judgment as owned infrastructure
Tom builds a judgement layer of how your best underwriters actually decide and it's yours*
Under the desk today. Plugged into your stack tomorrow.
* Tom is trained only on your own underwriters' recorded decisions. It surfaces, grades, and attributes the judgment already in your book — it does not invent risk appetite, and it does not replace the underwriter of record.
The model of how your desk decides. Owned by you.
Tacit captures how your most experienced underwriters actually make decisions, turns it into a structured model your institution owns and can interrogate, and serves that judgment to your people and your AI systems at the point of decision.
Decision capture
Tacit instruments the underwriting decision without changing how your people work. Decisions, prices, terms, and outcomes are recorded as a structured stream against the submission that produced them.
Deviation ledger
The single most important and least recorded act on your desk, turned into queryable data: every price away from the technical model, with the reasoning behind it.
Decline capture
The risks your underwriters walk away from are invisible to every pricing model. Tacit records the declines and their reasons — the refusals your actuarial function has never been able to see.
Referral memory
The reasoning exchanged when a junior escalates to a senior is, in most institutions, lost the moment the conversation ends. Tacit turns each referral and its resolution into a permanent node in your judgment model.
Judgment profiles
For each underwriter, a structured calibration signature: the risk types they handle most consistently, where their judgment is strongest, and where their practice diverges from the desk and the technical reference. Derived from observed decisions, validated against ground truth.
0.91
Practice model
How your institution actually decides — where your people converge, where they diverge, where collective practice has moved from the written guideline. Disagreement is preserved, not averaged away, because the disagreement is information.
Structured, not imitative
Not a black box that copies behavior. Every element is decomposable and attributable — an account of judgment you can defend to an examiner, transfer to a new practitioner, and interrogate yourself.
Decision support at the workbench
Your underwriters consult the model in the flow of work: the recommended action, the confidence your institution’s practice supports, the alternatives a dissenting senior would weigh, and the reasoning behind each.
Agent grounding
Your AI systems consult the same model through a programmatic interface at decision time — grounded in your institution’s actual practice, not a generic model’s defaults. Every response carries attribution, confidence, and provenance.
The self-writing file
The documentation your underwriters least want to produce becomes a byproduct they never write. From the captured decision stream, Tacit generates an audit-ready rationale for every decision.
Bench development
New underwriters develop against the captured practice of your senior desk: shown a real case, asked to commit to a decision, then shown how your most experienced people would have approached the same risk, and why. Judgment transfers through structured practice, not proximity alone.
Deployed in your environment.
Governed to your regulators’ standard.
Built for the security, integration, and governance requirements of the world’s largest regulated carriers.
A house model that carries cues the table cannot.
over time, on your book, a model of your own experts can begin to outprice the generic industry model — because it carries the cues your actuarial table never had a column for. we treat that as a result to prove on your data, not a claim to make on a homepage.
and the underwriter whose judgment the model keeps reusing can earn from contributing it — a recognition mechanism a carrier chooses to run, not a feature we administer. it flips capture from something seniors guard into something they want to feed, which is the difference between a capture product that adopts and one that dies on the vine.
what Tom does not do: invent risk appetite, replace the underwriter of record, or claim a loss-ratio number before it has earned one on your book. the model is only ever as good as the desk it learns from — which is precisely why it is yours, and not ours.
See the spread on your own desk.
if you are a Chief Underwriting Officer, Chief Data Officer, or head of underwriting inside a carrier or MGA, write to us with the subject line "Operator". the first step is a one-week noise audit on your own book — the real number, no slide deck, no pitch. we reply within two business days.