Industry
Banking
Banking Services Transformation: Unlocking Exponential Value
Highlights
- Rising complexity, administrative burdens, and outdated processes strain mortgage underwriting capacity within UK building societies, threatening their traditional relationship-based lending advantage.
- A human+AI underwriting model can address these pressures by automating routine administrative tasks and data checks, which frees expert underwriters to focus on complex risk assessments.
- Building societies must adopt a structured four-stage implementation pathway, transitioning from stabilisation to digitisation, augmentation, and controlled automation, while preserving crucial lending judgement.
- A benchmarking model must be established to track operational success across five specialised lenses—speed, underwriter effort, decision consistency, broker experience, and digital adoption—to ensure the solution serves the member-centric ethos.
- To successfully deploy this model, building societies must start by mapping existing bottlenecks (stabilisation) before introducing AI-led automation (digitisation). This staged approach builds internal trust and ensures underwriters remain in control of complex lending decisions.
On this page
On this pageinpage
Why should UK building societies adopt a human+AI underwriting model
For many UK building societies, underwriting remains manual. This is not just a legacy constraint but more a mutual model’s value proposition and strategic differentiator. Manual underwriting helps societies lend to borrowers with more nuanced profiles, apply common sense to edge cases, and serve niche customer segments, whose circumstances typically do not meet the demands of the standard mortgage lending model.
Replacing manual underwriting is not the goal. The objective is to redesign and enhance the existing model by leveraging artificial intelligence (AI) technologies, while preserving the flexibility and member-centric ethos of building societies.
Building societies continue to operate with technology infrastructures that are partially or wholly outdated. Manual underwriting capability is of value to the UK specialist and intermediary market, as complex cases still require significant human judgement and expertise, even as technology speeds up parts of the process. However, there are challenges as well: brokers continue to highlight delays, difficulty in reaching underwriters, and fragmented communication as real pain points in the UK mortgage journey.
Modernising the underwriting model is key. However, replacing manual underwriting is not the goal. The objective is to redesign and enhance the existing model by leveraging artificial intelligence (AI) technologies to assist, augment, and transform, while preserving the flexibility and member-centric ethos of building societies. However, this creates a distinct design challenge for UK building societies.
The way forward for building societies is to shift to a human+AI underwriting model, which offers a balanced approach, preserving human judgement at the point of decision-making. At the same time, the model leverages AI and workflow automation to enhance efficiency, consistency and service quality, and strengthens broker and member experience. This approach is particularly appropriate given the growing role of building societies in the UK mortgage market, including their substantial contribution to recent market growth and their continued emphasis on intermediary-led distribution.
How does a human+AI model preserve the human edge in underwriting?
UK mutuals must target assisted underwriting with AI augmentation rather than end-to-end automation, deploying AI solutions to prepare and enrich case information, leaving human underwriters to interpret and take the final decision. This approach is in line with regulatory expectations, where lenders are accountable for affordability and meaningful customer outcomes under the Financial Conduct Authority’s (FCA) Consumer Duty regulation. Additionally, with the FCA introducing a higher degree of flexibility in mortgage rules, lenders will also need to comply with the Mortgage Conduct of Business (MCOB) requirements around risk-based decisions.
Human+AI underwriting should not be interpreted as a stepping stone to autonomous credit decisions but as a governance model where decision authority rests with humans while manual tasks are increasingly machine-assisted.
In our view, a successful human+AI redesign of the existing underwriting model for UK building societies must focus on four outcomes:
- Protect the strengths of manual underwriting for complex and specialist cases, while reducing effort on routine evidence collation and checks.
- Reduce cognitive load on underwriters by giving them a structured, evidence-led case summary rather than forcing them to reconstruct the file from a variety of unstructured sources such as emails, PDFs, portals, and notes.
- Introduce decision augmentation before automation, using AI to support condition-setting, affordability evidence reviews, policy matching and case triage, while preserving human sign-off for material decisions.
- Measure productivity in a way that reflects mutual priorities, including turnaround time, broker experience, exception-handling quality, consistency of decisions, and customer outcomes, not just case volumes.
In the context of UK building societies, human+AI underwriting should not be interpreted as a stepping stone to autonomous credit decisions. Instead, it must be viewed as a governance model where decision authority rests with humans while manual tasks such as information gathering, evidence collection and case preparation are increasingly machine-assisted. This distinction matters as the objective of mutuals is to continue with a relationship-based model while improving efficiency and customer outcomes.
How can building societies implement the human+AI underwriting model?
