iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Dawid Kotur’s case for AI in lender compliance is to begin with a bounded task: automate repetitive, document-heavy checks against a lender’s own criteria, while keeping people responsible for judgment, exceptions and decisions that need explanation. In interviews published in July and August 2026, the Curvestone AI CEO argues that a system should show what evidence supported a result—not ask a lender to trust a black box. That is his and his company’s position, not independent proof that a particular tool is accurate or suitable for every lender.
What “built narrow” means in lender compliance
Rather than trying to automate an entire compliance function, Kotur recommends starting with a defined workflow and a clear set of lender-specific rules. His example is high-volume routine checking: work that involves reviewing substantial case documentation against established criteria. The proposed benefit is not that every case becomes an automated decision, but that staff spend less time on repetitive checks and more time on judgment and unusual cases.
This framing appears in the closest matching interview, Modern Lender’s 6 July 2026 interview with Kotur. Curvestone published an adaptation on the same date. A separate interview with The Intermediary, published 3 August 2026, discusses related themes. The supplied headline’s wording was not found as an exact headline, so this article addresses the closest matching interviews rather than claiming to reproduce a confirmed article with that exact title.
How the described workflow is meant to work
Curvestone says its software reviews mortgage and commercial-finance cases, reading materials such as scans, photographs, emails, call transcripts, fact-finds, bank statements and identity documents. It checks case information against criteria set by the lender, identifies exceptions and retains reasoning and source evidence for review. The Intermediary describes a human reviewer approving or overriding findings, with exceptions routed to a specialist.
Those are descriptions from the company and interview coverage, not independently tested product capabilities. The sources reviewed do not establish an accuracy rate, a controlled comparison with manual review, or performance across lenders and case types.
Why Kotur says people still need to own the decision
In Kotur’s account, AI is most useful for sorting routine evidence and surfacing potential issues; human reviewers remain important for interpretation, exceptions and accountability. A lender should be able to inspect the material behind a finding and decide whether to accept or override it. Kotur puts the test this way: “if you cannot explain and defend an automated decision after the fact, you shouldn’t be using it.” That is his practical standard, not a quotation from a regulator.
Rank #2
The interviews connect the discussion to Consumer Duty and the limits of manual spot-checking. They do not establish that Consumer Duty requires lenders to conduct 100% AI review. Kotur’s separate estimate that manual compliance sampling is “around 10%” comes from The Intermediary interview; it is his description of typical practice, not an independently verified industry-wide statistic.
Free tools Windows power users keep installed
One-click scans. No signup required.
What lenders should compare before building or buying
The interviews frame build-versus-buy as an ongoing operating choice, not simply a decision about who writes the first version. A lender considering either route can assess the same practical questions:
- Policy fit: Can the system apply the lender’s own criteria, and can policy changes be reflected without losing control over how checks work?
- Evidence and explanation: Can a reviewer trace a finding to the documents and information that support it?
- Human oversight: Can staff review, approve or override results, and can exceptions reach the right specialist?
- Integration: How will it work with the lender’s existing case and document systems?
- Ongoing maintenance: Who monitors and updates the system as documents, rules and models change?
- Operational proof: Has the lender evaluated the system on its own workflows and case material, with a process for identifying errors and handling them?
- Delivery capacity: Does the lender have the specialist staff and time to build, deploy and sustain an internal system?
Curvestone’s company-authored adaptation estimates that an internal AI system can take “12 to 18 months” to reach production-grade. That is a vendor estimate, not an independently validated sector average. The same adaptation makes the maintenance point: internal systems require continued work as source documents, regulation and models change. The sources offer no neutral cost model or independently validated build-versus-buy comparison, so the decision depends on each lender’s requirements, capabilities and evidence from its own evaluation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the interviews establish—and what they do not
Curvestone describes processing “thousands of checks a quarter.” The figure is a company claim repeated in the interview coverage, not an independently audited volume. It should not be treated as evidence of accuracy, suitability or results at another lender.
Rank #4
In Modern Lender, Kotur says: “Most teams don’t regret the initial build. They regret year two.” It captures his warning that operating and maintaining an AI system can matter as much as getting it into production. The sources do not provide an independently verified return on investment, detailed pricing, a complete security assessment or regulator endorsement of Curvestone’s approach.
The case for starting narrow is therefore also a case for making the first deployment assessable: choose a specific workflow, define what counts as a supported result, keep a human path for judgment and exceptions, and decide how performance and changes will be monitored. Expansion should follow evidence and operational readiness, rather than an assumption that success on one task proves the system is ready to make broader decisions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
