- Use the clinic website as source material, not unquestioned truth.
- Extract structured providers, services, prices and clinic details into a reviewable draft.
- Let the user correct the draft conversationally.
- Ask only for information that is missing or too uncertain to infer safely.
Most onboarding begins by making the clinic repeat itself
Create the clinic. Add every provider. Add every service. Enter every price. Write the description. Explain the same clinic again to the AI knowledge base.
For a small clinic, that can turn a product trial into an implementation project before anyone has seen value.
The website is a good first source because it already contains the public model of the clinic
Clinic websites commonly contain provider names, specialties, services, locations, contact details, prices and descriptions. That is enough to build a useful first draft.
The important word is draft. Websites are stale, inconsistent and written for marketing. AI should extract evidence from them, not silently promote every sentence into operational truth.
A better onboarding flow
The goal is not zero questions. The goal is zero unnecessary questions.
From URL to structured clinic setup
Send me your clinic website and I’ll prepare the first setup.
https://exampleclinic.com
I found 3 providers, 8 services and prices for 5 of them. Dr Levi is listed as an internist. Is that correct?
No. Dr Levi is a cardiologist, and remove physiotherapy — we stopped offering it.
Updated. I still need your clinic timezone and prices for the remaining 3 services.
The user reviews an understandable clinic model instead of filling an empty configuration form.
Corrections should update structured context, not just the chat
If the user says "Dr Levi only works Tuesdays," that information should land in the appropriate clinic/provider configuration. It should not remain as an orphan sentence in one onboarding conversation.
The same applies to services, prices, payer information and communication rules. Onboarding is valuable because it creates reusable operational context.
Show what was inferred and what still needs confirmation
AI onboarding becomes dangerous when it is invisible. The clinic should be able to see the proposed providers, services and facts before they become authoritative.
Uncertain fields should stay unresolved. "I could not find the price" is better onboarding than making one up to keep the progress bar green.
Automation should remove typing, not remove review.
The best onboarding agent does the boring extraction work and makes the remaining human decisions obvious.
How Menaela uses the result
In Menaela, the approved onboarding data becomes clinic context used by Main Menaela and patient-facing communication. You are not filling a one-time wizard that disappears after setup.
That is what makes the onboarding agent part of the product architecture rather than a clever signup trick.
