At 8am, the practice phone starts ringing before the first consultation has begun. Online requests accumulate, reception tries to establish who needs what, and the duty GP works through a queue that keeps growing while morning surgery is already under pressure. Digitising demand may change the queue's shape, but it doesn't automatically remove the triage work.
That is why the useful question isn't whether autonomous triage is faster. It's how much time does autonomous triage save a practice, which staff recover it, and does the work disappear or move somewhere else? The answer depends on the model being replaced, the practice's existing workflow, and whether the system completes triage and booking or only assists a human decision-maker.
This distinction matters in NHS primary care. A practice can reduce GP contact time without reducing total clinician workload, as the ESTEEM trial analysis of telephone triage demonstrated. It also needs a clear process for uncertainty and escalation.
What Triage Time Savings Really Measure
Time saved is not one number
At 8am, a request can pass through reception, a GP or duty doctor, and the booking team before the patient reaches the appropriate appointment. Reception collects enough information to direct the request. A clinician reviews the details and chooses the route. Administration may then need to contact the patient again and arrange the appointment.
That journey contains several separate workloads:
GP-led triage time, spent reviewing requests and deciding their priority or route.
Administrative booking time, spent matching the patient with an appointment and communicating the available options.
Rework time, caused by an unsuitable first route, appointment type, or incomplete information.
Downstream clinical time, which may stay the same when more patients are successfully directed into consultations.
Keep these categories separate when calculating savings. The 2025 integrative review of UK primary care triage distinguished total process time from GP time. It reported approximately 1.7 minutes longer overall for triage, while saving around 2.45 minutes of GP time. The distinction matters: the GP saving does not represent an equal reduction in total practice workload for every request.
Practical rule: State whether a claimed saving refers to GP time, combined clinical and administrative time, or operational hours across the whole practice.
What “released” really means
A task has been removed only when nobody else in the practice needs to complete it. If a GP stops handling the request but a nurse, receptionist, or another clinician takes over the same decision and follow-up, the workload has moved rather than disappeared.
The UK case evidence from Swanscombe Health Centre illustrates the difference. Its case study reports approximately 422 clinical and administrative hours returned during the first four weeks, alongside approximately 5,068 appointments booked autonomously during that period, as reported by GP Triage's Swanscombe case study. These figures describe work removed from the practice workflow. They do not mean every returned hour becomes extra consultation capacity.
The operational test is practical: map each request from submission to booking, record every staff touchpoint, then measure which steps remain after implementation. A system that recommends a route still leaves the queue with staff. A system that triages and books within defined safety boundaries can remove the manual triage step for eligible requests.
Five Models of Triage and Where Time Actually Disappears
The label “AI triage” covers materially different operating models. A practice should name the model precisely before comparing time savings.
The progression from reception to autonomy
| Triage model | Human touchpoint | Where time goes |
|---|---|---|
| Manual reception triage | Reception collects information and decides the initial route | Phone handling, questioning, callbacks and booking |
| Manual GP-led triage | A GP or duty doctor reviews each request | Clinical review, prioritisation and onward administration |
| Form-based online consultation | The patient submits information, but staff review and route it | Reading forms, deciding urgency and arranging care |
| AI-assisted total triage | AI suggests urgency or routing, but a human confirms every request | Opening the request, checking the recommendation and making the final decision |
| Autonomous AI triage | The system classifies, routes and books within the practice's configured safety model | Human attention is reserved for exceptions, escalations and cases outside the automated pathway |
The important dividing line is between AI-assisted and autonomous triage. AI assistance can make a clinician's review quicker or more consistent, and that can be a sensible improvement. It doesn't, by itself, remove the triage queue, because a human still needs to open, read and decide on each request.
Autonomous triage changes the workflow rather than accelerating it. The system makes the defined routing decision, selects an appropriate booking pathway, and sends the relevant summary into the booking record. The practice still owns governance, safety-netting and exception handling, but routine eligible requests no longer require manual triage.
Where the earlier models still help
Manual reception triage can protect clinicians from unsuitable requests and provide a human response for patients who struggle with digital access. Manual GP-led triage gives a doctor direct control over prioritisation. Form-based consultation can collect useful information before contact, reducing some back-and-forth.
The BMJ Open study of access in English general practice describes how patients can share relevant information before an appointment, enabling the practice to route them to the appropriate appointment and healthcare professional first time. That mechanism can improve workflow even when a person remains responsible for the decision.
The trade-off is structural. The first four models digitise, speed up or redistribute triage. Autonomous AI triage is the model that removes the manual triage step for requests that fall within its configured pathway.
