Built Technologies

The agent is real and doing the work rather than decorating it, reading loan agreements, budgets, inspection photos and reports, insurance certificates and historical draws to reach a funding judgement, and reporting 500,000 automated tasks completed. Underneath sits a construction loan administration platform that ran the market for a decade before the agent existed and which a trade association endorsed in its pre AI form. Apply the removal test and that platform survives intact.

This is the distinction that separates Built from the core insurance platforms rejected elsewhere in this index: those had AI as summarisation or stated intent, this has a shipped agent making the operational decision.

Built publishes an autonomy ladder the customer selects, which is the second instance of this pattern in the index and the clearest outside insurance underwriting. Audit mode reviews and recommends actions for human approval, assist mode handles routine steps while staff make the final decisions, and automate mode executes fully once policies are met while flagging exceptions. Each funding action is written to a compliance file, so the record exists whichever mode is running. Naming the levels, letting the institution choose among them and documenting every action is exactly what most vendors leave to inference.

The mode ladder doubles as a validation methodology, which is worth naming: a lender can run the agent in audit mode, compare its recommendations against what its own reviewers decided, and only promote it to assist or automate once the comparison satisfies them. That is a practical route to validating a model before relying on it, delivered as product rather than as a document.

Published accuracy carries a denominator, and every funding decision writes a compliance file an examiner can inspect. What remains absent is model documentation, a validation summary, error analysis and any stated position on supporting a lender's formal model risk process.

Scale is stated at more than 625 banks, private credit lenders, owners and contractors managing over 350 billion dollars of activity annually, and a major national bank is named as a customer with its lending services group manager quoted on record.

The agent's performance is quantified with a denominator rather than as a headline, at 99.9 percent accuracy across more than 500,000 automated tasks, alongside draw review time cut roughly 95 percent to as little as three minutes, two to five times capacity gain and materially more compliance risks surfaced than manual review. A national banking trade association endorsement and mainstream financial press coverage of the agent launch add independent visibility.

Two disclosures are better than the norm and one of them raises a question. The agent is described as processing against each lender's own standard operating procedures rather than a generic policy, which scopes behaviour to the institution, and the model layer partner is named openly. The training basis is also stated, more than three trillion dollars of real estate finance data, which is unusually specific provenance.

It is also the question: a platform handling 350 billion dollars a year across 625 institutions is the obvious source of a corpus that size, and nothing states whether one lender's draw history informs the agent serving another, or whether participation can be declined.

The consumer privacy surface is limited because borrowers are typically developers and contractors rather than individuals, which genuinely narrows exposure under consumer financial privacy rules. It is not absent, since residential construction lending reaches individual borrowers and the platform holds project financials, insurance certificates, lien records and payment instructions across the whole participant chain. No published privacy framework, retention schedule or subprocessor list was located.

No trust centre, enumerated certification list, attestation scope or audit period was located in this pass. The trade association endorsement implies a diligence process was passed, but that is a commercial review rather than a security attestation, and a platform executing payments and holding loan documentation for hundreds of banks would have supplied attestations privately. The grade records what an outside buyer can verify.

Built supplies technology and holds no licence, the expected posture. Its strongest external credential is an endorsement from the national banking trade association, which involves a review process and carries weight with the community and regional institutions that make up much of the customer base.

Product scope touches statutory instruments directly, since lien waivers, title and insurance validation are legal artifacts rather than administrative ones, and compliance files are produced per funding action. No supervisory instrument is named as a design target.

Borrowers here are businesses rather than consumers, so demographic fairness is a weaker frame and this reads as error governance. The published accuracy is high and the arithmetic still matters: 99.9 percent across 500,000 tasks implies roughly five hundred errors, and in draw funding an error either releases money against work not completed or withholds money from a project that is on schedule, stopping a build and costing a contractor directly. Nothing public breaks error rates down by document type, project type or lender, and no analysis addresses whether smaller or less documented borrowers see more exceptions.

The audit mode gives a lender genuine protection, since it can watch the agent before trusting it, and the compliance file written for every funding action means a disputed decision can be reconstructed with the documents that informed it. Neither is a commitment. No accuracy guarantee, no remediation term and no published obligation where an approved draw releases funds against work that was not completed. The party with least recourse is the builder whose draw is held or reduced, who is not the customer, does not see the policy applied and has no described route to contest the outcome.

The agentic layer's provider is named publicly through a joint launch announcement, which is a direct model chain disclosure and rarer in this index than it should be, and the integration partners are individually identified so a buyer can see which systems exchange data. The training basis is quantified.

What is not disclosed is the rest: no foundation model is named beneath the agent platform, no subprocessor list is published, and nothing states where loan documents and inspection imagery are processed.

The named integrations are the systems this market genuinely runs on, spanning accounting, real estate asset management and construction project management, so data moves between the lender's book, the owner's portfolio and the builder's site records without rekeying. That matters more here than in most lending because the participants are separate companies rather than departments, and the platform's core claim is connecting lender, borrower, builder and inspector in one workflow. Payments execution and disbursement routing are handled in the same system, so the decision and the money movement are not split across vendors.

Delivery is cloud hosted software as a service serving domestic lenders and construction participants, so the cross border complexity facing global vendors in this index does not arise. Residency still matters given the content, since loan files, inspection photographs, insurance certificates and payment records are retained for the life of a construction loan and beyond. No hosting regions, tenancy model, residency options or subprocessor chain were located.

No rates, tiers, billing unit or minimum were located. The platform spans lender software, borrower and builder collaboration, payments execution and diligence services including title, lien monitoring and insurance validation, so scope is the first commercial question, and nothing public indicates whether those are licensed together, charged per loan or per draw, or whether the payment rail carries its own economics.

On the capital side the coverage is complete for its market, reaching commercial banks, credit unions and private credit lenders, and it extends across the whole transaction to owners, developers, general and specialty contractors, title companies and other vendors, which is what makes the connected workflow possible at all. Commercial real estate asset and portfolio management extends it further into the investor side. The boundary is the vertical: this is construction and real estate finance only, with nothing addressing consumer lending, payments, insurance or capital markets.