Clause Work in progress

What will my health insurance actually pay?

Clause answers questions about Indian health insurance policies in plain words. Every line of the answer shows the part of your policy it comes from.

The problem

Ask a chatbot

"Is my mother's cataract surgery covered?"

Yes.

What the policy actually says

Yes, but:

  • only after the policy is 2 years old section 4.2
  • only up to ₹40,000 section 3.5
  • and you get less back if the room costs more than the limit section 3.2

Those three conditions sit on three different pages of a 30 to 80 page policy. Miss one and the family gets a surprise bill at discharge.

What Clause does differently

  • It writes down every rule in the policy once, ahead of time, and a person checks the list.
  • For each question, it goes through every rule on that list.
  • Ordinary code works out the money. The AI never does the maths.
  • Every answer quotes the policy's own words.
  • When it can't tell, it says what is missing and asks the one question that would settle it.

What an answer looks like

Every answer about a treatment has the same parts, in the same order. This one is real output from the code, on a made-up test policy.

Is it covered?

Partly. The policy pays ₹40,000 of ₹57,000.

From when?

Covered from 31 May 2025, when the 2-year waiting period for cataract (section 4.2) ended.

Limits

Cap: cataract is paid up to ₹40,000 (section 3.5).

Room rent: up to ₹5,000 a day. Your room is ₹3,000 a day, within the limit.

No co-payment: it starts at age 61, and the patient is 58.

What you pay

The policy pays ₹40,000. You pay ₹17,000, because of the cap for cataract.

Where this comes from

Section 3.5: "Claims for cataract treatment are payable up to Rs 40,000 per hospitalisation."

In plain words: cataract treatment is paid up to ₹40,000 per hospital stay, whatever the hospital charges.

The full answer also shows what Clause understood from your question, so you can correct it, and a line-by-line bill when you give the amounts. No AI writes the answer. Code fills it in from the rules.

How you'll use it

  1. Ask your question. Nothing else is needed to start.
  2. Clause asks only what that question needs, such as the policy's start date or the patient's age.
  3. Upload your policy schedule to fill those in, the PDF your insurer emails you. You check what it read.
  4. Get the answer. Your policies are remembered on your computer, so the next question is just a question.

The first version

In

  • Questions about a treatment, or a term in your policy
  • Which of two policies to claim from first, such as your company's and your own
  • The patient's details, such as a parent with a co-payment
  • 5 real policies, or your own policy wording uploaded
  • Runs on your computer

Not yet

  • Prescriptions, bills or other medical papers; you type the facts instead
  • Checking the cuts on a settlement letter
  • Photos or scans of papers
  • Follow-up questions, or languages other than English
  • An online app

How we'll know it works

Clause is tested against Claude given the whole policy, on 60 questions whose answers are worked out by hand and fixed before any system sees them. An answer counts only if the verdict, the amount to the rupee and the clauses it names are all right. Wrong answers given with confidence are counted on their own, because those are the ones people act on. If Claude does as well as Clause, we'll say so.

Where it stands

  • DoneThe code that applies the rules, the answer form, a made-up test policy, and 58 tests
  • Week 1Real policies, and their rules written down by hand
  • Week 2The test questions, fixed; the AI that writes down rules, and a person checking them
  • Week 3Asking only what's needed, schedule upload, and two policies
  • Week 4The comparison with Claude, the results, and a simple page to ask questions