Trusted speed. A faster route to mortgage offer

The starting point is a familiar handoff. A broker submits a case, a lender queries the property and another party has to find or reconcile the evidence. More data does not help if it describes the wrong flat, is too old for the question or cannot be used by the person reviewing it. We propose mapping those waits before buying more technology.

The government roadmap, published on 18 June 2026 and updated on 19 June, describes around 120 days from accepted offer to completion and around one in three transactions falling through. These whole-transaction estimates explain customer uncertainty, not mortgage application-to-offer performance. [13]

The scale makes small gains material. FCA and Bank of England statistics record £77.4 billion of gross advances in the second quarter of 2026 across around 340 regulated lenders and administrators. These are market measures, not offer-processing outcomes. [1]

Support for earlier information is broad. OPDA reports 1,133 responses to the consultation behind the roadmap, with the agreement levels it highlights at 79% and 82%. This describes consultation agreement, not achieved time savings. [2]

HM Land Registry reports more than 100,000 fewer requisition letters during 2025 to 2026 alongside work on avoidable requisitions. Registration is not mortgage underwriting, but this is a useful operational analogy. Identify recurring evidence defects, give the responsible party actionable feedback and measure whether the defect becomes less frequent. [11]

Five questions make evidence usable

A UPRN is a join key, not a verdict. Unit-level matching needs a confidence explanation and a way to resolve competing matches. Absence of a record must not be treated as absence of a risk. A mining, flood or licensing signal may prompt a specialist check rather than answer the lending question.

Three alternative property-assessment routes

These are alternatives selected under lender policy, not three steps every case takes. A referral can return to review when the required evidence arrives. No route itself issues an offer.

November discussion. Which handoff creates avoidable waiting? We propose selecting one residential or BTL cohort, mapping its evidence and agreeing the definitions needed to establish an application-to-offer baseline. No validated cross-lender benchmark for that cohort is supplied by this paper.

A fluent explanation is not proof. AI is useful here only if a reviewer can check its sources and the verified output removes more work than it adds. Candidate tasks are short document summaries, extraction of facts for checking, conflict flags and a consolidated evidence request.

The RICS responsible-AI professional standard took effect on 9 March 2026 for members and regulated firms. Its emphasis on governance, professional judgement and transparency is directly relevant to valuation work. It does not establish that an automated valuation is appropriate for every mortgage. [5]

Purchase illustration. A fictional first-time buyer is purchasing a flat. The application and record identify different units and lease evidence is incomplete. AI can describe the discrepancy and cite the supplied records. The reviewer resolves the unit and lease questions rather than allowing an inferred match.

Remortgage illustration. A fictional homeowner reports an extension after an earlier valuation. The reviewer checks what changed and whether fresh assessment is needed. An existing property record does not establish the current layout or condition.

Synthetic buy-to-let example. DEMO-002 declares monthly rent of £1,650 while an area estimate is £1,300 to £1,400 and the current tenancy agreement is missing. A useful draft identifies the conflict and requests evidence. It does not establish which figure is right or recommend approval or decline.

Human control in practice

We propose showing the original source beside every material extracted fact, separating quotations from interpretation and making rejection, correction and escalation straightforward. A changed fact, expired permission or materially changed model output should prompt review of the affected assessment, with the earlier version retained.

Reviewer workload belongs in the business case. Measure checking time, unsupported statements, incorrect citations, missed conflicts and failures to recognise missing evidence. Reviewer training, escalation capacity and the ability to switch assistance off are part of delivery cost, not incidental overhead.

December discussion. Which task is worth testing, what is prohibited and what failure would stop the test? Compare the proposed assistance with the existing work rather than assuming a shorter draft is a better result.

Governance, risk and compliance means being able to explain who used which evidence, for what purpose and who took responsibility for the next action. We propose five gates with an owner, stop condition and recovery route agreed for each. They sit within the lender’s existing mandates and do not create new lending authority.

Product and jurisdiction matter. Residential and buy-to-let cases do not necessarily share regulatory treatment. Legal and compliance specialists determine applicable obligations, evidence rights and processing conditions before an authorised test.

Partner roles and open interfaces

Trusted Offer Record combines Crosstide engineering, property intelligence and Thistle risk and governance expertise within an agreed engagement. Provider coverage, source reliability, commercial terms and permitted use need confirmation. Alto and LSL involvement is not assumed or required for the proposed test. No live Trusted Offer Record integration is implied.

Open, documented interfaces should allow evidence to be exchanged without losing identity, provenance or permissions. Sources and AI models should be replaceable, with the same validation and review boundaries applied after a change. The implementation architecture is chosen for the lender rather than prescribed by this paper.

An accountable history connects an action to the evidence. Preserve the observation date, reviewed version, reviewer, reason and timestamp when later information changes. A broker’s delivery receipt cannot clear an underwriting blocker and an AI summary cannot turn an absent document into a fact.

