Artificial intelligence is transforming mortgage lending from a slow, manual process into a streamlined, real-time service. As underwriting and decision-making become faster and more accurate, the traditional gap between cash buyers and mortgage borrowers is starting to close.
The mortgage market is not known for agility. For decades, the process has remained trapped in a maze of PDFs, spreadsheets, and manual checks. At best, it has been digitised, replacing paper with pixels. But digitisation alone has never solved the fundamental inefficiencies: fragmented data, human bottlenecks, and slow, inconsistent risk assessments.
The problem is not just bureaucracy; it is architecture. Traditional lenders operate workflows that rely heavily on people to collate, interpret, and act on data. Artificial intelligence, however, offers something different. Not merely faster execution, but a structural rethink of what mortgage underwriting could be if inefficiencies were designed out from the start.
“AI can provide efficiencies and automate aspects of work whenever there is a human workflow,” Stelios Constantinidis, Director of AI at MPowered Mortgages, explains. “Underwriting mortgages is a very manual process, which means there is a tremendous opportunity to apply AI to speed things up.
“Rather than replacing people, AI is reshaping the roles they play. One of the most obvious examples is document analysis. Instead of underwriters flicking through PDFs to compare payslips and tax returns, AI can extract and interpret the data in seconds. These are not chatbot-style language models guessing answers. They are highly specialised tools, built for accuracy, trained on specific tasks, and deployed in targeted areas of the process.”
Predictive and generative, not one-size-fits-all
The current hype around generative AI can create the illusion that all models are conversational, probabilistic and context-aware. In the world of lending, that misconception is both misleading and dangerous. Most of the real gains are coming from models that are neither general nor generative.
“There is an important distinction between digitised and automated,” Constantinidis adds. “The mortgage process has been digitised in many places, there are PDFs and Excel sheets, but AI takes that to the next level by doing the work of underwriters. It helps lenders underwrite mortgages within minutes or hours, not days.”
The result is not just speed, but consistency. Predictive models allocate workloads, assess complexity, and ensure that the right cases are assigned to the right people or systems. One example at MPowered Mortgages uses machine learning to decide whether a case is residential or buy-to-let, then intelligently routes it based on that classification and available capacity. This is not a job for an LLM. It is a classic recommendation problem that benefits from a narrow focus and a clear structure.
In contrast, when used, generative models tend to be smaller and fine-tuned. A typical use case might be extracting structured data from unstructured inputs, such as handwritten surveyor notes or mixed-format valuation reports. Rather than train a large, general-purpose LLM, Constantinidis’ team builds smaller, specialised language models that can be hosted in-house and controlled directly.
“Think of both as tools in a toolbox,” Constantinidis says. “You get to use what is right depending on the circumstance. For extracting information from applicant documents, we use specialised models trained for very narrow tasks. We do not use large-scale generative models to make decisions; they are not interpretable or auditable enough for that.”
Speed meets safety in a regulated industry
Financial services are tightly regulated, and for good reason. But this can breed inertia. Many institutions fear adopting new technologies because they equate automation with a loss of control. In mortgage lending, the stakes are high: get the risk profile wrong, and the consequences can ripple through a balance sheet or a borrower’s life.
What Constantinidis makes clear is that automation does not mean abdication; instead, it means a shift in focus. AI creates more robust audit trails and a clearer handover between machine decisions and human oversight. “The AI is not used to make critical decisions,” he continues. “It is either used to make low-risk decisions or to package information in the right format so that another, more accountable part of the system can make the decision.”
Critically, the team builds models that not only optimise for accuracy but also for reliability, specifically the ability to flag uncertainty and escalate to humans where needed. This kind of interpretability is not a luxury in regulated environments. It is a foundational requirement. “The key is having AI systems that know when to involve humans,” Constantinidis explains.
“When the system says it is 95 per cent sure, we need to trust that confidence level in the statistical sense, not just as a throwaway number.”
It is a practical approach that reflects the evolving regulatory climate, particularly in Europe. The idea of human-in-the-loop decision-making is no longer a philosophical preference. It is a compliance imperative.
The one-click mortgage is already here
At the heart of AI’s impact is not just efficiency; it is experience. The customer, historically treated as a source of paperwork and delay, is now being reimagined as the central user of a streamlined, almost consumer-grade service. “Mortgages are becoming a one-click process,” Constantinidis says. “At MPowered Mortgages, more than 50 per cent of our offers now go out in under a day. This is a fundamental shift from a process that used to take ten days or more.”
The benefits go beyond delight. In real estate, time is a powerful lever. Faster offers mean greater certainty. Greater certainty means a better negotiating position. For too long, cash buyers have had an unfair edge, not because they had better credit or more insight, but because they could move without delay. That advantage is starting to erode. “Cash buyers get a discount because they can move fast,” Constantinidis explains. “But as one-day mortgages become more widespread, borrowers will become just as competitive. That will have a huge knock-on effect in the property market.”
Control, compute and confidentiality
One of the quiet revolutions happening behind the scenes is the shift towards in-house infrastructure. AI, contrary to public perception, does not necessarily mean shipping sensitive data to large tech platforms. In regulated sectors, many of the most valuable use cases can be powered by smaller models, hosted on internal compute. Constantinidis explains. “There are tasks where you must rely on an external LLM provider, but there are many where you do not,” Constantinidis continues. “We host our own models for specific tasks using only a handful of GPUs. This gives us more control and ensures data security.
“That control is critical. Financial data is sensitive, and trust in AI systems will depend not just on how well they work, but also on where they work and who else has access. If a customer provides us with their payslips, they do not want that data sent to a third party for model training purposes. That is why we use smaller language models hosted internally. It avoids the opacity and risk that comes with larger, public models.”
Conveyancing and the collapse of silos
If underwriting can be transformed, so can everything around it. Conveyancing, another headache in the property transaction chain, is next. It involves its own labyrinth of documents, processes and regulatory checks, making it ripe for the same AI transformation. “Conveyancing is the same as underwriting in many ways,” Constantinidis says. “You have policies, data, and a set of processes to follow. AI can help there, too. We are already seeing this happen in legal tech, and the idea of bundling mortgage approval and conveyancing into one seamless process is very realistic.
“In the long run, this will not just reshape the customer experience, it will change how financial institutions operate. The innovators will not be the largest incumbents, but rather the smaller, more agile players who can move quickly, test, and scale. The big banks will eventually adopt it, but it will take some time. The real innovation is happening in smaller firms that do not need 17 sign-offs and a three-year roadmap to get started.”
A future with fewer frictions, not fewer people
That human connection does not vanish with AI. It becomes more valuable, reserved for edge cases, complex scenarios, or situations that require emotional reassurance. In a decade, mortgages may be as simple as booking a holiday online. However, the systems that make this possible will be anything but simple. They will be quietly orchestrating document analysis, risk assessment, compliance, and customer service behind the scenes, all powered by a layered, interpretable AI architecture built for speed, reliability, and trust.
Despite all the talk of automation, Constantinidis does not believe humans will disappear entirely from the process. In fact, he is adamant that for big financial decisions, most customers will still want human contact, even if the process itself becomes vastly more automated. “You might see 80 per cent of mortgages processed without a human, but people will still want to talk to someone,” he concludes. “It is the biggest financial decision of their life. They will want empathy, reassurance, someone who understands their situation.”
And they will make buying a home feel less like a war of attrition, and a little more like a decision people actually want to make.
