Watch ten flows from start to finish.

1. Insurer's systemThe privacy consent was saved to this session.
2. Insurer's systemThe details went to the insurer, and an SMS code was issued.
3. Insurer's systemThe insurer reported that the code had expired.
4. Insurer's systemThe new code was verified, and the customer's session opened.
5. Insurer's systemOn the first quote attempt, the insurer asked for health disclosure consent first.
6. Insurer's systemConsent was saved.
7. ModelIt requested the quote again from where it left off.
8. Insurer's systemPremiums for two plans were calculated in about 7 seconds.
9. Insurer's systemA scheduled message went out because the customer hadn't written.
10. ModelIt read this as a provider search and took the city from the message.
11. Insurer's systemThe provider list was queried by city and product network.
12. ModelIt searched the list for that hospital and said plainly it wasn't there.
13. ModelIt separated two questions: frequency and where to pay.
14. Insurer's systemThe frequency changed, and the premium was recalculated.
15. Insurer's systemA purchase link for this quote was created.
16. Insurer's systemThe insurer's purchase page opened in a new tab.
17. TeamPayment and the policy are completed on the insurer's page; the result doesn't come back to the chat.

The assistant's limits in this flow: It doesn't take card details or payment in chat, doesn't issue policies and doesn't promise exact cover.

1. ModelIt understood the customer wants a motor liability quote.
2. ModelIt opened the right vehicle and registration steps.
3. Insurer's systemIdentity received. The code wasn't sent yet because the marketing preference is missing.
4. Insurer's systemThe preference was written to the authorized system, and an SMS code was sent.
5. Insurer's systemCode verified. Customers whose phone is already verified never see this step.
6. Insurer's systemThe vehicle and previous policy were looked up in 3 to 5 seconds.
7. ModelThe lookup found the address missing, and it tried to complete it with one question.
8. Insurer's systemThe address was updated, and the vehicle was looked up again.
9. Insurer's systemThe authorized insurer list came back, and 16 separate requests went out.
10. Insurer's systemAnswers came in about 12 seconds; 8 insurers quoted.
11. Insurer's systemThe server worked out the best price and the best online price; the model can't change them.
12. Insurer's systemA scheduled message went out because the customer hadn't written.
13. ModelIt read this as the difference between the legal limit and additional cover.
14. Insurer's systemQuote details were fetched.
15. ModelIt reported only what the quote data says.
16. ModelIt linked the request to the chosen quote.
17. Insurer's systemThe quote was checked and still valid.
18. Insurer's systemInstallments came back, and a payment session opened.
19. Insurer's systemThe insurer rejected the installment option.
20. ModelIt stated the error as it is and didn't say a policy exists.
21. Insurer's systemThe customer approved on their bank's 3D Secure screen.
22. Insurer's systemThe bank reported the result, and the insurer issued the policy.
23. Insurer's systemThe policy summary came back, and the PDF was prepared.

The assistant's limits in this flow: It doesn't email quotes, doesn't change prices and doesn't say a policy exists after a payment error.

1. ModelIt understood the DASK request.
2. Insurer's systemThe customer isn't logged in.
3. ModelOn WhatsApp it didn't ask for the phone separately.
4. Insurer's systemLogin started, and an SMS code was sent.
5. Insurer's systemCode verified.
6. Insurer's systemSaved homes were checked; none found.
7. ModelIt split the text into province, district and neighborhood.
8. Insurer's systemEach part was checked against official address records.
9. Insurer's systemThe street was found in official records.
10. Insurer's systemThe request came back with a missing field, and the model asked for the next one.
11. ModelIt asked about and confirmed residential use.
12. Insurer's systemThe home record opened, and the DASK quote request went out.
13. Insurer's systemInsurer quotes were collected.
14. ModelIt made no price comparison because the premiums are the same.
15. Insurer's systemThe pre-contract information form was prepared.
16. Insurer's systemIt checked the form was opened and saved the consent. Consent isn't accepted before opening.
17. Insurer's systemThe summary, quote and call time went to the agency's system, and automatic replies stopped.
18. TeamThe representative completes payment and the policy.

The assistant's limits in this flow: It doesn't suggest DASK prices differ between insurers and doesn't say belongings are covered.

