The Financial Conduct Authority has opened “AI Live Testing” to let firms trial artificial intelligence in live environments with supervisory oversight. For mortgage lenders and building societies, this is the safest path to move beyond prototypes and see how AI performs under real pressure with more advanced PoC’s - affordability checks, document triage, fraud flags, vulnerable-customer prompts and Consumer Duty explanations.
This is a pivotal step not just for digital mortgage lending, but also for the wider shift towards digital home buying, digital conveyancing, and even how AI in the mortgage industry and AI in lending can reshape the end-to-end process.
Three Key points for the AI-first lender
1) This is a controlled runway for PoC’s
Live Testing gives you a framework to define a narrow use case, spell out risks and controls, and agree what “good” looks like with the regulator. If you want an intake agent to gather consented bank data, validate documents and draft an underwriting note, you can do that with a human reviewer and a kill switch - and capture evidence for model and conduct risk committees.
The FCA has also been clear that it does not need new rules to oversee AI, so the existing principles for fair treatment, data use and model monitoring still apply. For firms exploring AI in mortgage lending, this provides reassurance and a safer environment to test without fear of regulatory missteps.
2) Design for explanation from day one
Consumer Duty has raised the bar on understandable outcomes. That means every AI-influenced step should generate a clear, human-readable explanation: which data points drove an affordability nudge, why a document was flagged, how an exception was routed to a person.
Teams that bolt this on at the end will struggle; teams that treat explanations as product features will find it easier to evidence fair value and suitability. The FCA’s mortgage rule review also underlines the need to consider borrower circumstances beyond headline ratios. AI home buying tools - when designed with transparency in mind — can deliver clarity to customers as well as to supervisors.
3) Build machine-readable products
Agentic tools are already being championed in building societies. If your criteria, fees and service levels are not available via stable, versioned APIs with provenance, you will be invisible to those agents.
Start with read-only endpoints for criteria and affordability, then add booking for valuations and IDV under supervised write access. Keep humans in charge of judgement calls and maintain replayable logs. This aligns with the wider proptech push and the principles of Horizontal Digital Integration, where systems connect seamlessly to cut rekeying, reduce friction and improve customer journeys.
Where to start
Pick one journey where AI can remove rekeying or ambiguity. Define the human gate, the data minimisation rules and the explanation format. Apply to Live Testing with a measurement plan that tracks time to AIP, exception rates by reason, abandoned cases and customer understanding scores. Share findings with your board and risk functions as you iterate.
Used well, AI will shorten cycle times, reduce friction for applicants and give underwriters better files, not fewer decisions. This is where AI for estate agents, lenders and conveyancers come together within a digital home buying ecosystem - supported by digital conveyancing and horizontal digital integration - to deliver faster, smarter and more predictable outcomes.
Live Testing your PoCs using the FCA sandbox is the safest way to prove that - and to bring your regulators with you.