On a new receptionist's first day you hand over a sheet: the cancellation policy, whether you can book for tomorrow, prep before an ultrasound, which insurance cards you take. The knowledge base in Kashfy is the same idea, but for the AI that answers your patients on WhatsApp.
This post covers what goes into it, how an answer gets built from it, how to test any question before a real patient asks it, and where the limits are. The feature page is Knowledge Base.
The problem it solves
The AI already knows your schedule, prices and services because they live inside Kashfy. But plenty of things aren't on any screen: what you ask patients to do before a certain test, how late a reschedule is allowed, the rules for an unusual insurance case. Without a place to read from, it either hands every such question to staff or gives a generic answer that isn't yours.
What to put in it
Anything you would give a new receptionist:
- Cancellation and rescheduling policy
- Instructions before a visit or procedure (fasting, what to bring, what to wear)
- Your own insurance rules
- Frequent questions with ready answers
- Plain-language descriptions of procedures
Supported files are PDF, DOCX, Markdown and plain text. A single file can be up to 50MB, and there is no limit on the number of documents. Plain-text notes and FAQs can also be edited right in the dashboard, without uploading a new file.
How it works, step by step
- You upload the file. For example, your cancellation policy as a PDF.
- Kashfy indexes it. This takes seconds. Upload at 2pm and by about half a minute later the AI is answering from it.
- A patient asks on WhatsApp. Say, how many hours before the appointment can they cancel.
- The AI pulls the relevant document and replies in the patient's language and style.
- You can check the source. An answer that draws on the knowledge base cites where it came from: the file name, the section and the exact quote. You always know what the AI told the patient and why.
If the policy changes, upload a new version in place of the old one. The old one is retired immediately, so there's no risk of the AI quoting last year's rule. There's also a version history if you want to see what it was quoting last month.
Test before patients ask
The dashboard has a test tab. Type a hypothetical patient question and see which documents the AI used to answer. If it missed one, rephrase the file or split it into smaller ones so it's easier to find. Do this after every important upload, especially for anything involving money.
Tone and language
A tone slider inside the knowledge base lets you pick casual, professional or formal, and it shapes every reply. On language, you can upload Arabic or English documents. The AI translates between them internally, so an English policy is quoted correctly to a patient writing in Arabic.
With more than one branch
Policies often differ by branch, so the knowledge base can be scoped per branch. Maadi answers with its policy and Zayed with its own. More on branches in Running Several Branches? One Patient File, One Number Each. You can also search across all documents from the dashboard and categorize them by topic: policies, procedures, insurance.
The limits, honestly
- It doesn't replace the doctor. For a real medical question or an angry complaint, the AI hands off to a human with the full conversation.
- Answers are only as good as the document. A vague or contradictory file gives vague answers, so write policies in clear sentences.
- If you spot a wrong answer, you can jump into the conversation and correct it. Every conversation is visible in the dashboard.
Where to start
Start with two files: your cancellation policy and the prep instructions people ask about most. Upload them, test them in the test tab, then add the rest. You can also try the AI yourself on the live demo page. Every feature is on every plan, and the first 30 bookings are free with no card. Details are on the pricing page.
