The licence fee is the number in the proposal, and in a regulated company it is frequently not the largest cost of the project. That is not a criticism of vendors — it is a structural feature of putting any system into a GxP environment.
Here is the honest cost structure, so your business case survives contact with year two.
The costs people budget for
- Licence or subscription, usually per user or per volume.
- Implementation and configuration.
- Integration with the systems it must read from or write to — routinely underestimated, and where timelines actually go.
- Initial training.
The costs that decide the business case
Validation. The package is yours regardless of vendor evidence: intended use, risk assessment, test set with agreed ground truth, execution, report. Building an honest test set is usually the largest single line, because establishing the correct answer for a few hundred real cases takes qualified people — see building the test set.
Review time that never goes away. The most commonly ignored cost. If a qualified person must review every output, that time is permanent. A tool that drafts in seconds but takes twenty minutes to verify has not saved twenty minutes — it has moved the work. Measure verification time in the pilot, not generation time.
Revalidation on model changes. Every version bump is a change with an impact assessment and, often, a rerun of the test set. With a vendor shipping quarterly, that is a recurring annual cost — see change control and revalidation.
Monitoring. Performance against thresholds, periodic review, someone looking at the numbers. Small per month, permanent.
SOP changes and retraining — procedures revised, read-and-understood records refreshed, competency evidenced.
Supplier management. Qualification up front, then periodic reassessment, audit rights, subprocessor tracking.
Exit. Retention obligations outlive the contract. Getting records out in a readable form is a real cost, usually discovered late — see decommissioning.
How to build a case that survives
Model five years, not one. The first year flatters any tool because validation is capitalised in enthusiasm and the recurring costs have not arrived yet.
State the review time explicitly as an ongoing cost rather than hiding it in "business as usual". A case that pretends review is free will be dismantled by the first quality reviewer who reads it — and they will be right.
Compare against the real alternative, which is rarely "do nothing". It is usually contractors, overtime, or a programme slipping. Costing the status quo honestly is what makes the comparison credible — see the real cost of manual validation.
And be straight about what you do not know. A range with stated assumptions beats a single number nobody believes. The returns side is covered in GxP AI software ROI; this is deliberately the cost side of the same page.
Where to go next
Explore GxP Copilot for AI-native validation, TraceDraft for source-traceable clinical documentation, or book a demo to see either on your own data.