The proposed model allows UK building societies to improve operational efficiencies without compromising the human element that is the foundation of their relationship-based lending model. However, it gives rise to a critical redesign question: how do building societies and mutuals retain the judgement-heavy benefits of manual underwriting, while eliminating the operational friction that frustrates brokers, slows decisions, and increases the cost-to-serve? In our view, the following design principles must guide the roadmap to a human+AI underwriting model.
The medium-term target for building societies and mutuals must be AI-augmented underwriting, rather than touchless underwriting.
Design principle 1: Adopt AI-led automation for preparatory activities while retaining human control over judgement-based decisions aimed at optimal allocation of responsibilities between AI and humans (see Figure 1).
Figure 1: AI-augmented underwriting model
The medium-term target for building societies and mutuals must be AI-augmented underwriting, rather than touchless underwriting. This is a more practical and achievable path and is consistent with their relationship-led and member-centric value proposition.
Design principle 2: Reduce cognitive load in manual underwriting. Underwriters are constrained by administrative tasks in their risk assessment process. Such tasks typically include following up for documents, reading emails, interpreting broker notes, switching systems, and repeatedly reopening files. A human+AI underwriting model must therefore focus heavily on improving cognitive efficiency, prioritising the following:
- Highlight exceptions and unresolved issues instead of revalidating information that has already been gathered.
- Link all data points to source document or broker inputs ensuring traceability for quick verification.
- Adopt system-generated draft communications, allowing underwriters to edit rather than compose from scratch, improving communication efficiency.
The objective is to empower underwriters to shift to risk adjudication rather than spending time on administrative functions.
Design principle 3: Adopt AI for decision augmentation. For UK building societies, decision augmentation is the correct starting point as underwriting is manual. Many societies compete through specialist and nuanced lending decisions, and governance around models and customer outcomes still requires strong oversight. We recommend adopting an approach with four levels of decision support:
- Level 1 - Assist: AI extracts data and summarises the case without offering recommendations.
- Level 2 - Recommend: AI proposes likely outcomes while underwriters remain mandatory approvers for all cases.
- Level 3 - Controlled pre-clearance: AI reviews and performs low-risk tasks but the overall decision still remains human.
- Level 4 Narrow automation: AI-led targeted automation for low variance segments, where evidence policy and model governance are mature.
The final step is to define a practical roadmap with the primary objective of creating a human+AI underwriting model where technology-driven automation handles routine processing and analysis, enabling underwriters to focus on complex judgements and deliver a scalable, consistent, and member-centric lending experience. Let us examine the stages in a pragmatic transformation roadmap (see Figure 2).
- Stabilise and measure the existing underwriting process.
- Digitise case preparation with details from the application, documents and third-party data sources
- Augment decision-making with AI recommendations, while preserving human accountability.
- Automate low risk, rules-based activities with clearly defined guardrails.
Figure 2: Governed model for moving from stable manual control to selective, trusted automation
How is the productivity of the human+AI underwriting model measured?
In a building society context, measuring underwriter productivity purely on the basis of the number of mortgage loans reviewed and decisioned per day is overly simplistic and understates the strategic importance and responsibilities of the role. When mutuals are trying to balance process efficiencies with judgement, flexibility, broker service, and customer outcomes, productivity measurement metrics must reflect that operating reality. The benchmarking model must therefore incorporate five proof lenses to demonstrate whether automation is improving flow, effort, quality, experience, and adoption, while ensuring that accountable human judgement remains central to mortgage decisions (see Figure 3).
This benchmarking model is important because the objective of redesign is not to maximise raw volume at the expense of case quality but to increase underwriting capacity without diluting judgement or member outcomes.
Figure 3: Benchmarking model to measure the productivity of a human+AI underwriting model
This benchmarking model is important because the objective of redesign is not to maximise raw volume at the expense of case quality. It is to increase underwriting capacity without diluting judgement or member outcomes.
What is the future of mortgage underwriting in a human+AI world?
Manual underwriting will continue to remain a strategic differentiator for UK building societies, but its role will evolve significantly in the years ahead. While people will continue to remain accountable for the decisions that matter the most, AI will become a trusted co-pilot for manual underwriters, delivering efficiency and insights by continuously analysing data, validating information, highlighting risks, and recommending actions in real-time. Underwriters will increasingly shift from administrative processing to higher-value decision-making and relationship-driven outcomes. To retain an edge over their peers, building societies must take the first crucial step of embracing a human+AI underwriting model.