Lorikeet
for Fetch Pet Insurance
Contact us
We rebuilt Pip on Lorikeet to show what it looks like when your AI agent is version-aware about PDS content, actions plan changes on real customer data, and handles both member and vet conversations from the same brain. Give either side a try below.
The chat widget opens with Pip primed to help a pet parent. Ask a coverage question and Pip pulls the right PDS version. Ask about a claim and it looks up real mock data. Ask to change your plan and it quotes back before it writes.
Click a prompt to copy it, then paste it into the chat widget.
Two demo customers live behind Pip: sarah@example.com is on PDS v3, james@example.com is on PDS v4. Any other email works too but comes back as a generic demo customer.
The same Pip picks up when a vet clinic drops in. It verifies the clinic, confirms FetchPay eligibility, and lodges the claim direct so the pet parent never touches an invoice.
Click a prompt to copy it, then paste it into the chat widget.
The vet flow assumes FetchPay eligibility for any clinic name you use, so you can freestyle. Pet parent lookups work best with the two demo customer refs above.
The five things we heard were most worth showing in this build.
Pip answers 27 of Fetch's real help centre articles out of the box: waiting periods, pre-existing checks, FetchPay, how to lodge, cat and puppy cover.
Every PDS article carries a policy version tag. Pip looks up the customer's version first, then filters retrieval to that version. A v3 customer never gets a v4 answer, and vice versa.
One Pip handles both audiences. Pet parents get coverage answers, claim status, and plan changes. Vets get identity verification and a FetchPay-lodged claim.
Pip pulls the customer's plan, sees their pets, quotes their claim history, and adds new pets with quote-back and explicit confirm before the write fires.
Lorikeet takes a seat in your Intercom inbox and handles the traffic you route to it. Not a rip-and-replace. Your seats and routing rules stay.
Every skill, prompt, and mock in this demo was built in the platform itself. No engineering ticket. You can open Pip, change the PDS logic, and republish in minutes.
One agent, four skills, five mock tools, and one tagged knowledge base. Around a day of work end-to-end.
27 real Fetch articles indexed for Pip to search. No tags on these because they apply to everyone.
Four PDS articles seeded manually with a policy_version tag on each. Same-question-different-answer, shared, and v4-only cases all covered.
Customer lookup, claim history, plan updates, vet verification, and FetchPay claim submission. Real systems would replace these on go-live.
One agent, one channel, one widget. The widget on this page is the same one that would sit inside Intercom in production.
From this demo to a Pip that pays out and updates plans on real data.
Click every pill. Try Sarah, then James, then your own email. See where the PDS filter holds and where it needs tightening. Send us anything that surprises you.
Replace the customer, plan, and claims mocks with your real APIs. Pip can now answer any coverage question with a customer's real version and quote their real claim history.
Wire the plan update and vet claim submission mocks to your real systems, with the confirm gates staying in place. This is the point Sam's team is waiting for the platform rebuild.
Lorikeet takes over the tags you route to it. The other 60% of your inbox stays where it is. You can move more over as fast or slow as you like.
Every PDS answer is traceable to a tagged article and a customer's version. Legal can review the tags, not the LLM.
Simple coverage questions and claim status lookups stop needing a person. Your team focuses on the calls only a human can take.
A vet clinic can lodge a claim in under 30 seconds through Pip. FetchPay pays out on the same rails.
Pet parent versus vet is a skill difference, not a separate bot. As pet health expands to a third audience, you add a skill.
Everything Pip does was written in the platform. Add a rule, change a benefit, tighten a check. No engineering escalation.
Today Pip pulls from mocks. When the customer platform rebuild lands, the same skills work against real records. No re-build.
A short workshop to shape the versioned PDS logic against your v3 and v4 for real.
Talk to the team