Costs can fall on the purchasing organisation, staff who check and repair outputs, and patients who must repeat work. The subscription is only one part: implementation, verification, correction and maintenance also consume resources. A low model-inference cost can coexist with an expensive service, while a higher licence price may be reasonable when it reliably removes other work.
The useful unit is cost per appropriately completed workflow, measured against the real alternative. Counting generated paragraphs or multiplying an advertised time saving by every appointment will not establish that an organisation has saved money.
Start with the work being replaced
Define the task and its endpoint before constructing the business case. "Produce a draft letter" is different from "complete and communicate an accurate letter through the correct pathway". An evaluation that stops at the draft excludes review, filing, transmission and subsequent correction.
Describe current practice honestly. It may already involve templates, administrative support, delayed documentation and occasional errors. Do not compare an assisted workflow with a deliberately inefficient version of the existing process.
Then identify the new steps. Staff may need to check outputs, resolve ambiguous status information, manage exceptions or review updates. Some of that work may be offset by better drafts or fewer clarifications; some may be genuinely additional. The model should allow either outcome.
This is an economic evaluation framework, not a claim that a particular supplier creates hidden costs or fails to save time.
Separate implementation, routine use and continuing assurance
Implementation can involve configuration, contracts, privacy assessment, local clinical-safety work, staff training and integration testing. These activities should not vanish from the calculation because different teams undertake them.
Routine use includes both successful work and exceptions. Count review time, editing, abandoned outputs, repeated tasks and support contacts. Consider patients' time where the tool changes what they are asked to do.
Continuing assurance includes monitoring, regression testing after consequential changes, retraining staff and maintaining a viable alternative. The DCB0160 deployment standard, checked on 10 October 2026, concerns deployment and use rather than treating safety as a one-off supplier document.
How these costs are allocated depends on the organisation. A small practice, a hospital and a shared service may distribute them differently. Reusing a common assessment can reduce duplication, but it does not make local workflow differences disappear.
A hypothetical monthly calculation
The following model was constructed on 10 October 2026 for illustration. Every workload, duration and staff-cost assumption is hypothetical; none represents measured supplier performance.
Assume 1,000 notes per month, with current documentation taking four minutes per note. That is 4,000 staff minutes. Value staff time at an illustrative £60 per hour, so the baseline staff-time value is £4,000.
Now assume an AI-assisted process requires 2.2 minutes of review and 0.8 minutes of correction per note, plus 400 minutes of allocated support and assurance work per month. The total is 3,400 minutes, with a staff-time value of £3,400. Add an illustrative £300 monthly licence cost and the assisted total becomes £3,700.
| Illustrative monthly component | Current process | Assisted process |
|---|---|---|
| Documentation or review and correction | 4,000 minutes | 3,000 minutes |
| Allocated additional support and assurance | Included in baseline assumption | 400 minutes |
| Total staff-time value at assumed £60 per hour | £4,000 | £3,400 |
| Additional licence charge | £0 | £300 |
| Total modelled cost | £4,000 | £3,700 |
Under those assumptions, the assisted workflow releases ten staff hours and has a £300 lower combined modelled cost. That is not automatically £300 removed from payroll. The staff may use the capacity for other work while the organisation pays an additional cash licence fee.
Sensitivity matters more than an attractive central estimate
Change just one assumption in the example: correction takes 1.5 rather than 0.8 minutes per note. The assisted workload becomes 4,100 minutes. At the same illustrative staff-time valuation and licence price, the total becomes £4,400, compared with the £4,000 baseline.
The lesson is not that this is the likely outcome. It is that a plausible business case can be sensitive to the effort needed after generation. That effort should be observed rather than assumed away.
A real model should also vary use rates, encounter complexity, staff roles, implementation costs and the frequency of exceptions. It should distinguish fixed costs from costs that grow with activity. If the service becomes easier to access, demand may also change; that can represent beneficial access without being a reduction in total workload.
Report a range and the assumptions driving it. An organisation can then decide what evidence it needs before committing to wider deployment.
Released time, extra capacity and cash savings are different
Time released from typing may improve consultation quality, reduce after-hours work, increase capacity or create room for training. Those are different benefits and should not be counted repeatedly as though the same minute can be spent several times.
A cash saving requires a credible financial mechanism. A service may reduce overtime or avoid a planned expense, but those changes need evidence. Multiplying minutes by an hourly rate estimates the value of time; it does not establish that the budget falls by that amount.
Distribution matters too. A doctor may gain time while a pharmacist, administrator or nurse inherits checking. Evaluate the complete team, including whether people doing the additional work have the authority and information to resolve errors.
The WHO's January 2024 guidance announcement includes wider concerns about affordability, access and dependence. An economic assessment should therefore consider who receives the benefit and who bears the cost, not only aggregate activity.
Apply the same cost discipline to professional learning
This article is published by iatroX and includes its own offering in the discussion. Educational value should be assessed by the learning goal and the methods a person will use, not by the number of unrelated examinations listed in a catalogue.
Under iatroX's UK offer stated for October 2026, Ask-iatroX and free question access are genuinely free, without a trial expiry or verification gate. The paid subscription costs £99 paid upfront for a year, equivalent to £8.25 per month when billed annually, or £29 for a monthly subscription.
The paid bundle includes question banks, Socratic Tutor, the study planner, iatroX Simulations and CPD tools together. Simulations and CPD are included rather than separate paid add-ons. Three monthly payments total £87 and four total £116, so the annual option is cheaper from the fourth month of continuous use at those published prices.
For someone needing only free reference access, a paid bundle may be unnecessary. For a short period of paid learning, monthly payment can cost less overall. For a learner expecting at least four months of relevant use, annual payment has the lower subscription total, provided the upfront commitment is appropriate.
Those are price comparisons, not promises of examination success or clinical productivity. The CPD workflow also requires the learner to review and personalise the record; draft generation does not remove the underlying learning activity.
Build renewal around observed value
Before renewal, review whether the intended users still use the relevant functions, whether the work is appropriately completed and whether the expected benefit appeared. Include correction burden, support effort and the cost of leaving.
For a clinical organisation, the strongest case may be better records and less unfinished work rather than more appointments. For a learner, it may be a coherent practice routine rather than multiple disconnected subscriptions. For an occasional user, the simplest adequate tool may remain the best economic choice.
A sensible purchase is not the cheapest interface or the broadest feature list. It is the option whose complete workflow delivers enough relevant value to justify its total cost, with the assumptions open to scrutiny.
Frequently asked questions
Are AI-generated time savings the same as cash savings?
No: saved time may create capacity or reduce workload without reducing expenditure. Cash savings require an identifiable and evidenced change in the organisation's costs.
What is commonly missed in a healthcare AI business case?
Review, correction, implementation, support, ongoing assurance and downstream work can be omitted. The baseline and assisted workflow should be measured on comparable terms.
When is iatroX's annual subscription cheaper than monthly payment?
At the UK prices stated for October 2026, four monthly payments cost £116 compared with £99 paid upfront annually. Three monthly payments cost £87, so annual payment becomes cheaper from the fourth month of continuous use.
Match your professional learning goal to the relevant iatroX pathway →