AI proposes, a person disposes
No AI output becomes a claim, a payment, a coverage decision or a cancellation without a human acting on it.
Artificial Intelligence
AI proposes. A person disposes. Every call leaves a receipt.
Where Waypoint uses models, what they are allowed to decide, what they are never allowed to decide, and how the cost and the reasoning are both recorded.
Request a DemoThe rules that do not bend
Default-deny
Every AI capability is an entitlement, off until it is bought and switched on for your organization.
Every call is metered
Cost is recorded per call — capability, model, tokens — so spend is a number you can look up rather than a surprise.
Intake, step by step
Step one
It reads what arrived
Documents arriving by email or upload are read and the loss facts extracted: parties, dates, loss description, policy references, claimant count.
Step two
It builds a draft, not a claim
The extraction becomes a draft intake sitting in the queue. Nothing is assigned, no clock starts, no reserve exists — because no claim exists yet.
Step three
A reviewer confirms or corrects
Confidence is shown per field, so attention goes where the platform was unsure. On confirmation the claim is created through the same door as a typed FNOL.
How a model gets used
Training
Runs as a background job on your organization’s own data.
Registry
Every trained model is versioned and fingerprinted.
Promotion
Models advance a ladder — candidate to active — deliberately, with a reason recorded.
Serving
The active model scores, and both the score and its reasoning are recorded on the claim.
Staleness
Operators are alerted when an active model ages past its threshold.
Remy
The assistant, and its boundaries
| Property | Behavior |
|---|---|
| Scope | Segmented per portal — each Remy answers only for its own portal |
| Availability | Visible only while signed in |
| Position | Draggable; grab the header bar and move it anywhere on screen |
| Input | Text only. There is no microphone and no audio capture |
| Customer Portal | Absent by design — Remy is the internal assistant, and a customer session is not a staff session |
| Entitlement | Sold separately, and off until it is switched on |
Every call is receipted
Per-call receipts
Capability, model, token counts and cost, written for every call.
Your consumption
Your own usage against your own entitlements, visible to you.
Identifiers stripped
Our fleet view of spend carries no claim identifiers, so cost management is not a reason to read your data.
No silent overage
Entitlements gate capability; the ledger records what was used. Neither depends on trust.
What is bought, and what a person still decides
Each capability is sold separately and arrives switched off. Turning one on never moves a decision away from a person.
| Capability | What a person still decides |
|---|---|
| Remy in-product assistant | Advisory only — it explains the system, it does not act in it |
| Agentic FNOL | A reviewer confirms the draft before a claim exists |
| Document intelligence | Extracted fields are confirmed or corrected, with confidence shown |
| Guided FNOL wizard | The intake is submitted by the person taking it |
| Predictive scoring | Promotion of a model, and every action on a score, is a person’s decision |
| BI datasets | Exports run with the caller’s own permissions — no elevation, no way around a wall |
Where we are honest
Bring your hardest intake
We will show what the model reads, what it proposes, and who signs.
Request a Demo