AI doctors could reduce some GP work, but they could also create or transfer work elsewhere. The answer depends on which consultations are genuinely replaced, what remains unresolved and who handles the next step. Counting AI encounters or minutes saved inside one service cannot establish a reduction in total workload.
This is a question about the design and evaluation of emerging care pathways, not evidence that Nolla has increased or reduced NHS demand. The important unit of analysis is the complete episode, including work undertaken by patients, clinicians, pharmacists and administrative teams.
A service can be valuable even when total activity rises. Equally, an apparently efficient consultation can be poor value when another team must repeat it. The distinction matters for patients seeking access and practices deciding how to respond to externally generated plans.
Begin with the genuine opportunity
Consider a fictional patient who would otherwise request an appointment mainly to organise a straightforward next step in an existing plan. A suitable digital service might obtain the relevant information, resolve the request and provide a clear record without another appointment.
Better preparation could also make a necessary consultation more effective. An organised history, an accurate account of previous treatment and a clear statement of the patient's concern might reduce repetition while preserving the need for professional assessment.
These are plausible routes to benefit, not measured savings attributed to a named product. A proper evaluation should allow them to succeed rather than starting from the assumption that every digital contact generates unnecessary demand.
The same applies to follow-up. A service that identifies a practical problem early could prevent a series of confused messages. Its value would depend on resolving the problem, not merely sending more reminders about it.
Identify the work that can move outside the platform
Now consider a different fictional encounter. A patient receives a treatment plan but cannot determine whether it accounts for an existing medicine. They contact their GP to check it. The practice cannot see the assessment, so a receptionist requests information, a clinician reviews the concern and another message is needed to explain the outcome.
The original service may correctly report a quick consultation. That does not capture the time required to establish what happened elsewhere.
Other proposed examples include reconciling conflicting advice, reviewing an uncertain finding and arranging an examination that the remote pathway could not provide. These are not allegations about Nolla or any other company. They are foreseeable types of work that an evaluation should track rather than exclude from its boundary.
Importantly, some transfers are appropriate. Escalating a patient who needs an examination is not a failure simply because another clinician becomes involved. The question is whether the transfer is justified, informative and completed without avoidable duplication.
Substitution and newly met need are different outcomes
Suppose one patient uses a remote service instead of a planned GP appointment. That may represent substitution, although any later contacts still matter. Another patient seeks care who would otherwise have tolerated the problem without consulting anyone. That is additional access, not an appointment saved.
Additional access could produce substantial benefit. It might address a concern earlier or help someone who found conventional routes difficult. But those benefits should be measured directly rather than counted as displaced NHS activity.
A useful evaluation would ask what people would plausibly have done without the service, while recognising the limitations of retrospective self-report. It could combine patient reports with available service-use data and a suitable comparison design.
The worst shortcut is to multiply every AI consultation by the duration of a conventional appointment. That calculation assumes all encounters replace the same alternative, require no subsequent work and deliver an equivalent outcome. Those are empirical questions, not arithmetic facts.
What earlier online consultation experience can teach
The HSSIB investigation into online consultation in general practice, published on 25 July 2024, examined digital access tools rather than Nolla or autonomous prescribing. It identified benefits for access and demand management alongside concerns about implementation, resourcing and patient safety.
The report also discussed possible longer-term demand consequences when digital systems do not support sufficiently holistic and proactive care. It did not establish that all online consultation increases workload, and its findings should not be presented as a causal evaluation of current AI care services.
Its relevance is methodological: implementation and the surrounding service affect what a digital tool achieves. A favourable result at one point in the pathway does not settle the question of overall benefit.
A new AI service should therefore be evaluated against the actual local alternative, including that alternative's delays, duplication and limitations. Comparing it with an imaginary error-free traditional system would be as misleading as ignoring downstream work.
Two identical headline savings, different consequences
The following figures are entirely hypothetical, constructed on 5 October 2026 to illustrate accounting. They are not Nolla results or estimates of typical consultation times.
| Illustrative measure per episode | Hypothetical service A | Hypothetical service B |
|---|---|---|
| Clinician time removed from the original task | 5 minutes | 5 minutes |
| Additional clinician review elsewhere | 1 minute | 7 minutes |
| Net clinician time difference in this simplified example | 4 minutes saved | 2 minutes added |
| Further questions before judging value | Was care appropriate and complete? | Was the extra review necessary and beneficial? |
The calculation deliberately excludes other costs to show why even a simple headline can mislead. A real evaluation should also include administration, patient effort, pharmacy work and implementation support.
Time is not interchangeable merely because it shares a unit. Removing a predictable clerical step and adding an urgent complex decision could change the intensity of the day. Equally, an additional review could prevent harm and be well worth doing. Time measurement must sit beside appropriateness and outcomes.
Build a whole-episode scorecard
A proposed scorecard should record repeat consultations for the same concern, time spent obtaining missing information, clinician review, administrative coordination and unresolved escalations. Patient time matters too: a service that reduces professional work by requiring the patient to repeatedly reconstruct their history has not eliminated that work.
Break results down by the type of encounter and the route taken. Patients managed entirely within the service should not be the only group analysed. Those referred out, unable to complete intake or lost to follow-up are essential to understanding the system's consequences.
Outcome measures should include whether the patient understood the plan and whether the relevant clinical concern was resolved or appropriately transferred. The scorecard should also identify which organisation did the work. This prevents an apparent saving for one provider from disguising a new obligation for another.
None of these measures requires the assumption that fewer contacts are always better. The aim is less avoidable work and better care, not the smallest possible contact count.
How to review a patient who brings an AI-generated plan
For a fictional teaching exercise, give a GP trainee an external plan with a missing medication history. Ask them to identify what can reasonably be accepted, what requires clarification and which parts need reassessment. Neither automatic rejection nor uncritical acceptance is a satisfactory response.
The source of the plan matters for provenance and accountability, but it does not replace examining its content. A polished recommendation can be incomplete; an unfamiliar service can also provide useful information.
As described on 5 October 2026, Ask-iatroX offers free clinical reference with linked sources. It can support a general learning question arising from this exercise, but it cannot supply patient information that was never obtained. Reference access does not itself establish that a transferred treatment plan is appropriate.
The practical goal is to remove unnecessary duplication while retaining the reassessment that the patient's situation genuinely requires.
The companion iatroX article on care after the prescription examines where these downstream tasks arise. The whole-pathway evaluation article explains how to keep incomplete and escalated encounters visible in the evidence.
Frequently asked questions
Does each AI consultation save a GP appointment?
No. Some consultations replace existing care, some meet previously unmet need and others lead to further assessment.
Is increased healthcare activity always a negative outcome?
No. Additional appropriate care can benefit patients, but it should not be described as workload reduction without evidence of substitution.
What is the most useful workload measure for AI care?
Measure total work across a defined episode alongside appropriateness and outcomes. Report which people and organisations undertake that work, rather than relying on platform activity alone.
Explore the clinical questions behind an unfamiliar care plan →