infoMVP in development · Not a live clinical service
Three things we won’t build

Three things we won’t build

Monday, October 12, 2026schedule11 min read
translateRead in Arabic

Product decisions are known by what you say no to. Medigize has three: no diagnosis, no lowered urgency, and no silent guessing.

Product decisions are known by what you say no to. Medigize has three.

Each one closes a door that would be easy to open. A health product that names conditions, tells people not to worry, or acts without anyone able to check it can look more impressive in a demonstration. This article explains each limit, what it costs us, and how you can hold us, or any health app, to it.

In short:

  • No diagnosis. No screen will tell a patient “you have X”.
  • No lowered urgency. When in doubt, priority goes up.
  • No silent guessing. Every output passes a validation gate and leaves a review trail.

What are Medigize’s three design limits?

Medigize commits to three limits before any feature: no diagnosis, no lowered urgency and no silent guessing. In practice, a licensed clinician makes every medical decision; when the answers leave room for doubt, the tool points to more care, not less; and every output passes a validation gate and leaves a review trail that can be looked at again.

Three things Medigize will not build: no diagnosis, because diagnosis is the clinician’s responsibility; no lowered urgency, because when in doubt priority goes up; and no silent guessing, because every output passes a validation gate and leaves a review trail.

Three limits, set before any feature.

No diagnosis: the clinician names the condition

No screen will tell a patient “you have X”. Diagnosis is the clinician’s responsibility.

What we build sits before and around that decision: organising the complaint, flagging warning signs, pointing to a suitable level of care, and preparing the case so the clinician starts with the information. The diagnosis itself happens with a clinician who can examine the patient. It is the line we drew in what care operations means: the operations around treatment are ours to organise, and the treatment is not.

A diagnosis is more than a name. It draws on the history, an examination, often tests, and the judgement of someone who answers for it. Even with all of that it is hard: the World Health Organization lists diagnostic errors among the common adverse events that can cause avoidable harm in health care. A tool that reads what a patient types has less to go on than a clinician in the room, not more.

WHO’s 2021 report on artificial intelligence in health puts the principle plainly. The first of its six guiding principles, protecting human autonomy, says that in health care, people should stay in control of health-care systems and of medical decisions.

So what does a patient see instead? Questions that help them describe the complaint clearly, a plain warning when a sign calls for an ambulance, and a suggested level of care with the next step. What they will not see is the name of a condition offered as the answer. That has a cost: a name feels like an answer, and a tool that gives one can feel more useful. We accept it. The name belongs to the person who can examine you.

No lowered urgency: when in doubt, priority goes up

When in doubt, priority goes up. The tool never reduces a case’s urgency because it is “probably minor”.

This is a deliberate trade. A cautious tool will sometimes point someone to more care than they needed, and we accept that cost rather than the opposite one. We set out the reasoning in urgency is the clinician’s call.

Any tool that points people to care can be wrong in two directions. It can send someone to more care than they needed: a trip that ends with a doctor’s reassurance, an afternoon lost, perhaps a fee. Or it can send someone to less care than they needed: a “wait and see” for something that should not have waited. The first mistake costs time and money. The second can cost far more, and it cannot always be undone. When a tool has to choose under doubt, it should choose the mistake that can be recovered from.

There is a second reason for caution in our region. The same WHO report warns that systems trained mainly on data from people in high-income countries may not perform well for people in low- and middle-income settings. A tool used across the Middle East and Africa should assume it can be wrong in ways nobody has measured yet, and lean towards care. It is also why we make no claims about results before a pilot measures them.

The rule does not replace you. If you have severe chest pain, trouble breathing, fainting or loss of consciousness, or signs of a stroke, call an ambulance or go to the nearest emergency department. Do not wait for an app, ours included.

No silent guessing: checked before, reviewable after

Every output passes a validation gate and leaves a review trail. No decision is taken that nobody can look back at.

It is a commitment about how the system is built: outputs are checked before they are shown, and the way they were reached can be reviewed afterwards. A system that cannot be looked back at cannot be corrected, and in healthcare, being correctable is part of being safe.

How an output reaches a patient: structured questions, then a mandatory validation gate, then a review trail at every stage, and the clinician makes the medical decision.

Checked before it is shown, reviewable after.

Two terms, in plain words:

  • A validation gate is a mandatory check between what the system produces and what a person sees. An output that does not pass is not shown.
  • A review trail is a record kept at every stage, so that someone can look at a case again later and see how an output was reached.

“Silent guessing” is what happens when a system fills a gap with its best guess and nobody can tell. A gate stops what should not be shown. A trail turns what was shown into something people can find, question and fix.

This is how patient safety is understood today. WHO’s patient safety fact sheet says most mistakes that lead to harm come from system or process failures rather than from one person, and that understanding them means moving from blame to system-based thinking, with incident reporting for learning and continuous improvement. It also lists technology among the factors that can lead to harm, from problems with health information systems to the misuse of technology. And WHO’s AI principles ask for effective ways for people affected by decisions based on algorithms to question them and to seek redress. None of that is possible without a record of what happened.

How do the three limits work together?

