Fair, alternative credit assessment for underserved people

Merit AI is a London-based financial AI system that recognises the potential of underserved people through behavioural and alternative financial data. Every assessment is evidence-first and advisory only. A human makes every final decision.

Make an enquiry Read our vision

Common questions

Does Merit AI make credit decisions?

No. Merit AI is an advisory system only. It helps predict what someone needs to build independence and contribute, and it never issues, recommends, or determines adverse action. Every output is there to support human decision-making, not replace it. Final decisions are always made by qualified people.

How does Merit AI support financial decisions?

Merit AI organises verified financial evidence, alternative financial data, personal context, current affordability and predictive trajectory analysis into an explainable advisory view. A qualified person weighs that evidence alongside policy and regulatory requirements and makes the final decision. Merit does not automate or replace financial judgement.

What does future potential mean in a financial assessment?

Future potential is an evidence-based view of how income, costs and opportunities may change over time. Merit anchors projections to verified starting figures and UK reference data, shows the reasons and uncertainty, and keeps them separate from current affordability. It is not a promise, a judgement of personal worth or an automated credit score.

What data does the model use?

Merit AI analyses 45 features across financial behaviour, employment history, education, goals, stability indicators, social networks, and behavioural patterns. It integrates Salesforce data, caseworker assessments, in-house document parsing, and reaccreditation pathway information.

How does bias monitoring work?

Merit AI runs continuous fairness metrics across protected characteristics in real time. Automated checks flag discriminatory patterns as soon as they appear, and every model output keeps a bias audit trail to support FCA reporting.

What is the dual neural architecture?

Version 2.1 separates two functions: the Signal Encoder learns from human judgments to create interpretable concept scores, while the Outcome Modeller uses those concepts plus financial features to predict trajectories. This separation makes every output explainable.

How is privacy protected?

Every assessment runs entirely in-house, so no applicant data is ever sent to an external AI. Sensitive fields are protected with post-quantum encryption (CRYSTALS-Kyber-1024, AES-256-GCM) and searched via HMAC-SHA256 blind indexes. A pseudonymisation layer supports internal auditing, and raw neural embeddings are never exposed externally. Only the derived concept scores are visible.