Service mandate and decision rights · Lesson 1

Define a digital health service mandate and its decision boundary

6 min read

Lesson objectives

  • Frame a digital health service decision around users, workflow and harm boundaries.
Reading tools

Before a governance group decides whether a digital scheduling and follow-up service works, it must say what the service does. A reminder, an appointment change and a message requiring follow-up have different users, handoffs and potential harms. An operational mandate links a user need to an observable function, assigns the next decision owner, and states what the software cannot decide. More clicks do not establish better access when people with urgent needs disappear from the pathway.

Trace one journey from a request to its closure. Record the entry channel, identity check, minimum required data, the action visible to the user, and the owner who takes over when the step fails. Separate use from completion: opening a link is not completing a request. Define the denominator before calculating a rate. Repeated requests may come from one person, so a request completion rate cannot be reported as a patient completion rate. Separate an operational failure, such as a broken link, from a potential care-safety failure, such as a time-sensitive message with no accountable receiver. The latter needs a specified human escalation under the institution's own policy. The digital service does not make a clinical judgment.

The mandate separates four decisions. The service owner decides whether non-urgent rescheduling addresses a real access problem. The data steward determines what information is needed and how retention is governed under the applicable local rules, which must be verified before implementation. The technical owner tests delivery and records failures. The relevant clinical team retains judgment about signals that may call for care. Separation does not mean isolated decisions: expanding a feature requires evidence across these owners. When an operations report says the service has 'reached patients', first ask whether it counted unique people, sessions or requests. A comparison across weeks is unreliable if the indicator definition changed. Governance should fix the definition, data owner and validation step before attaching a success threshold.

A committee can test the boundary in three layers. First, does the service possess inputs sufficient for the claimed action? A scheduling form knows a booking number and date, but may know nothing about clinical condition. Second, is there a verifiable output? A screen saying 'sent' shows the system attempted delivery, not that an accountable receiver accepted the message. Third, what follows if this step fails? Delayed non-urgent rescheduling differs from a message flagged as time-sensitive with no receiver. Those layers explain why a bounded pilot can continue for one function while expansion of another remains on hold. They do not declare that a particular message is medically urgent; they define an operational boundary requiring a handoff to the qualified team under institutional procedure. Test the alternative pathway both when connectivity fails and when identity does not match, recording what the user sees and who takes over.

After the provisional decision, specify evidence that could disprove the working explanation. If the team thinks name entry causes identity mismatch, review a suitably minimized sample of failed cases and compare the system reason code with what support staff actually found. If a 'success' message was delivered despite no confirmed appointment, the completion indicator itself needs correction. If no receipt time is logged for follow-up messages, do not infer it from dispatch time. Record receipt as unverified and establish a way to demonstrate it before claiming safety. The decision record names the assumption, test, evidence owner and review date, so the pilot can be stopped or redesigned when observations contradict the prediction.

The choice is broader than 'launch' or 'cancel'. Continuing every function could spread access rapidly, but would tolerate a messaging hazard with no verified receiver. The committee rejects it on the present evidence. Stopping everything would remove that digital hazard, yet it would also deny users a non-urgent rescheduling pathway that has completed many requests, without fixing the failure modes. A third option separates the functions: continue a bounded rescheduling pilot with an alternative channel, and hold messaging until receipt is demonstrated. That option permits learning while preserving a clear harm boundary. It is not proof of effectiveness. A later cycle needs the same denominator definition, complaint review, and an examination of who cannot use the digital channel. If failed requests simply move to telephone support and are not logged there, the digital completion rate can look better while the underlying burden persists. The review should therefore consider both channels, including staff workload and whether users can actually finish the task.

Of 500 requests, 410 completed and 90 did not. Causes sum to 35+25+30=90. Completion uses all requests: 410÷500=82%. Identity mismatch uses 35÷90≈38.9% when describing incomplete requests, or 35÷500=7% when describing all requests.

The request completion rate is 82%. Test identity, connection and form changes in a non-urgent pilot; promptly escalate the time-sensitive report to a named clinical owner and do not expand messaging until receipt is verified.

Requests may recur for one person. Response times, health outcomes and verified receipt of the flagged report are absent. The completion rate is therefore neither a unique-person access rate nor evidence of clinical safety.

If 460 of 500 requests completed while the time-sensitive report still lacked an owner, completion would rise to 92%, but containment and the hold on expanding messaging would remain necessary. Improving the completion metric does not resolve the receipt hazard.

Lesson summary

Of 500 requests, 410 completed and 90 did not. Causes sum to 35+25+30=90. Completion uses all requests: 410÷500=82%. Identity mismatch uses 35÷90≈38.9% when describing incomplete requests, or 35÷500=7% when describing all requests. The request completion rate is 82%. Test identity, connection and form changes in a non-urgent pilot; promptly escalate the time-sensitive report to a named clinical owner and do not expand messaging until receipt is verified. Requests may recur for one person. Response times, health outcomes and verified receipt of the flagged report are absent. The completion rate is therefore neither a unique-person access rate nor evidence of clinical safety. If 460 of 500 requests completed while the time-sensitive report still lacked an owner, completion would rise to 92%, but containment and the hold on expanding messaging would remain necessary. Improving the completion metric does not resolve the receipt hazard.