How Autonomous Triage Releases Clinical and Admin Hours
Autonomous triage releases time through a chain of completed actions. Removing only one part of the chain produces a smaller result.
Clinical decision-making
The first mechanism is automated clinical triage. The system gathers the patient's reported information, assesses urgency and routes the request according to the practice's configured safety model. A GP or duty doctor doesn't need to review every routine request to determine where it should go.
That saving is GP-specific time when it removes GP-led triage. It must not be presented as a saving of total practice workload unless administrative and other clinical effects have also been measured. The distinction is important because historical telephone triage evidence shows that reducing GP contact time can still leave total clinician time broadly unchanged, with work redistributed elsewhere in the pathway.
Booking and routing
The second mechanism is autonomous booking. Once the route is determined, the system can match the request to the appropriate appointment pathway rather than sending the patient back to reception for another exchange. This removes administrative handling associated with routine booking, although the exact result depends on appointment configuration and exceptions.
That isn't the same as claiming that every booking becomes a new appointment. A practice may use the released time to manage demand, protect same-day access, support staff resilience or improve oversight. The benefit is the removal of manual processing, not an automatic promise of additional clinical capacity.
Write-back and exception work
The third mechanism is structured write-back. GP Triage states that it pushes a triage summary into the booking record and integrates with the major UK clinical systems. It doesn't read full patient records, create SNOMED coding, or create tasks directly in the practice's clinical system, so the operational claim should remain limited to the supported triage summary and booking workflow.
The hours affected here are administrative, including manual transcription and record preparation. Exception handling remains part of the operating model. High-priority presentations and requests outside the configured pathway need a defined human route, and that oversight should be counted when calculating net savings.
A sound measurement therefore records three separate outputs:
GP hours removed from manual triage.
Administrative hours removed from booking and data handling.
Hours retained for exceptions, governance and oversight.
What the UK Evidence and Case Studies Show So Far
UK evidence does not support one universal figure for time saved. The operational difference lies in what happens to the manual decision step. Traditional or AI-assisted triage may redistribute work between clinicians, reception and administration. Fully autonomous triage can remove that step for requests within the configured pathway, although exceptions and oversight still require staff time.
Swanscombe Health Centre reports approximately 422 or more clinical and administrative hours returned in the first four weeks, with approximately 5,068 appointments booked autonomously during that period. Its case study also describes a week in which approximately 120 hours were saved while approximately 1,445 patient requests were processed autonomously, using an NHS benchmark of approximately five minutes for each manual triage and booking action. These figures describe one implementation and should be tested against the practice's own request volume, staffing model and exception rate.
Langton Medical Group reports approximately 30 hours of GP-led triage removed per week to date across a practice serving approximately 14,000 patients across three sites, according to its published case study. That figure covers GP time only. It does not represent combined clinical and administrative hours, so the two measures should not be added without checking what each workflow includes.
| Metric | Reported range | Evidence basis | Caveats |
|---|---|---|---|
| Clinical and administrative time returned | Approximately 422 or more hours in four weeks | Swanscombe case study | Practice-specific, early post-go-live period, combined hours |
| Autonomous appointments booked | Approximately 5,068 in four weeks | Swanscombe case study | Booking activity, not guaranteed net new capacity |
| GP-led triage removed | Approximately 30 hours per week | Langton Medical Group case study | GP time only, across a multi-site practice |
| GP contact time under GP triage | Approximately 4.0 minutes versus 9.5 minutes under usual care | ESTEEM trial analysis | Overall index-day clinician time was not reduced |
| Total clinician contact time under GP triage | Approximately 10.3 minutes versus 9.6 minutes under usual care | ESTEEM trial analysis | Individual time reductions can mask workload redistribution |
The NHS England primary care access delivery plan reports a median request closure time of approximately 120 minutes over the stated period and says patient contacts are concluded on the same day unless the patient chooses a later appointment. That benchmark makes the full request-to-resolution pathway the relevant comparison. Measuring only the duration of one consultation will miss booking, routing and follow-up work.
Demand is already distributed across several access routes. NHS England recorded approximately 5,781 practices with at least one online consultation system and approximately 4.7 million submissions in April 2025, at around 86.8 submissions per 1,000 registered patients, in its April 2025 online consultation data. An AI assistant for clinics may support parts of this administrative environment, but assisted processing remains different from autonomous triage when a human still makes the routing decision.
The reported AI triage evaluation found approximately 91% of requests automatically booked, with administrative time falling from approximately seven hours to four hours and waiting times from approximately 11 days to three days. Consultation length remained approximately 18 minutes, while total appointments were approximately five fewer per day. The practical lesson is specific: automated booking can reduce handling time without creating equivalent clinical capacity. Any business case should therefore separate manual triage hours, booking hours and appointments released.
