

The checks matter more in regulated work because the reviewer has to accept or reject each flag in seconds. Every flag therefore cites the rule or reference it came from. We built a scanner for a European pharmaceutical regulator that reviews advertising against health-claims rules, and review time per asset went from two or three hours to about two minutes. Time came down because each finding pointed at the rule it broke, so the reviewer was confirming rather than searching.

Every result is logged with its input, the checks it passed, the approver, the time taken and the model and prompt version that produced it. In a GxP setting that log is part of the audit trail, and in a per-result contract it is also the basis of the bill.

Change control belongs in the contract

A model update can move the pass rate, and in a per-result contract a moving pass rate moves the bill. Systems that support GxP work should be built to the company's validation expectations, so the contract should say who approves a model or prompt change and how it is tested before it reaches production. We run the acceptance checks against the new version on a held-back sample and switch only when it passes. This slows model upgrades down a little, which in pharma is usually what the quality team wants anyway.

Which workflows suit per-result pricing

The best candidates have steady volume, a reviewer who already exists and a result that can be described in one sentence. Promotional material review fits well, because every asset already passes through a committee and the claims library gives the agent something approved to check against. Medical information responses drafted from approved standard response documents fit for the same reason: the source is controlled, and a medical information specialist signs off on each one.

Literature screening for safety monitoring can work too, with one result defined as an abstract screened and classified with its reasons, ready for a safety scientist to confirm. The volume is high and the decision rules are written down. Work that depends on scientific judgment from the first line, such as a clinical study report narrative or a regulatory strategy memo, is a poor fit, because nobody can write a check that says whether it is good.

Pricing a result when review time is long

In regulated work the reviewer's time does not disappear, so the saving comes from how much shorter the review becomes. Measure it during the acceptance test by timing the same reviewers on a sample with and without the agent's draft. If a reviewer takes 40 minutes to check an asset from scratch and 12 minutes to confirm a cited draft, the 28 minutes in between is the value a price per result has to share.

Price from that difference and the loaded cost of the reviewer's time, not from the volume alone. Reviewers in pharma are expensive, so even modest volumes can justify a build, and the price per accepted draft is usually higher than in back-office work. That is fine as long as it sits well under what the shorter review saves, and the calculator on our Results as a Service page will run the sums on your own numbers.

Why pharma buyers end up preferring it

Pharma budgets are set a year ahead and procurement is slow, so a variable bill can look like a problem at first. In our experience the per-result model is still easier to defend to finance than an hourly statement of work that grows when a project slips, because the price per accepted result is fixed and the monthly minimum is known in advance. Several of our pharma and life sciences clients pay us on results-based terms. We do not name them, which is normal in this industry.

Our Results as a Service page sets out the terms we use after launch, and the pharma and life sciences page shows the kind of work we build.