An NHS team buying clinical AI is purchasing a change to work. It needs to know what will happen to its appointments, documentation, decisions and staff time. A large international audience can make a supplier worth examining. It cannot answer those local operational questions.
The persuasive slide usually travels more easily than the benefit. A product may have an impressive footprint elsewhere while arriving in a service with different records, referral routes, staffing and responsibilities. The relevant question is whether the evidence follows the product into that setting.
NHS buyers should be able to say this directly: show us the improvement we are purchasing, in a workflow sufficiently close to ours. Global popularity should help a company get a hearing. It should not decide the procurement.
Start with the work that needs to change
The purchasing conversation should begin with a defined problem. A practice might need referral letters completed sooner. A department might need shorter delays before notes are available. A clinician might need faster access to an applicable recommendation.
These sound similar in a general presentation about productivity. They have different endpoints. Generating a draft faster may do little for a referral delayed by missing information or unclear responsibility. A quicker answer may add work if the sources are difficult to check.
The buyer should describe the current process before describing the desired software. Who starts the task? What makes it complete? Where does it wait? Who checks it? This creates a fair comparison and prevents a product demonstration from quietly redefining the problem around its strongest feature.
It also leaves room for simpler solutions. A revised template, clearer ownership or a better route to an existing reference may solve part of the problem. AI should compete with those options as well as with other AI.
A safety pack and a benefit claim answer different questions
The current NHS Digital Technology Assessment Criteria guidance covers clinical safety, data protection, security, interoperability, usability and accessibility. It also states that DTAC works alongside other relevant approvals. Those checks matter; they are not a calculation of the benefit a particular service will receive.
NHS DCB0160 guidance addresses organisations deploying and using health IT, alongside the manufacturer-focused DCB0129 standard. The distinction recognises that a product and its deployment create related but different responsibilities.
For buyers, this means two conversations must stay visible. Is the proposed configuration appropriately assured for use? Does it produce enough useful improvement to justify its cost and disruption? A strong answer to one should not be used to end the other.
Ask which differences could change the result
A local evaluation should investigate meaningful differences, not repeat every experiment from scratch. If a supplier has already demonstrated a bounded capability under comparable conditions, that evidence deserves weight. The buyer's job is to identify what could stop the result transferring.
For an ambient tool, that might include the consultation format, multiple speakers, recording conditions or the way staff finalise notes. For clinical search, it might include the guideline jurisdiction, availability of investigations or the organisation's referral criteria. For automation, the critical difference may be who owns an exception when the system cannot finish.
These are hypotheses to examine in the proposed workflow. The number of hospitals in a supplier's announcement does not identify which of them resemble the buyer's service. A short description of one comparable deployment can be more useful than an impressive count of unspecified sites.
The supplier should explain what it expects to transfer and why. That is a stronger sales argument than assuming that geographical scale makes local questions parochial.
Measure the finished task and include ordinary users
A demonstration is usually organised around successful use. Procurement needs to understand ordinary use, including interruptions, incomplete information and staff who are still learning the system.
An illustrative referral pilot could follow the task from the clinical decision to a reviewed, complete referral ready for the receiving service. It would record drafting time, corrections, missing information and any work returned to the practice. The important unit would be the completed referral, not the number of generated paragraphs.
The participating group should reflect the people expected to use the tool after purchase. If the benefit depends on exceptional prompting skill or a clinician spending substantial time customising templates, the implementation plan needs to include that work. Averaging it away does not make it disappear.
Pre-agreed review criteria also matter. The team should know what counts as a material omission or an unresolved task before it sees the results. Otherwise enthusiasm for the project can shape the definition of success after the fact.
Convert saved time into something the service can use
NICE's evidence standards framework considers evidence of economic impact as well as performance relevant to purpose. For a local buyer, the economic question is what usable benefit remains after implementation and ongoing work are included.
Suppose, illustratively, that an evaluation identifies several short reductions in work throughout a session. Those minutes might allow a clinician to finish on time, spend longer with patients or complete other necessary tasks. They should not automatically be converted into additional bookable appointments unless the timetable and staffing make that possible.
Different benefits deserve different descriptions. Less work after the clinic may improve the working day without immediately reducing expenditure. Faster turnaround may help patients even if staff hours are unchanged. Additional activity may be valuable but can also create downstream demand.
The business case should specify which benefit is expected, who receives it and how it will be observed. Licence fees are only one part of the cost; training, support, integration and continued checking belong in the same account.
Set the decision before starting the pilot
Local evidence should lead somewhere. A pilot without a decision rule can become a permanent performance of innovation, consuming time while neither supporting adoption nor ending the experiment.
A useful proposal would specify the workflow, the evaluation period, the outcomes that matter and the conditions for expansion. It would identify the people who can decide to continue, modify or stop. The thresholds should match the function's consequences and the service's actual priorities.
That does not mean demanding perfection. It means deciding what level of improvement and residual work would justify the purchase. A narrow success can support a narrow deployment. A negative result can identify a problem worth fixing. Both are more useful than a general statement that staff found the technology exciting.
The product version and configuration should travel with the findings. When a material change arrives, the organisation can then decide what needs checking again instead of treating every update as either trivial or a complete restart.
Share evidence so every practice does not repeat the same work
There is a legitimate frustration behind suppliers' complaints about repeated local assessment. Buyers should not demand identical documents in incompatible formats or ignore relevant findings from comparable services. Better shared evidence can make deployment faster and the purchasing decision stronger.
The answer is a reusable account of the evaluated task, context, results and implementation requirements. Local teams can add the differences that matter. That approach gives successful suppliers a way to accumulate useful evidence while preserving the buyer's responsibility for its own workflow.
The same questions apply to a source-linked reference tool such as Ask-iatroX. Its published methodology provides something concrete to examine; local usefulness still depends on the task and the effort needed to check and apply an answer.
NHS procurement should reward companies that make their results transferable and inspectable. The most persuasive supplier is the one that can explain what will improve here, what the team must do to obtain that improvement and how both sides will know it happened.
Frequently asked questions
Should NHS buyers ignore international adoption figures?
No. They can indicate experience, distribution and the opportunity to learn from other deployments. Buyers should then ask which settings are comparable and what measured results are available for the proposed local use.
Does meeting DTAC establish cost effectiveness?
DTAC addresses core assurance areas for digital health technology. A local economic assessment still needs to examine the useful benefits, implementation requirements and ongoing costs of the particular deployment.
Does every NHS service need a completely new clinical AI trial?
No. Relevant evidence should be reused where it transfers. Local evaluation should concentrate on differences that could change performance, safety or operational benefit, with a clear decision about what the findings will support.
Explore how iatroX approaches clinical reference and learning ->