1. Insurer's systemThe privacy consent was saved to this session.
2. Insurer's systemAn SMS code was issued.
3. Insurer's systemCode verified, and the customer's policies were listed.
4. Insurer's systemThe selected policy's documents were opened.
5. ModelIt read this as roadside assistance and towing.
6. ModelIt searched the document for the towing clause.
7. Insurer's systemID, plate and phone were masked in the excerpts.
8. ModelIt limited the answer to the clause it found.
9. ModelIt suggested a related right as a follow-up.
10. ModelIt found both clauses and reported the negative condition too.
11. ModelIt saw the answer is in usage records, not the document.
12. ModelIt didn't guess without a source and pointed to the right channel.

The assistant's limits in this flow: It doesn't open claims, cancel or change anything, and doesn't guess at what the document doesn't say. This flow is in pilot today.

1. ModelIt understood this as a claims question. No ID is needed; the guide needs the insurer's name.
2. Insurer's systemThe name was searched in the 27-insurer claims guide.
3. Insurer's systemThe phone and link from the guide were passed on unchanged.
4. ModelIt concluded: two vehicles, property damage only, both can move.
5. ModelIt suggested the mobile report only because the conditions fit.
6. ModelIt wrote the text from the customer's account. The customer sends it.
7. ModelIt read the document and the image in 15 to 20 seconds.
8. ModelIt compared the insurer the customer named with the document.
9. ModelIt described what's visible without commenting on amount or fault.
10. Insurer's systemInsurer B's line came from the guide.
11. ModelIt recognized a question that needs a decision.
12. ModelIt made no decision and said what it depends on.
13. ModelIt said what it can't do and asked for consent and a time.
14. Insurer's systemThe reason, summary and phone went to the agency's system, and automatic replies stopped.
15. TeamThe representative calls the customer and helps with the report.
16. TeamThe insurer's claims team decides fault, cover and payment.

The assistant's limits in this flow: It doesn't open claim files, make up file numbers, decide fault, cover or compensation, or give legal opinions.

1. ModelIt read the document in 15 to 20 seconds.
2. Insurer's systemID and plate were masked.
3. ModelIt limited the answer to the policy lines.
4. ModelIt understood the staff member wants a short draft.
5. ModelIt wrote this without promising payment.
6. Insurer's systemInsurer details came from the claims guide.
7. ModelIt saw the question belongs to a different policy.
8. ModelIt didn't guess about something outside the document.
9. TeamThe technical staff member explains in their own words and enters the record.

The assistant's limits in this flow: It doesn't decide cover, doesn't talk to the customer itself and doesn't log in to insurer screens with staff passwords.

1. ModelUnderstood a cancellation request and its reason.
2. Insurer's systemAn SMS code was issued.
3. Insurer's systemThe code was verified.
4. Insurer's systemThe document this reason needs was read from the request types.
5. Insurer's systemThe document was added to the request and not passed to the model.
6. ModelUnderstood that amount and date are the insurer's decision.
7. ModelGave no amount.
8. Insurer's systemThe request and document went to the organization's customer system, and a request number was created.
9. TeamThe team checks the document; the insurer decides on cancellation and refund.

The assistant's limits in this flow: It doesn't cancel the policy itself, doesn't state a refund amount or date, and doesn't interpret the uploaded document.

1. ModelIt read this as a product and rule question.
2. Insurer's systemThe bank's instructions and product documents were searched.
3. ModelIt answered with the source. No customer data was needed.
4. ModelIt looked up the rule on another insurer's policy.
5. ModelIt refused, and gave the reason and the correct wording.
6. ModelIt didn't guess without a source.
7. TeamThe staff member talks to the customer and takes the explicit request.

The assistant's limits in this flow: It doesn't recommend insurance to customers, doesn't write text presenting insurance as a loan condition and doesn't take explicit requests.

1. Insurer's systemEnding policies were fetched from the agency system.
2. Insurer's systemCustomer identities were masked.
3. ModelOrdered them by end date, claims and last year's renewal day.
4. Insurer's systemThe policy and claim record were opened.
5. ModelExplained the reason for the price change from the record.
6. Insurer's systemThe quote request went to the insurers.
7. ModelAdded the package difference next to the prices.
8. TeamYour staff talk to the customer and explain the quote.
9. ModelDrafted it from what the staff member wrote.
10. TeamThe staff member checks and approves the record.

The assistant's limits in this flow: It doesn't call customers itself, doesn't renew a policy without approval and doesn't promise discounts.