The £10,000 decision: how we built Seedr's AML threshold

The question nobody asks until it matters

When you build a payments product, you inherit other people's paranoia. The Financial Conduct Authority doesn't exist to annoy founders. It exists because financial crime is real, and the cost of getting it wrong is ruinous. We knew that going in.

But knowing and acting are different things. Early on, we could have set our AML threshold anywhere. £50,000. £5,000. We could have waved through every payout and hoped Stripe's backend would catch anything dodgy. Most indie platforms do something close to that. They bundle compliance into "we're using Stripe, so we're fine." That's not wrong, exactly. But it's not us.

The FCA roadmap we published commits MRVL to Payment Institution authorisation by 2028. That's not marketing speak. It means every schema decision, every payout, every audit trail has to be defensible before a regulator. From day one, Seedr was built to that standard, not retrofitted to it later.

Where £10,000 came from

Our threshold isn't arbitrary. In January, I sat down with our compliance advisor and we mapped out creator behaviour. What does a typical tip volume look like? How often does a single creator hit four figures in a week? What's the noise floor where manual review becomes actually useful, rather than just theatre?

Faith creators, who've been our strongest cohort from launch, typically see tipping patterns that cluster between £50 and £500 per stream. A really successful Sunday service might pull £2,000 across 200 supporters. Giggl comedians and Foundr makers follow similar curves. Very few hit £10,000 in a payout window without a specific event: a conference, a product launch, a moment that breaks their usual rhythm.

That spike is exactly when you want a human eye on the transaction. Not because the creator is suspicious, but because the behaviour is statistically unusual. Manual review at that threshold catches the outliers without creating friction for 99% of creators. We don't charge extra for it. We just flag it, review it, and process it if it's clean.

The integer pence trail matters too. Every transaction, every fee calculation, every payout sits in our database as an audit-ready record. When (not if) a regulator asks us to justify a payout, we can. That's what FCA-ready means in practice: you don't scramble. You show them the data.

The creator who nearly made us second-guess

That Streamr creator I mentioned, the one asking about £8,500? I called her back instead of templating a response. She'd been fundraising for her church's new community space. Over six weeks, her audience had tipped £34,000 total. But she was asking about one specific payout that sat just below our threshold.

She wasn't angry about the system. She was curious. She'd actually thought through the compliance angle herself. "I get it," she said. "You're protecting me as much as you're protecting you." That conversation shifted how I talk about this internally. The threshold isn't a restriction. It's a guardrail. It says: we're taking money movement seriously enough that we won't automate away the human judgment.

We still processed her full amount. The manual review took an hour. She got her payout on Monday, same as everyone else. But we knew the source was legitimate, the use case was straightforward, and the money trail was clean. That's the workflow the threshold protects.

Why this matters for scaling

Here's what I've learned building a payments platform inside a creator ecosystem: compliance is boring until it isn't, and then it's catastrophic. The platforms that survive are the ones that bake it in early, when the cost of changing it is low and the number of creators affected is manageable.

Every time we add a new creator app to MRVL (Streamr, Giggl, Foundr), the tipping volume grows. Our AML threshold doesn't change. But our processes have to evolve. We've documented exactly how a manual review works, who can trigger it, and how we escalate if something looks genuinely suspicious. That playbook exists before we need it, not after.

The schema decision that underpins this is living in supabase/functions/_shared/fees.ts. It's boring code. It calculates platform fees (5% standalone, 1.5% for Foundr Free creators, 1% for Foundr Pro), and it's the single source of truth for how money moves. No surprises. No shadow calculations. When an audit happens, that file is exhibit A.

Trust is the product

I could write a blog post about Seedr and talk about weekly Monday payouts, no-account tipping for fans, creator dashboards, or the three lines of code it takes to embed SeedrButton into a MRVL app. All of that's true. All of it matters.

But the thing that separates us from a hundred other tipping platforms is that we're not hiding behind regulatory theatre. We're building toward actual authorisation. That means our creators can tell their audiences: the money you send goes through a system that's designed to be compliant, not just compliant-adjacent. That's a differentiator that compounds over time.

The £10,000 threshold is a line in the sand. It says: beyond this point, a person looks at it. Not a process. Not a machine. A person. That's expensive and slow. It's also the thing that makes trust real.