Take a composite example, built from many cases and with no identifying detail. Late in the evening, a woman writes about a complaint that has lasted two days. She is not sure whether it can wait until morning. Here is what each limit means for her:

  1. The questions. She is asked structured questions about the complaint: since when, how it has changed, what she has taken, and any long-term conditions, so nothing important is skipped because she forgot to mention it.
  2. The warning signs. If an answer points to a sign that calls for an ambulance, the tool says so plainly, and the call comes first.
  3. The doubt. If her answers leave room for doubt, she does not hear “it is probably nothing”. The tool points to the more urgent of the levels of care that could fit, and to the next step.
  4. The gate. Before anything reaches her screen, the output passes the validation gate.
  5. The trail. Each stage leaves a trail, so the case can be looked at again later.
  6. The clinician. If she chooses to share her case, the clinician who sees her starts with the information instead of a blank page, and makes the medical decision.

At no point does the tool tell her what she has. At no point does it talk her out of care. At no point does it do something nobody can check.

What we will do, and what we won’t

Saying no is only half of it. The other half is what those limits leave room for:

Comparison: Medigize will organise the complaint, flag warning signs, point to a suitable level of care and share a record only with consent; it will not diagnose, lower urgency, release outputs nobody can review or share data without consent.

The line, drawn in public.

  • We will help a patient organise their complaint, flag warning signs, point to a suitable level of care, and share a record only with their consent.
  • We will not diagnose, lower urgency, release outputs nobody can review, or share data without consent.

Would we build it? Test yourself

Six things a health product could do. For each one, decide whether it fits inside these limits.

Try it

Would Medigize build this?

Six things a health product could do. Decide whether each fits inside our three limits.

  1. 1. A screen that tells you “you have X”

  2. 2. Help you organise your complaint before a visit

  3. 3. Lower a case’s urgency because it is probably minor

  4. 4. Show an output that never passed a check

  5. 5. Share your record with a new doctor without your consent

  6. 6. Keep a licensed clinician in every medical decision

0 of 6 answered

Your answers stay on this page.

How can you tell whether a health app has safe limits?

You do not have to take our word for any of this, and you should not take any health app’s word without asking. Seven questions work for any product, ours included:

  1. Does it say it diagnoses? A tool that names your condition from what you typed is doing a clinician’s job without the examination.
  2. Will it ever tell you not to worry? Reassurance about something that needs an examination is not an app’s to give.
  3. What does it do when it is unsure? Look for a stated rule. “When in doubt, more care” is one. Silence is not.
  4. Who makes the medical decision? The answer should be a licensed clinician, every time.
  5. Can its outputs be checked? Ask whether they are checked before you see them, and whether a record is kept of how they were reached.
  6. Who sees your data, and with whose consent? A record should move only when you agree, and only what is needed.
  7. What does it say about emergencies? A responsible tool tells you to call an ambulance when a warning sign is there, rather than keep you typing.

Clinicians and clinic managers can add one more: does it work alongside your judgement and your systems, or try to replace them? Medigize Dock, for providers, works on top of the systems a clinic already uses. See Medigize for providers.

Why say no in public

These are not marketing lines. They are design limits we commit to before we talk about features.

Writing them down in public does two things. It gives patients and clinicians a way to hold us to them, and it makes every feature we build later pass the same test: does it keep the diagnosis with the clinician, keep urgency from going down, and leave a trail?

We are not alone in thinking limits belong in public. WHO’s principle of transparency says enough information should be published or documented before an AI technology is designed or deployed, so that there can be meaningful public debate about how it should, and should not, be used. WHO’s Global Patient Safety Action Plan 2021–2030 makes patient and family engagement one of its strategic objectives. And the Patient Safety Movement Foundation collects stories of preventable harm told by patients and families themselves, because that is where the lessons are clearest.

A limit written in public is also a promise that can be checked. If you ever see a Medigize screen that seems to cross one of these lines, tell us. For how we approach data and consent, see governance and privacy. For the series so far, start with Every patient journey starts with a question.

Frequently asked questions

Does Medigize diagnose patients?

No. No screen tells a patient which condition they have. Diagnosis is the responsibility of a licensed clinician.

Will Medigize tell me my symptoms are not serious?

No. It will not reassure you about something that needs examination, and it never lowers a case’s urgency because it seems minor.

What happens when the tool is not sure?

Urgency goes up, never down. When the answers leave room for doubt, it points to the more urgent level of care that fits, and a clinician makes the medical decision.

What is a validation gate?

A mandatory check every output passes before it is shown. Together with a review trail, it means outputs are checked first and can be looked at again later.

What is a review trail?

A record kept at every stage, so a case can be looked at again later. It is what makes an output possible to question and, if needed, to correct.

Who makes the medical decision?

A licensed clinician, in every case. Medigize prepares the case and organises the journey around that decision.

Is Medigize an AI doctor?

No. Medigize is AI-assisted navigation and care operations software. It does not diagnose or treat, and a clinician stays in every medical decision.

Should I use an app in an emergency?

No. If you have severe chest pain, trouble breathing, fainting or loss of consciousness, or signs of a stroke, call an ambulance or go to the nearest emergency department. Do not wait for any app, including ours.

Medigize is a non-diagnostic navigation tool. This is general information, not medical advice. In an emergency, call an ambulance or go to the nearest emergency department.

References

  1. World Health Organization: Global Patient Safety Action Plan 2021–2030
  2. World Health Organization: Patient safety (fact sheet)
  3. Patient Safety Movement Foundation: Patient stories
  4. World Health Organization: WHO issues first global report on AI in health and six guiding principles for its design and use (2021)

Last updated:

Explore Medigize

Medigize is in development. Explore our current scope and express your interest.

Share:
arrow_backBack to Blog