MQube

A three week manual process is what remains without the AI, and that is precisely the baseline the company measures itself against. Document extraction analyzes bank statements, identity documents and supporting paperwork automatically. Decisioning covers affordability and underwriting end to end, with offers described as issued in seconds. A large language model chatbot ingests a lender's own policy documents to answer criteria questions. A valuation model draws on around 180 property data points to flag likely mismatches before a physical valuation is instructed.

A human checkpoint is described at the decisive moment, with the assembled case presented to an underwriter who checks it before an initial binding offer is issued, which matters because the offer is legally binding on the lender. The valuation model warns brokers early when figures look unlikely to hold rather than failing silently later, and the chatbot is offered with a sandbox so a lender can test behaviour against its own policies before deployment.

One customer description says the platform automates the entire underwriting process, and offers are elsewhere described as delivered in seconds, so where the checkpoint sits in the fastest cases is unclear.

The validation approach is the strongest feature: the company ran its own regulated lender on the platform for years and the results were confirmed externally by industry body data rather than self reported, which is a live market test rather than a benchmark. The sandbox lets a prospective lender evaluate the chatbot against its own policies before committing, which is customer owned testing.

No extraction accuracy, decision error rate or valuation variance figure is published, and the chatbot's performance is stated inconsistently across sources, at 90 percent of broker criteria questions automated in one place and over 40 percent of criteria queries resolved in another.

The evidence is a working lender rather than pilots: the company built a regulated lender on its own technology and let the market test it. That lender reached more than 97 percent of customers receiving a decision within one day, against a three week industry average. The industry trade body's data confirmed it as the fastest growing lender in the country by end 2024, and it reported over two billion pounds in applications and 1.3 billion in completions since launch.

External adoption is now named, with a building society's broker portal running on the platform and a major adviser network partnership in place. Backers include three large financial institutions alongside venture funds.

MQube owns a mortgage lender that competes directly with the institutions it sells the platform to. A building society running its broker portal on this technology is therefore routing application data through a competitor's system. No boundary statement is published, and nothing states whether client lender data is isolated, whether models learn across lenders, or what separation exists between the technology business and the lending business.

The platform ingests bank statements, identity documents and full financial circumstances for mortgage applicants. Together they form a near complete picture of an applicant's finances, and none of the handling terms are published. There is no public data protection agreement, retention schedule, subprocessor list or deletion commitment.

A regulated building society has connected its broker portal to the platform, and a regulated lender operates entirely on it. Security assessment has therefore been passed at a serious standard. Yet MQube publishes no attestation, certification, trust center or enumerated framework for other institutions to examine.

The technology has run inside a fully authorized mortgage lender since 2022. The platform has therefore operated under regulatory supervision in a live market rather than merely being designed for compliance. Binding offers imply the affordability and disclosure standards that govern regulated mortgage lending. No regulator, rule or handbook provision is named for the technology business itself, and a lender licensing the platform carries its own obligations that are nowhere mapped.

Mortgage lending is among the most consequential automated decisions any consumer faces and nothing is published about how the models behave across applicants. Automated valuation drawing on property data carries locational loading, since historical price patterns encode past investment and disinvestment, and affordability assessment from bank statement analysis can read irregular income, benefits or informal work differently from salaried employment. No fairness testing, outcome analysis by applicant type, or explanation of how declines are reasoned appears.

The applicant is the affected party and is unaddressed, and no guarantee, indemnity or correction process is published. Someone declined by automated affordability assessment, or whose property is valued lower than expected by a model reading 180 data points, has no stated route to an explanation, to correcting a misread bank statement, or to human reconsideration. The binding nature of the offer protects the customer who succeeds, not the one who does not.

No base model, provider, hosting arrangement or subprocessor is identified. Large language models are named as the basis of the chatbot and a retrieval framework appears in a conference discussion the company participated in, which is incidental rather than disclosure. Property data feeding the valuation model, and the sources behind bank statement categorisation, are likewise unnamed, and both determine how the system performs on unusual cases.

Deployment reaches into customer environments rather than sitting alongside them, with a building society's broker portal built on the platform and the chatbot offered as a plug and play component for lenders' internal systems, broker and customer facing portals and existing chat interfaces. A major adviser network distributes products originated through it. No core banking, servicing, bureau or valuation provider system is named, and there is no public developer documentation.

The company states an intention to take the technology international, which would make residency a live question. Building societies and lenders handling applicant financial records would examine it during procurement. No hosting provider, region selection, residency commitment or private deployment option is published.

The company publishes a commercial claim about outcomes rather than prices: it describes its own lender as the lowest marginal cost originator in the market. That speaks to the economics the platform produces without indicating what a lender pays to license it. No pricing, packaging or basis of charge is published for the technology business.

Product coverage within its chosen market is thorough, spanning residential and buy to let, individual, limited company and portfolio landlords, affordability assessment, portfolio lending and product switching, and the buyer set reaches lenders, building societies, brokers and adviser networks. The constraint is deliberate concentration on one product in one country. International expansion and a tokenization offering are stated as intent rather than delivered, so the coverage is deep rather than broad.