January discussion. Agree the control map and responsibility matrix, then challenge them with a mismatched flat, changed remortgage evidence and a missing BTL tenancy. Specialists also need to examine access revocation, retention, supplier processing, incidents and recovery. The paper is a discussion proposal, not a security or compliance certification.

Lenders need defensible reviews and sustainable processing cost. Brokers need a clear evidence request and a reliable next update. Customers need to understand what is outstanding, why it matters and when they will hear again. A faster internal task has little value if it shifts work to the broker or creates a misleading promise of an imminent offer.

The government roadmap describes wasted fees, stress and plans unravelling when purchases fail. [13] Its reform ambitions include halving fall-throughs and cutting buying times by around four weeks. These are forward-looking policy targets, not achieved results or a benefit attributable to Trusted Offer Record. [15]

One commercial estimate puts April to June 2026 fall-throughs at 71,959 at an average cost of £3,584 each. It comes from a cash home-buying firm rather than official statistics and does not identify the share caused by mortgage evidence. [14]

A dependable business case needs a lender baseline. One quarter of commercial fall-through estimates does not establish an annual trend or identify the share caused by mortgage evidence. A pilot should measure its own costs and outcomes rather than borrow a national headline.

Communication has an owner

We propose agreeing who owns each promised next update, who approves wording and how a missed date is explained. Exact wording should say what is needed and what happens next without making an unsupported lending assurance. Delivery and customer understanding are separate measures.

Consumer Duty provides a design lens through products and services, price and value, consumer understanding and consumer support. Its applicability depends on the activity, product and customer, including the distinction between regulated residential lending and different BTL categories. [8]

Commercial measures belong beside speed. Application-to-offer conversion, completion pull-through, broker preference and retention may be useful if definitions, attribution and observation periods are agreed. Faster offers alone do not establish higher revenue or better retention.

The portal customer-friction example uses 20 fictional applications. Eight have repeat requests, median avoidable waiting is three working days and 14 have clear next-action communication. Three of four flagged support needs are addressed and two process-related complaints are recorded. Every value is invented and should only help participants design measurement.

February discussion. Test whether customers understand a referral and can act on the request. Include people needing additional support and avoid equating a sent message with a successful explanation. The desired result is more predictable service, not a promise that every case can be fast.

We propose starting with one clearly defined cohort and synthetic cases, followed by separately authorised data and a controlled comparison. Residential purchase, residential remortgage, BTL purchase and BTL remortgage should be analysed separately unless the evaluation design justifies combining them.

Agree the baseline before changing the work

The first room should establish definitions and data owners for application-to-offer time, valuation referrals and requests per case. Clock rules mean agreed start, pause and stop points. Include or disclose declined, withdrawn and incomplete applications so a result cannot improve simply by excluding difficult cases.

Application-to-offer is distinct from a decision in principle, valuation turnaround and completion. Report median and upper-tail waiting, manual touches, reviewer time and net cost after data, technology, checking, training and specialist costs. The baseline collection itself requires capacity and a named owner.

Agree decision quality and customer measures alongside commercial outcomes. An independent quality reviewer should check omissions, identity errors, unsupported AI statements and reasoned overrides. Compare similar cases or agree a staged design. A faster pathway earns no credit if material checks are missed or unpaid work is shifted elsewhere.

Proceed, revise or stop

Lender owners set thresholds before results are known. Proceed only when evidence quality, controls and measured value support it. Revise where coverage gaps or extra checking absorb the benefit. Stop where identity failures, unauthorised reuse or material omissions cannot be contained. No generic AI score replaces the lender’s risk appetite.

Change management is part of adoption. Agree staff mandates, training, escalation cover and monitoring, keep a route back to manual work and test the effect of a supplier or model change. The right outcome may be a narrower cohort or clearer evidence request rather than more automation.

March discussion. Produce a bounded pilot proposal with cost assumptions, owners, comparison design and proceed, revise or stop criteria. Prove the join, prove the decision, then earn the right to scale.

Book a 45-minute exploration with David Titterton of Crosstide. Bring three things.

You leave with a candidate cohort, draft baseline definitions and a view on whether discovery is worth scoping. The next step is a separately scoped discovery, not a commitment to a fixed-duration programme, a supplier or a product purchase.

The five rooms from November 2026 to March 2027 are proposed invitation-led discussions using a proposed Chatham House format. No attendee, venue, regulator participation or delivery commitment is confirmed. We welcome a lender and broker voice to challenge the assumptions and define a useful baseline.

Competitor separation and ownership of outputs need agreement before participation. We propose no sharing of identifiable customer records, confidential pricing, future competitive strategy or another firm’s lending policy. Participants should agree information barriers, attribution and rights to use shared outputs rather than assuming discussion grants reuse permission.

Each short theme chapter supports one room. Participants can start with the question most relevant to their role while contributing to the same evidence map and proposed test. Architecture, supplier involvement and the cohort remain choices to agree together.