Ask a registrar what they actually need at three in the afternoon and it is rarely "what does the national guideline say" in the abstract. It is "what is the pathway here" — which referral form, which local threshold, which clinic the patient goes to. National guidance from NICE, CKS and SIGN sets the principle; the local system tells you what to do with it. The gap between the two is where a lot of clinical AI quietly underperforms.
What the gap looks like, concretely
National guidance gives you the rule. The local system tells you how it is applied here: local referral criteria, two-week-wait pathways, local laboratory reference ranges, which investigations are available and how to order them, local formulary choices, and the pathway quirks every department carries. Two hospitals following the same national guideline can run materially different local processes. A tool that knows only the national layer can be correct in principle and useless in practice.
The market is racing to add local knowledge
This is not a hypothetical concern, and 2026 has made it a competitive battleground:
- Tandem has made local-protocol upload part of its clinical decision support: clinic administrators can upload referral pathways, local lab ranges, two-week-wait pathways and clinic-specific protocols, which Tandem says remain within the organisation and are not used to train its models.
- Heidi Evidence is explicitly positioned around "region-specific guidance and formularies", built with named partners including NICE, BMJ Group, MIMS and HealthPathways, and lets clinicians upload their own sources.
When two of the most prominent clinical AI products both build toward local and regional knowledge, that is a clear signal: national guidance alone does not answer the clinician's real question.
The hard part is not storage — it is usability without cognitive load
Holding local documents is easy. Making them genuinely usable in the moment is the difficult part, and it turns on a few things that are easy to underestimate:
- Source citation. A clinician needs to know which document an answer came from — and ideally the supporting passage — not just that "the system says so." An uncited answer is not inspectable, and an answer you cannot inspect is one you cannot fully trust.
- Versioning and dates. Local protocols go stale faster than national guidance. A pathway from three years ago presented as current is worse than no answer. The date and version have to travel with the answer.
- Conflict handling. Sooner or later local policy and national guidance disagree. A trustworthy system surfaces that conflict and shows both, rather than silently choosing one. The clinician, not the model, resolves it.
- Next-step reasoning over recall. The useful output is not a quoted line but "here is the threshold, here is the local pathway, here is what to do next" — with the sources attached.
How iatroX approaches the knowledge layer
iatroX is built on this discipline at the national level first. AskIatroX surfaces source-linked answers from validated UK guidance — NICE, CKS, SIGN and the SmPC — with the source attached, so the answer is inspectable rather than merely fluent. The local layer is the natural extension of the same principle, and the principle does not change: cite the source, show the date, surface conflicts, and reason to the next step rather than reciting text.
AskIatroX is a UKCA-registered, MHRA-listed Class I device — it informs the clinician's judgement rather than directing the decision. That is exactly the relationship local knowledge needs: a clinician deciding which pathway applies, with the tool making the relevant guidance fast to find and verify.
That is the harder, more honest version of "local guidelines support" — not a folder of uploaded PDFs the clinician still has to read under time pressure, but a knowledge layer that gives a trustworthy, cited, current answer at a glance.
An honest caveat
None of this is fully solved by anyone yet. "Local knowledge without cognitive load" is a genuinely hard problem, and the failure modes — stale protocols, uncited answers, silent conflicts — are exactly the ones that erode clinical trust fastest. Getting it right matters more than getting it shipped quickly, and any vendor claiming otherwise is worth a second look.
The missing layer between NICE and the corridor is local, current, source-linked knowledge a clinician can trust without stopping to verify it. National guidance will always be the foundation. But the tool that earns clinician trust is the one that knows the difference between what the guideline says and what actually happens here — and is honest about both.
Frequently asked questions
Why aren't national guidelines like NICE enough on their own? National guidance sets the principle, but care is delivered through local processes — referral criteria, local thresholds, available investigations and pathway specifics that vary between sites. A clinician's practical question is often about the local pathway, which national guidance does not specify.
Which clinical AI tools are adding local guidelines? Among others, Tandem allows clinics to upload local protocols into its decision support, and Heidi Evidence is positioned around region-specific guidance and formularies via partners including NICE, BMJ Group and MIMS — both signalling that national guidance alone is insufficient.
How should a clinical AI handle a conflict between local and national guidance? It should surface the conflict and show both sources rather than silently selecting one. The clinician remains responsible for resolving the discrepancy, so the tool's job is to make the disagreement visible and clearly sourced.
What sources does AskIatroX use? AskIatroX draws on validated UK guidance, including NICE, CKS, SIGN and the SmPC, with the relevant source attached to each answer so it can be verified rather than taken on trust.