A claim is not an event. It is a trajectory.
Everything expensive about a claim happens after somebody decides it is one. The cheap moment is earlier — and it is legible earlier, in language, before it is legible in paperwork.
It shows up in a support thread first.
A rider says they were hurt. A customer mentions a lawyer. Somebody describes a failure, not a complaint. The signal is already in your helpdesk — it is simply sitting in the same queue as a refund request.
By the time it is a file, the options have narrowed.
Notice starts clocks. Counsel changes tone. The set of things you could have done cheaply closes quietly, and nothing in the workflow marks the moment it did.
So the product is a window, not a verdict.
Forecede does not decide anything. It finds the matters worth a human's attention while attention is still worth something, and puts the context on the desk with them.
The same claim, drawn twice.
One drawing, two sheets. The first is the shape a claim takes when nobody is watching it. The second is what changes when someone steps in. The whole product is the distance between them.
Illustrative. The drawing explains the shape of the argument; it does not report a measured outcome. Both dimension lines are deliberately unlabelled — we do not yet have a time-to-notice figure we would stand behind, and an unlabelled span is worth more than an invented one.
Four things happen, and a person is in three of them.
Read
Conversations arrive from your helpdesk on a schedule and on webhook. Personal details are stripped before any model sees the text, and our build fails if a model call site isn't covered by that check.
Weigh
Each conversation is compared against your own reviewed history, then assessed for severity by a multi-vendor model ladder. Where the assessments disagree with each other, the matter is routed to a person rather than resolved by a tiebreak.
Resolve
A governed early-resolution path, with the jurisdiction, the exposure and the similar matters already assembled. No model settles anything. A person approves every figure.
Record
Every decision writes an audit line — who acted, on what evidence, when, and what the system believed at the time. The file that reaches the TPA is one somebody can check afterwards.
The claim is one half. The book is the other.
Early resolution only matters if the operation around it is real. These surfaces are in the product today — you can walk through them on the demo, and we would rather you did that than take a diagram's word for it.
The book
Accounts, insureds, agency structure and servicing roles — who owns what, and who is allowed to see it.
Renewals
A staged pipeline with prep built in, so the work in front of a renewal is visible before the date is.
Certificates
ACORD forms, holders and issuance — with the underlying policy in the same system as the certificate rather than a different one.
Contract check
Reads an agreement against the policy and marks each insurance requirement met, gap — or unverifiable. The third answer is the one that makes the first two worth reading.
Commissions & proposals
Statement reconciliation, and a proposal path that carries the comparison work through to the client-facing document.
Triage inbox
The email that runs a brokerage, filed against the account it belongs to instead of a folder somebody remembers.
What we will and won't claim.
Most of this industry sells certainty. We would rather be the vendor you can check — including on the parts that are not finished.
What's connected
Helpdesk connectors are live: Gorgias, Zendesk, Intercom, Freshdesk.
Carrier-side ingestion is roadmap, not product. We will say so every time we are asked, including in the room where it would be easier not to.
What's governed
Personal data is redacted before any model sees it, and that is enforced by a build gate rather than a convention.
Where the system is uncertain it hands the matter to a person. A person approves every figure, on a record that can be audited afterwards.
What we haven't proven
Our evaluation rests on one founding tenant's reviewed corpus. Until a second tenant's data exists we do not publish generalization claims, and we will not imply them.
We are in pilot recruitment: conversations, not signatures. When that changes, this page changes with it.
If a number isn't on this page, it's because we couldn't yet stand behind it. That is the same discipline we would apply to a reserve.