// Who we partner with
Partner is a word we can define.
Everyone in this category says it. Most of the time it means nothing, so here is what it means here: your engineers and ours on the same team, you own the asset, and we are measured on you not needing us. Four conditions, all checkable.
Three axes sit below this page — the organization you run, the line you write, and the desk you sit at. Start wherever your problem is already named.
// What partnership means here
Four conditions. Check them against anyone else.
- 01
The people who sold it are the people who build it.
Named engineers in the proposal, with the systems they have built, and you meet them before you sign. No delivery team behind them. The opposite arrangement has a name, a pyramid, and its economics require juniors to bill.
- 02
Your engineers are on the team, not on the other side of a handover.
We build in your shop, on your systems, next to your people. Not because it is a nicer way to work, but because it is the only way the capability transfers. A team that watched a system get built can run it and extend it. A team that received a handover cannot.
- 03
You own the asset from the first commit.
Code, schema, transformations, tests, runbook and documentation live in your repository, on your infrastructure, from day one rather than at the end. Continuing to run and improve it is a separate agreement and your call.
- 04
We are measured on you not needing us.
Exit criteria are written at kickoff, not negotiated at the end, and an owner is named on your side before go-live. A firm whose incentive is the length of the engagement will build you a dependency. Ours is the opposite, and it is in the scope document.
That is the whole definition, and it is deliberately unflattering to us in one place: a partner who is doing this properly is trying to become unnecessary. It is also the reason we can only run a handful of engagements at once.
// What it does not mean
You still hire us.
Not a software purchase.
Partnership here is a description of how the work runs, not a softening of the commercial relationship. You hire Dearborn Labs; you don’t buy software from us. The build fee is fixed before work starts, and continuing to run and improve what we built is a separate agreement and your call.
Not shared risk.
It also does not mean shared risk in the sense a capacity provider would use the word. Your capacity partners, your program partners and your reinsurance panel are counterparties with money moving between you. We are a firm you engage to build something, and the distinction matters enough to state it in the first week.
Not for everyone.
Two conditions have to be true before any of this is worth reading, and both are about you rather than about us.
// Who it is for
The constraint is people, not appetite.
Roots · State of AI Adoption 2025 · vendor research.
Enthusiasm is not what is missing. Senior engineering capacity next to the people who run the work is what is missing, and that is what an engagement here supplies.
So the screen has two parts, and both are checkable before a call.
You already buy outside engineering.
You already buy engineering or software services from an outside firm. Budget exists, a procurement path exists, and nobody has to be convinced that a build can be bought as a service.
You have engineers to own what gets built.
You have enough internal engineering to own what gets built. Our engineers embed with your team and build alongside them. Ownership is only real if there is somebody on your side to run and extend the thing afterward. Without that, the output is a dependency, which is the opposite of the point.
// The stack you actually run
Three kinds of organization, one recurring failure.
The systems differ. The failure does not. In all three the same document arrives in a format nobody controls, gets keyed by a person under a deadline, and everything downstream inherits whatever came in. That is where these pages start.
By organization
Carriers
Combined ratio, regulatory examination, and a modernization program already consuming more than half the IT budget. Your core vendor is now also an AI vendor, which changes what is worth building yourself.
Carriers →By organization
MGAs & program administrators
Binding authority, capacity renewal, and a monthly bordereaux deadline against a capacity panel that each want the same period in a different layout.
MGAs and programs →By organization
Reinsurers & fronting carriers
Portfolio accumulation, ceded recovery, and statutory reporting for business you did not originate and whose data you do not control.
Reinsurers and fronting →Three other routes reach the same pages from a different direction. By line of business → By role and desk → Sponsors with insurance assets →
// The build
Four builds that recur across all three.
- Governance fileSurface: your compliance file, in the states you write in.
- Write-backSurface: the sanctioned, versioned API — verified GA before anything is architected on it.
- Normalization you ownSurface: inbound files, plus a mapping layer that lives in your repository.
- Intake and reconciliationSurface: the document channel plus your policy or book of record.
- 01
Submission and inbound intake with reconciliation
Not extraction — reconciliation. When the schedule total disagrees with the application total, the system says so and shows both instead of silently picking one.
Surface: the document channel plus your policy or book of record.
- 02
Loss-run and bordereaux normalization you own
Per-counterparty field mapping held as editable configuration rather than as one person’s memory, with valuation-date alignment and ALAE treatment stated rather than assumed.
Surface: inbound files, plus a mapping layer that lives in your repository.
- 03
Write-back into the system of record
A decision that survives the next endorsement, appears in the audit trail, and does not break the effective-dated model. Reading is easy. Writing is the step that separates a demonstration from a deployment.
Surface: the sanctioned, versioned API — verified GA before anything is architected on it.
- 04
The governance file, assembled by the build
Model inventory entry, data lineage, pre-deployment testing results, drift thresholds with remediation triggers, and a named human decision-maker.
Surface: your compliance file, in the states you write in.
What we measure
Baseline before build
On your data, with the cohort definition written down.
Touch count per file
Time to first decision
From receipt, measured the same way on both sides.
Rework rate
How often a file goes back for missing data.
Correction rate
On system output, tracked as a first-class number rather than hidden.
// What’s hard about this
Two things are true across every segment.
The segment boundary is not where the data boundary is.
- MGA
- Fronting carrier
- Reinsurer
editable mapping layer, held in your repository
One book of business is read by an MGA, a fronting carrier and a reinsurer, and each names the same fields differently. Lloyd’s Coverholder Reporting Standards v5.2 is the only published specification of its kind; US program and fronting carriers set their own, contract by contract.
So a build scoped to one relationship is a build you repeat per counterparty. Scoping it at the artifact — one canonical record, N outputs from an editable mapping layer — is what makes the second and third relationship cheap.
Lloyd’s CRS v5.2 and US program contracts · current at September 2026.
Delegating the work does not delegate the responsibility.
Roughly half the states have adopted the NAIC AI Model Bulletin, and New York’s Department of Financial Services states the position plainly: an insurer cannot rely solely on a third party’s claim of non-discrimination.
Whoever holds the paper answers for a model somebody else built. The answer is to scope the governance file to the entity that files, and to assemble each party’s evidence where the data originates rather than reconstructing it at examination.
NAIC and state bulletins · current at September 2026.
// What ships with it
The compliance file is part of the build.
- Connecticut
- Requires an annual AI compliance certification attested by a named officer.
- Iowa
- Formally defines bias and outcomes testing.
- Colorado
- The expanded Regulation 10-1-1 compliance deadline passed on 1 July 2026.
State AI bulletins and regulations · current at September 2026.
Every build leaves with the artifact its regulatory surface requires, and it is yours to file, extend and re-test without us.
What the governance file contains →Bring us the workflow you’ve already tried to fix.
A decade of running an AI-native carrier, pointed at one of your workflows for four weeks.