AI use-case prioritisation worksheet
Score up to three operational AI use cases for a UK dental practice on value, risk, data, handover and measurement. Essentials such as keeping clinical judgement with clinicians are flagged automatically.
How to use it
Name up to three tasks you are thinking of handing to AI, such as "out-of-hours enquiry capture", "appointment reminders" or "website questions". Answer each question honestly, and leave anything you have not checked as "Not asked yet". Start with the use case that scores highest and has no essentials missing. The reasoning behind each question is in the AI operations guide. Nothing you enter leaves your browser.
What each group checks
- Value. Whether the task is frequent and repetitive enough to be worth automating, and whether it covers a real gap, such as lunchtime or evenings, rather than work your team already does well.
- Risk. Whether the task can be done without clinical judgement, whether urgent messages will reach a person, how much harm a wrong answer could cause, and whether patients are told they are talking to an AI.
- Data. Whether the task works with the minimum personal data, and whether the supplier will sign the contract the law requires.
- Handover. Whether a person is always within reach, and whether someone named picks up each handover in good time.
- Measurement. Whether you can measure the task before launch and review the conversations after it.
A use case that scores well on value but badly on risk is usually one to redesign rather than drop: narrow the scope until the risky part is handed to a person. The rules for doing that are in human handover and booking-rule design.
Worksheet
Value
Is this a task the team repeats many times a week, with much the same answer each time?
Opening hours, parking, how to register, how to change an appointment. Repetitive, rule-based work suits automation best.
Would it help at the times reception cannot answer: busy periods, lunch or out of hours?
Value comes from covering gaps, not from replacing the conversations your team already handles well.
Risk
Does the use case keep clinical judgement with clinicians: no diagnosis, no suitability, no treatment advice? Deal-breaker
If the task cannot be done without a clinical opinion, it is not an operational use case.
Can urgent or worrying messages be recognised and passed to a person straight away? Deal-breaker
Pain, swelling, bleeding, injury or distress always go to a person, in hours and out of hours.
If the tool gets this wrong, is the harm small and easy to put right?
A wrong car park answer is easy to fix. A wrongly cancelled treatment appointment is not.
Will the tool say it is an AI at the start of every conversation?
Patients should never have to guess whether they are talking to a person.
Data
Can it work with the minimum personal data, without medical history or clinical notes?
Name, contact details and the reason for contact are usually enough for operational tasks.
Will the supplier sign a written processor contract covering instructions, security, sub-processors and deletion? Deal-breaker
UK GDPR Article 28 requires a binding contract when a supplier processes personal data for the practice.
Handover
Can patients always reach a person, and is that offered without them having to insist? Deal-breaker
Asking for a person should work at any point in the conversation, first time.
Is a named person responsible for picking up every handover, with a set time to respond?
A handover nobody owns is a missed enquiry with extra steps.
Measurement
Can you measure this task today, before launch, so there is a baseline to compare against?
Without a before figure, nobody can say whether the tool helped.
Will every conversation and handover be logged so you can review a sample each week?
Logs are how errors are found and fixed, and how you answer a patient who queries what was said.
Results
- Use case A
- No answers yet0 of 12 questions answered. Score = weighted points / maximum points for the questions answered.
- Use case B
- No answers yet0 of 12 questions answered. Score = weighted points / maximum points for the questions answered.
- Use case C
- No answers yet0 of 12 questions answered. Score = weighted points / maximum points for the questions answered.
A score only compares the answers you entered. Treat any deal-breaker as decisive, and do not compare two columns until you have asked both the same questions.
Why these are essentials
- Clinical judgement stays with clinicians. Valid consent depends on a dental professional explaining the relevant options and possible costs before treatment starts [3]. A task that needs that conversation is not one to automate.
- Urgent messages reach a person. A tool that cannot recognise and pass on pain or distress should not be answering patients.
- A processor contract. UK GDPR requires a binding contract when a supplier processes personal data on the practice's behalf [2].
- Patients can always reach a person. Where a significant decision is made solely by automated means, UK GDPR safeguards include enabling the person to obtain human intervention [1]. I treat reaching a person as essential for every use case, whether or not that article applies.
For the use case you choose, run the trial in the AI receptionist evaluation checklist before launch.
Sources
- legislation.gov.uk, UK GDPR, Article 22C: Safeguards for automated decision-making. Accessed 1 November 2026.
- legislation.gov.uk, UK GDPR, Article 28: Processor. Accessed 1 November 2026.
- General Dental Council, Standards for the Dental Team, Principle 3: Obtain valid consent. Accessed 1 November 2026.