What prompted you to write the AI Charter in the first place, and what did you feel was missing from the industry conversation?

The noise. The mortgage technology space has become saturated with AI claims that, when examined closely, do not hold up. Almost every conversation centres on capability – what AI can do. Very few address governance. What happens when it goes wrong, who owns the liability, where the data goes, and what it costs at scale. Brokers and lenders deserve a clearer picture. We needed a documented framework that customers, compliance teams, and partners can actually interrogate. 

There’s a lot of hype around AI right now. Where does the mortgage industry sit on that spectrum and are we seeing real transformation yet? 

Closer to hype than transformation, if I am being honest. There are genuine use cases delivering real value, but there is also a significant volume of solutions that amount to a chatbot skin over a general-purpose AI model, dressed up as sector-specific innovation. The industry is not yet good at distinguishing between the two, and that creates real risk for firms making procurement decisions under pressure to be seen doing something with AI. Real transformation requires consistent, reliable, compliant outcomes at scale. Most deployments today do not meet that bar. The firms building properly now will be the ones delivering genuine transformation in 3-5 years. 

In the charter, you describe many AI solutions as “a thin layer on top of someone else’s engine.” What does that mean in practice and why should brokers and lenders care?

It means a large number of AI products in this sector are essentially a shiny user interface connected to OpenAI, Google, or Anthropic’s API. The vendor has not built any underlying AI capability. They have written a prompt, wrapped it in a product name, and taken it to market. The implications for brokers and lenders are material. Client data passes through infrastructure you do not control, the vendor’s pricing is entirely at the mercy of their upstream provider, and you have no guarantee of consistent outputs because the underlying model is probabilistic by design.

For a regulated advice process operating under Consumer Duty, those are not minor considerations. 

Cost control is not something that usually comes up in AI discussions. Why is it one of your four pillars and how does building your own AI change the picture? 

Because it will determine whether AI in mortgage technology remains commercially viable. 

There are two cost risks, but right now, the industry is only concerned with one of them. 

The first is more obvious. As usage grows, costs grow with it and if you are dependent on a third-party model, you have no mechanism to control them. 

The second is more fundamental and almost entirely overlooked. The price you pay for AI today is artificial. Major providers are running at significant losses. The token costs firms are building their business models around are not sustainable. Some analysts forecast price increases of 10-30 times current rates as providers are forced toward commercial viability. That is not a fringe view – it is simple arithmetic. 

So it’s not just about what happens when your usage scales. It is what happens when your supplier’s economics change. Firms committing to third-party infrastructure today are making a long-term bet on a short-term price. 

We use the analogy of a jet engine powering a bicycle. Deploying a large general-purpose language model to perform a narrow, structured mortgage task is wasteful by design. Smaller, purpose-built systems are more accurate, more consistent, and significantly cheaper to run. The engineering discipline is in knowing which tool applies where. Companies routing everything through OpenAI or Google are exposed to pricing decisions entirely outside their control – today, and far more tomorrow. 

We at Mortgage Brain are not. 

Mortgage data is among the most sensitive personal financial data there is. How should firms be thinking about data sovereignty and what does Mortgage Brain do differently?

As a governance question first, and a technology question second. Before any AI system touches client data, firms should be able to answer: where is that data processed, who has access to it, could it be used to train the provider’s model, and what happens in a breach? 

The question of whether client data is potentially training other models is certainly one for concern. When a broker or lender sends client data through a third-party AI system, there is a real risk that data, your clients’ sensitive financial information, may be used to train someone else’s model. Most firms have not read the terms carefully enough to know whether that is happening. Many of the major providers reserve the right to use inputs for model training unless enterprise agreements explicitly prohibit it, and even then, the contractual protections are not always watertight. For regulated firms operating under Consumer Duty, that is not a compliance footnote. It is a material risk that needs to be owned and answered. 

Data processed outside the UK or EU compounds this further. Different legal frameworks apply, different authorities have oversight, and in a dispute the firm may have very limited recourse. Under UK GDPR, that exposure is not theoretical. 

Because Mortgage Brain builds and runs its own AI, none of that applies. Data shared with Mortgage Brain exists only within Mortgage Brain’s systems. There are guardrails. It goes nowhere else. The FCA’s direction of travel on AI oversight is clear, and firms with proper data governance frameworks in place now will be considerably better positioned as those requirements develop. 

The charter makes a strong point about consistency. Why is AI variability a specific problem in mortgage advice?

Because mortgage advice is a regulated activity. If the same client scenario produces different compliance interpretations or different risk flags depending on when the system is queried, you do not have a reliable process – you have a probabilistic one. That is not compatible with a compliant advice framework, and it would not survive Financial Conduct Authority (FCA) scrutiny. Where consistency is required, we design systems that behave deterministically. The same input produces the same output, every time. An adviser can rely on that. A compliance auditor can rely on that. Variability might be acceptable in a general-purpose chatbot. In mortgage technology, it is a liability. 

One piece of advice for mortgage firms thinking about AI right now? 

Ask harder questions before you buy. Specifically, where does client data go, who controls the underlying model, what happens to your pricing if the provider changes their rates, and can you demonstrate consistent outputs to a compliance auditor? If the vendor cannot answer all four clearly, the solution is not ready for a regulated environment. AI in mortgage technology should be infrastructure – robust, governable, and defensible. If it looks more like a demo than a system, treat it accordingly. 

Mortgage Brain has launched a dedicated AI Zone on its website, giving brokers access to practical tips on using AI safely and the full AI Charter. Visit www.mortgagebrain.com to find out more.