A Simple Way to Model Savings for Your Own Practice
A practice doesn't need a vendor forecast to begin modelling. It needs a clean baseline and a definition of what counts as a saved hour.
Step one, record the current workflow
Start with your registered list size, average daily requests, request channels and the people who handle each stage. Use a representative period rather than a single unusually quiet or pressured day. Separate GP-led triage from reception processing and other clinical review.
Then record the time spent on:
Manual triage: GP or duty doctor review per request.
Booking administration: reception handling after a routing decision.
Data handling: copying or checking information in the practice's clinical system.
Exceptions: requests that require escalation, clarification or safety-netting.
The BMJ Open access study reports approximately 4.0 patient-initiated requests per person-year in 2021/2022. That can provide context for demand planning, but your own request data should drive the calculation because access routes and workflow maturity vary.
Step two, calculate gross hours
Use this basic structure:
GP hours saved = requests removed from GP-led triage × average GP triage minutes ÷ 60
Administrative hours saved = requests removed from manual booking and data handling × average administrative minutes ÷ 60
Net operational hours = GP hours saved + administrative hours saved minus exception and oversight hours
Don't insert a percentage for autonomous eligibility unless you have evidence from your own baseline or a clearly defined implementation assumption. The approved UK evidence includes examples of autonomous processing and booking, but it doesn't establish one universal proportion that applies to every practice.
Step three, test the assumptions
A worked example can show the method without pretending to predict a particular practice. Take a notional practice with a list of approximately 10,000 patients. Record its actual daily request volume, then create a low, central and high scenario using locally agreed assumptions for the number of requests that would complete autonomously.
For each scenario, enter the practice's measured GP triage duration and administrative booking duration. If a request takes approximately five minutes for the combined manual triage and booking action, that benchmark can be used only where the workflow matches the NHS benchmark described in the Swanscombe case study. It shouldn't be presented as a universal duration or split into GP and admin minutes without local measurement.
Model the people, not just the requests. A request can leave the GP queue while still requiring receptionist intervention. Your calculator should show both GP hours and administrative hours before adding them together.
Finally, subtract implementation activity, exception handling and ongoing oversight. Compare the result with the relevant case evidence, but don't extrapolate Swanscombe or Langton directly to your practice. The most sensitive variables are usually request volume, the proportion completing autonomously, average manual handling time and the number of exceptions that return to staff.
Where Time Savings Can Be Overstated or Misunderstood
The most common mistake is treating a shorter GP queue as proof that the whole practice has saved time. UK telephone triage evidence provides a useful warning. In the ESTEEM analysis, GP triage reduced initial GP contact time to approximately 4.0 minutes compared with approximately 9.5 minutes under usual care, yet total clinician contact time across the index day was approximately 10.3 minutes compared with approximately 9.6 minutes. The workload shifted rather than disappearing.
A second mistake is counting every digital improvement as a benefit unique to autonomous triage. Online forms can reduce repeated questioning, and structured submissions can improve routing before a consultation. Those benefits are real, but they don't prove that the manual clinical decision has been removed.
Implementation also has a cost. Staff need to understand the new route, exceptions may require dual handling during transition, and the practice must review whether the configured pathway matches its safety model. Net savings should be measured after these activities are included, not only during a successful demonstration.
Safety governance needs equally precise language. GP Triage uses Infermedica, a Class 2B medical device, with UKCA and MHRA context and DCB0129 and DCB0160 documentation. It supports adult and paediatric presentations. Its stated safety wording is “to date, zero recorded clinical safety incidents across approximately two million triages” and “around 97% concordance with GP decision-making”. These are reported performance figures, not guarantees, and every practice still needs local governance, escalation and safety-netting arrangements.
When Autonomous Triage Is the Right Operational Move
Autonomous triage is most likely to create meaningful operational savings where a practice has sustained inbound demand, a substantial manual triage queue and the governance discipline to define exceptions clearly. The benefit comes from removing the manual triage step, not merely helping staff complete it more quickly.
Practices with mature digital workflows may see a smaller marginal gain than practices still relying heavily on manual reception and GP-led triage. Smaller practices with lower request volumes may also find that implementation effort outweighs the released time.
The right decision is made by comparing your own baseline with the model's actual scope. Separate GP hours, administrative hours and downstream clinical demand, then test the result against a realistic implementation plan.
GP Triage provides autonomous AI triage and booking for NHS primary care, with support for adult and paediatric presentations, integration with the major UK clinical systems and a triage summary pushed into the booking record. Visit GP Triage to book a modelling session using your practice's request volumes and staffing data.
