In regulated work, the prize isn't automation - it's the decision.
Where money and decisions flow in US lending and insurance, how large the addressable work is, and who already sells into it.
Illustrative estimates from cited public inputs. The first figure is a cost pool (what the industry spends doing this work), not a software market; the obtainable figure is the revenue a vendor could plausibly capture.
How the money - and the decisions - flow
Toggle lending vs insurance. Highlighted boxes are where Arbiter overlays.
The lender rarely keeps your loan - it makes it, then sells it. The decision is the moment of truth. Arbiter overlays the decision points - the highlighted boxes - without replacing the systems around them.
The cost pool, the serviceable slice, the obtainable share
Drag the assumptions to re-run the model. Every figure is reproducible from the inputs shown.
Selling into top IMBs, banks and credit unions is a long enterprise cycle; a realistic 3-5yr capture is ~4% of the $22B SAM ≈ $0.9B. Cross-check: ~5.1M loans x ~25% automation penetration x ~$700 platform fee per automated decision ≈ $0.9B. The platform fee pays for the work; a decision warranty, where written, is a separate premium priced from the measured defect rate (see the Platform page).
Sources: mba.org · sf.freddiemac.com · mba.org · statista.com
The competitive landscape, by layer
Who sits where in the stack, from each company's public material, and where Arbiter would have to win.
The platform of record where a loan or policy actually lives end-to-end: application data, documents, workflow, conditions, closing, servicing. It is the operational backbone lenders/insurers build everything else around. High switching cost, deep integrations, slow to rip out. The insurance equivalent is the "core" suite (policy admin, billing, claims).
Their edge: Horizontal decision tooling that credit and risk teams author themselves, with agents, rules and models in one flow and a case manager for the human side. Customers across lending and insurance make it the most direct horizontal comparable to Arbiter.
Arbiter: The closest horizontal rival, and the hardest fight. Taktile gives risk teams a platform to build decisions on; Arbiter would ship the credit-specific parts as product: every extracted fact quoted and verified in code, prebuilt credit gates, and a committed adversarial suite that gates each policy change. Whether that beats a well-staffed team on Taktile is the question a pilot has to answer.
Their edge: Owns the system of record for a huge share of US originations; data gravity, deep third-party integrations, regulatory entrenchment, and very high switching cost. Distribution moat: AI ships to an installed base rather than needing new sales.
Arbiter: ICE keeps AI out of final approval, pricing and disclosure. So does Arbiter. The difference is what reaches the human: a computed, gated, evidence-cited decision draft with a replayable record, on Encompass or any other system of record.
Their edge: Patented expert-systems IP and a buyback/repurchase warranty that puts money behind the decision - a differentiated risk-transfer moat few competitors match; truly autonomous decisioning rather than copilot.
Arbiter: Candor showed lenders will pay for a warranted mortgage decision. Arbiter's thesis is that the same pattern, verified facts, deterministic gates and a priced warranty, extends to commercial credit, where no comparable warranty exists today.
Their edge: Embedded POS at many regulated banks/CUs with deep core/LOS integrations and account-opening footprint; multi-product (mortgage + consumer + deposit) gives cross-sell and stickiness.
Arbiter: Blend calls Autopilot a non-decisioning tool: it reviews documents and builds the needs list. Arbiter sits one step later, turning verified facts into a gated decision draft, and can run on top of Blend and the system of record underneath it.
Their edge: Scale and incumbency in credit decisioning across many markets, with data integrations and policy tooling that risk teams already know.
Arbiter: Provenir and FICO are decision engines a bank configures. Arbiter would sit in front of or alongside them, reading unstructured credit files into verified facts, which is the part general engines leave to the lender.
Coverage against autonomy
Breadth of coverage against how much of the decision each player automates. Positions are an analyst judgement, not a measurement.
Wedge and moat
A newcomer cannot out-distribute ICE or out-data Rocket head-on, so the plan does not try.
Every override and outcome flows into versioned policy + memory + recalibrated thresholds. Incumbents capture data - not structured judgment lineage with a closed loop to convert it.
Their guidelines, overlays, RBAC, approval matrices and audit history live in our ledger. Leaving means rebuilding their control framework and losing the replayable trail exams depend on.
A warranty on the financial loss from defects in auto-rendered decisions, written only on the slices where shadow-mode measurement supports it and priced from that measured defect rate. Regulatory accountability stays with the lender; only financial loss can be transferred.
Who is positioned to win
A read of the field, including where this concept is weakest.
AI ships into an installed base that rarely switches. Slow to automate the decision itself, and hard to dislodge.
Risk teams already author credit strategy here. The strongest position for an agentic decision layer if they close the extraction gap.
Win on what generalists cannot fake quickly: a warranty backed by a track record, or fair-lending defensibility.
Not to topple any of them. Win one decision (commercial credit) on verified extraction and replayable gates, earn a measured defect rate, then price a warranty on it.
Company facts are from the public sources linked on each card, with the date they were checked. None of the companies named is affiliated with Arbiter or has reviewed this analysis. Market sizes are illustrative estimates assembled from cited public inputs, not official figures or vendor revenue.