Your insightsour automations
We build and deploy AI for insurance companies. From data architecture to production systems. Embedded with your team.
We build it. You own it.
The AI infrastructure and workflows we build are yours to keep. No license, no subscription.
// Why seven and not nine
Why this is a practice rather than a menu of projects.
A list of nine builds prices like nine builds. That is the wrong shape for the thing a carrier is actually deciding, because none of these workflows is finished when the first system ships.
Take underwriting. Year one is usually intake reconciliation, because it is the highest-confidence work in the category and it produces the data everything else needs. Year two is the exposure model that intake made possible, and the appetite screen that the underwriting manual can now be checked against. Year three is a portfolio view that is worth building only once the first two exist — decisioning on a book with no operating data behind it is a dashboard, not a decision.
That sequence is not a roadmap we sell. It is the order the work has to happen in, and it is why the discipline is the unit that matters and the build is the increment. A carrier that hires us for one build and never comes back has still got a system its own team owns, which is the deal either way.
The seven are the seven budget lines this work gets funded out of: intelligent document processing, workflow and process automation, integration, data platform, model risk and AI governance, application development, and statutory and delegated reporting. That is not a coincidence. It is the test we applied.
// If you arrived looking for something specific
The named builds, and where each one sits.
Nine builds recur often enough to be written up on their own. Six carry a full write-up and three are shorter notes, each listed with the discipline it belongs to.
| The build | Discipline | What it is |
|---|---|---|
| Submission intake | Underwriting workflow automation | Cross-document reconciliation across ACORD 125, 126 and 140, the SOV, the loss runs and the broker’s email body — with conflicts surfaced rather than silently resolved. |
| SOV normalization | Underwriting workflow automation | Arbitrary broker schemas to a canonical model, geocode resolution recorded as a field, and COPE gap filling with provenance on every filled value. |
| Document summarization | Claims operations automation | Medical records, demand packages and forensic reports, with page-level citation as a design requirement rather than a feature. |
| Loss-run normalization | Claims operations automation | Valuation-date alignment, ALAE treatment, open-reserve development, and claim-versus-occurrence counting. |
| Write-back | Core system integration | A decision that survives the next endorsement, appears in the audit trail, and does not break the effective-dated model. |
| Semantic layer | Core system integration | A Guidewire Cloud Data Access or Duck Creek replica turned into an analytics model that respects effective dating. |
| Bordereaux | Delegated authority and ceded reporting | N formats from one book of record, and the reported-versus-booked reconciliation nobody currently owns. |
| Exposure translation | Delegated authority and ceded reporting | CEDE, EDM and OED across the versions actually in circulation, with a validation report on every conversion. |
| Governance package | AI governance and model risk | Model inventory, lineage, bias testing, drift thresholds, override specification, vendor validation. |
Swipe to see the full table →
If the workflow you care about is a reinstatement queue, a premium audit, a surplus-lines filing or something else that is not on this list, it still belongs to one of the seven. The list is what recurs, not what we are limited to.
Bring us the workflow you have already tried to fix.
We will tell you which discipline the workflow belongs to, and what we would build first.