JammJar adds AI document checks for UK mortgage advisers

JammJar, a UK mortgage-advice software provider, has expanded its case-management platform with AI tools that draft fact-find entries from client calls and analyse bank statements and income documents, which JammJar says an adviser reviews before they enter the case file, according to a report published by trade title Mortgage Soup on 18 September 2026. The update follows the company's Summer '26 launch webinar. It affects advisers who use JammJar's software, and by extension their clients, whose calls and financial documents may pass through an AI extraction layer before the results reach a regulated case file. Public material does not establish whether the update is available to all customers or is being rolled out in stages.

JammJar frames this as time saved on administration. The more consequential question for a regulated market is narrower: whether the platform's human-review and audit-trail controls actually support the accuracy and record-keeping that UK mortgage advice requires, or simply move the same risks somewhere less visible. JammJar Limited is registered in England and Wales under company number 15294462, according to Companies House. No evidence in the public record establishes that JammJar itself holds FCA authorisation; it is a software supplier to regulated firms, not a regulated adviser, and the firms using it remain the accountable party.

How the AI turns conversations into proposed fact-find entries

According to Mortgage Soup's report and JammJar's own site, the platform analyses client conversations and generates proposed fact-find changes. Before any change enters the case record, the adviser sees the source wording the suggestion was drawn from and a confidence level attached to it. JammJar's home page separately describes calls becoming fact finds, though it gives less detail on the approval step than the dedicated document-check page does. The stated design is that the adviser, not the software, decides whether a suggestion is correct.

Bank statements and income documents: a four-stage review

JammJar's document-check page says the platform classifies and assigns uploaded files, then analyses income, spending and commitments, turning the results into proposed fact-find updates that link back to their source. Mortgage Soup describes this as a four-stage review covering commitments, spending, flagged transactions and income.

JammJar says the system can identify missing pages, income gaps, unexplained deposits and other outstanding evidence before a case is submitted. It says it calculates average income, applies evidence rules for cases such as weekly-paid applicants, categorises spending and surfaces commitments shown in bank statements. Where information is missing, JammJar says it is recorded as "Unknown" rather than inferred — a design choice that, if it holds in practice, would avoid the software quietly filling a gap with a guess.

Separately, JammJar says identity checks are provided through Onfido, with document and biometric verification, and that referred results require an adviser to approve or reject them rather than pass automatically.

Human review and the audit trail: how much oversight is real

JammJar's document-check page says advisers can inspect the exact source excerpt and confidence score behind a suggestion, then accept or reject it, and states plainly that "nothing writes to the file on its own." It says every accepted entry links to its source through an append-only trail, and that queried transactions keep a message thread rather than being silently resolved.

This is worth setting against JammJar's own home page, which separately says information extracted from documents "updates the fact find on its own" — language that sits awkwardly next to the document-check page's insistence that nothing writes itself. It is not clear from the public material whether these statements describe different stages of the same workflow, different product configurations, or simply inconsistent marketing copy, and whether adviser approval is compulsory in every case is not established.

The design nonetheless echoes a concern the industry itself has raised. The FCA's Mortgage Rule Review feedback records respondents worrying about AI errors, hallucinations and incomplete information, and says respondents considered human judgement essential, supporting the principle that qualified advisers should review and sign off AI-based recommendations. A source excerpt, a confidence score and an accept-or-reject step are consistent with that principle in outline. Whether they deliver it depends on things the packet cannot establish: what the confidence score actually measures, how advisers respond to a low one, and whether daily use turns "review and approve" into a habitual click-through that offers little real scrutiny. No public evidence describes how advisers in live deployments actually use the control.

Efficiency claims: three figures, no visible methodology

JammJar and Mortgage Soup's report between them give several timing claims, but none comes with a disclosed sample, case definition or independent benchmark.

It is not clear whether the three-to-four-minute figure and the five-to-ten-minute figure describe the same task measured differently, or two different tasks. Readers should treat all four figures as the company's own claims about its own product, not as independently verified results.

What FCA rules still require, regardless of the software

Using AI tools does not change a firm's underlying obligations. Under FCA rule MCOB 4.7A.25, effective from 31 January 2020, a firm giving regulated mortgage advice must retain the customer information used to assess suitability, and an explanation of why the advice given was suitable, for at least three years. A traceable digital record could help a firm evidence that it met this duty — but possessing software logs is not the same as the underlying information being accurate or the advice being suitable.

The FCA said on 8 June 2026 that it did not plan to introduce separate AI-specific regulation. Instead it intends firms to govern AI use through existing frameworks: the Consumer Duty, the Senior Managers and Certification Regime, and established governance and control expectations. In practice, that means responsibility for an inaccurate fact-find entry or a poorly reviewed AI suggestion sits with the advising firm and its accountable senior managers.

Data protection: unresolved questions about who controls what

Processing client calls and financial documents through AI raises separate obligations under UK data protection law, which the ICO regulates. ICO guidance says organisations must identify and document whether each party is a controller, joint controller or processor for each specific processing operation, and that a controller appointing a processor needs a written contract setting out responsibilities, liabilities, security requirements and assistance with data-subject rights. It is not established in the public material which role JammJar occupies for transcription, extraction, or any model training or support use of client data, nor whether firms have such contracts in place.

ICO guidance also says AI processing frequently creates risks that require a data protection impact assessment, though this depends on the specifics of each deployment rather than being automatic. A UK mortgage firm adopting JammJar would need to make and document that assessment itself.

Whether this workflow counts as solely automated decision-making under UK data protection law, and which safeguards that would trigger, depends on whether adviser review is genuine rather than nominal and on how each firm configures the system — a classification that cannot be settled from the public material alone. ICO guidance on automated decision-making is also under review following the Data (Use and Access) Act 2025, so firms should check current ICO material rather than rely on older summaries before drawing conclusions.

JammJar's document-check page also says that only redacted facts leave a case, never the raw file. No data-flow diagram, subprocessor list, or independent security assurance was publicly available to verify what that boundary covers in practice, and the company's main site separately references UK and EEA data centres without clarifying which data types sit where.

What's not yet established

Several things a UK firm would need before relying on this workflow are not answered by the material currently public: independent accuracy or error-rate testing of the conversation extraction, bank-statement analysis or income calculations; how the confidence score is calibrated and what threshold should prompt caution; whether the audit trail is technically immutable, exportable, and configurable to FCA retention periods; who can amend or delete an audit entry; and evidence from actual firms using the platform day to day, rather than the company's own description of its features.

Where to check the detail

Firms considering this or any AI-assisted advice workflow can review the FCA's suitability record requirements directly in the Handbook at MCOB 4, and the ICO's guidance on AI accountability and automated decision-making, rather than relying solely on a software supplier's own description. What happens next is whether JammJar publishes fuller technical documentation on its audit trail and data handling, whether named firms describe their experience of the update in practice, and how the FCA's ongoing work on AI in financial services — including any further examples of good and poor practice arising from its Mortgage Rule Review — treats tools of this kind.