Why, How, What

Why: The world is producing more interrupted lives, and its systems only read continuous ones

Every society runs on gatekeeping systems that decide who gets access to what: credit, housing, work, education, professional recognition. Almost all of them measure the past. Credit history, employment records, qualifications from institutions the assessor happens to recognise: these are proxies that work tolerably well for people whose lives have been continuous, local and well documented, and fail everyone else.

The number of people they fail is growing, and every macro trend points the same way:

• Displacement is at historic highs. Well over a hundred million people are forcibly displaced worldwide. Each arrives somewhere with their capability intact and their paper trail gone. A surgeon who fled a war is still a surgeon; every conventional scoring system reads her as a blank.
• Work is becoming less linear. Career changes, retraining, gig and portfolio work, industries restructured or automated away. The continuous single-employer record that legacy systems were designed around is becoming the exception, not the rule.
• Lives cross borders. More people will live, qualify, work and borrow in more than one country than ever before. Records do not travel; potential does.
• AI is being bolted onto the old proxies. The dominant response has been to automate the past-facing systems. Automating a system that cannot see interrupted lives does not fix its blindness; it industrialises it.

The result is a widening gap between what people can do and what the records say they can do. That gap is not just individual tragedy. It is squandered capacity at a civilisational scale: surgeons driving taxis, engineers stacking shelves, whole communities locked out of contributing what they are able and eager to contribute.

The insight: humans already know how to see potential

A good caseworker sitting across from an applicant can see what no file records: persistence through disruption, sacrifice for family, rebuilding after loss, steady effort toward a self-chosen goal. This recognition is real, it is skilled, and it is remarkably reliable. The near-perfect repayment record of RefuAid across years of lending to people every conventional scorer would refuse is the proof. But that recognition is trapped: it lives inside individual conversations, cannot scale beyond the hours in a caseworker’s day, and vanishes when the caseworker moves on.

Merit exists to learn from those acts of recognition. Not to imitate their surface, and never to replace them. The purpose is to genuinely, deeply learn how one human recognises the drive and resilience of another, so that this way of seeing people can be honoured, defended and extended.

Why it had to be built this way

• A mission lock in law. Merit AI is a non-profit whose Articles of Association apply all income solely to its objects, prohibit any distribution of profit to members, and place a Guardian Share over the mission itself.
• No external AI, ever. Nothing an applicant shares ever leaves Merit’s own process. There is no call to any outside AI provider anywhere in the system.
• The human is the teacher, not the failsafe. Every prediction is advisory, and a human makes every decision. Remove the human and you remove the teacher.

How Merit works today

Merit is live, currently supporting RefuAid’s interest-free lending. An application moves through six stages:

1. Evidence is read entirely in-house. Bank statements, CVs and documents are parsed by Merit’s own software on Merit’s own servers. Every figure carries a label saying whether it was verified, declared or assumed.
2. An honest financial picture, with the working shown. Merit builds monthly income and spending from the statements themselves, and is deliberately cautious: where the evidence is thin, it says so and withholds figures rather than dressing up a shaky number.
3. A forward trajectory, anchored in official UK data. Using published reference data, Merit projects how the applicant’s income and outgoings are likely to evolve, including real-life events: a visa renewal fee, family reunification, the housing shift from benefits into work.
4. A deterministic responsibility gate. Fixed, written-down rules grounded in FCA requirements run before anything else. These rules are ordinary logic, not machine learning. They never change themselves, each cites its regulatory source, and every run is audited.
5. Learning that explains itself. Merit’s models offer an advisory view with the reasons attached: which evidence pushed the view, and by how much. When a caseworker disagrees with Merit, that disagreement is treasured training material.
6. A human decides, and Merit helps them say it properly. The caseworker sees everything and makes the decision. Merit then drafts the decision letter by written rules, checks it line by line against FCA and UK GDPR requirements.

AI models in finance, with context and human oversight

Merit uses AI to organise verified evidence, alternative financial data and personal context into an explainable view of current affordability and future trajectory. Future potential means an evidence-based projection of how income, costs and opportunities may change; it is kept separate from today’s financial position, carries uncertainty, and is never treated as a promise or an automated credit score. A qualified person weighs that context alongside policy and regulatory requirements and makes the financial decision.

What Merit will enable, and the change in the world

1. Now: inclusive lending, proven in hard mode. RefuAid’s lending is the proving ground, deliberately difficult: a population conventional systems serve worst, with real money and real lives.
2. Next: a shared capability for mission-locked lenders. Community lenders, CDFIs and social finance bodies face the same gap Merit closes.
3. Then: from credit to resources. Creditworthiness is one narrow slice of the real question: what does this person need to fulfil the future they are working toward?
4. Throughout: public knowledge, not proprietary advantage. What Merit learns about fairly recognising interrupted lives should change how the field works, not sit in a vault.

Merit exists to prove that the way one person recognises another’s potential can be learned, honoured and scaled, under a constitution that makes it legally impossible to betray the people it serves.