PreCredits is a B2B2C tokenised store-credit platform I built and run end-to-end — concept and regulatory model through to product, launch and the day-to-day of operating a small Australian fintech. The shorter story below is about how the build happened: the regulatory frame I stood it up on, the AI delivery method (Claude + Lovable + Supabase) that did most of the engineering, and what eighteen months of that has actually taught me.
The observation that started PreCredits was simple. Australian merchants issue an enormous amount of store credit every year — refunds for change-of-mind returns, gift cards, loyalty balances, goodwill credits when something goes wrong. To the merchant, it's a liability sitting on the balance sheet. To the customer, it's money that only spends in one place, often forgotten, frequently lost, and almost never as useful as cash.
Both sides lose value in the gap. Merchants carry breakage, but pay for the working-capital and reconciliation overhead. Customers hold balances they don't redeem. And the broader ecosystem treats store credit as a marketing afterthought rather than the liquid, programmable instrument it could be.
The product question that followed was specific: what would store credit look like if it were tokenised, portable across a defined merchant network, and built on payment-grade rails? Not crypto for the sake of it. A boring, regulated, ledger-backed token that behaves like money inside a closed loop, and like data everywhere else. That's the idea PreCredits has been working on since 2019.
Store credit isn't a winner-take-all market — it's a long tail of liability that every merchant has, that no one is incentivised to fix, and that benefits from a neutral platform sitting between the merchant and their customer. That made it tractable for a small-team, AI-accelerated build in a way that, say, payments or lending wouldn't have been.
Anything that looks like money in Australia attracts attention from a small set of regulators, and the structure you choose at the start determines what you can ship later. PreCredits had to be defensible to four overlapping regimes from day one: ASIC for financial-product framing, AUSTRAC for AML/CTF obligations as they apply to the product, ACCC for consumer-protection and gift-card rules, and APRA for any prudential exposure depending on how value is held.
The work that mattered most was the boring part of fintech: a clear written position on which exemptions and carve-outs applied, where the platform sat in the payments value chain, who held funds at any given moment, what consumer disclosures were required, and what record-keeping and reporting obligations applied to the product. None of that is glamorous, and all of it has to be in writing before a regulator asks.
PreCredits has filed three consecutive years of R&D Tax Incentive claims with AusIndustry, all certified. The discipline of writing those claims — articulating the technical uncertainty, the systematic experimentation, and the new knowledge produced — has been a useful operating habit independent of the rebate. It forces a real engineering log: what you tried, what failed, what you learned, what you'd do differently. Most early-stage products don't keep that record. Once you have it, the rebate is a side benefit; the artefact itself is the thing that compounds.
"Compliant by design" is a phrase founders use to sound serious. The honest version is that you write a thoughtful position paper, you stay in regular conversation with advisers who do this for a living, and you update the design when the law moves. The work is continuous, not a one-time tick.
The honest framing for the build: a regulated fintech in 2019 would have needed a much larger engineering team and a meaningful seed round to get to a working product. By the time the AI tools matured to the point I'd trust them with production code — late 2024 onward — the same work could be done by a smaller core team with the right scaffolding and an advisory layer for the regulatory and specialist gaps. PreCredits in its current form is the result of that shift.
The delivery method, in plain terms:
Architecture conversations, schema design, code review, refactors, edge-case enumeration, and the kind of multi-file changes that used to need a senior engineer's afternoon. The conversation interface matters: the value isn't the autocomplete, it's the dialogue.
React/Tailwind front-ends shipped fast. UI iteration loops collapsed from a day to an hour. The honest test: does the output need a human-engineer pass before it goes to production? Increasingly, no.
Postgres, auth, row-level security, realtime, storage, edge functions. Every layer a small fintech needs, with the audit-grade primitives a regulator expects. RLS policies are the unglamorous bedrock.
Edge deploy, DNS & security, payments. Best-in-class infrastructure rented at marginal cost. The right answer for a fintech is almost always to rent the boring layer and own the differentiated one.
The interesting thing about that stack isn't any individual tool — it's the lean-team shape it enables. A regulated product that would have taken a much larger engineering team and a meaningful seed round in 2019 can now be stood up by a small core team with an advisory layer and a willingness to write the regulatory position paper themselves. Whether that's good or bad for the broader fintech ecosystem is a separate question. It's certainly the world I'm operating in.
PreCredits sits between merchants and their customers. Merchants integrate the platform to issue, manage and reconcile store credit as a tokenised balance — auditable, programmable, and durable across the customer's relationship with the brand. Customers see a balance that behaves the way they expect money to behave: it's there, it's clear, and it doesn't disappear into a forgotten gift-card folder.
The B2B2C shape was a deliberate choice. A pure-B2C consumer wallet would have meant fighting the giants — Apple Pay, Google Pay, the bank apps — for surface area on the phone. A pure-B2B accounting tool would have been useful but invisible to the end customer. The wedge that made sense was sitting between the two: merchant-facing for distribution, customer-facing for the experience that proves the model.
The product surface is intentionally narrow. Issue credit. Hold credit. Spend credit inside the loop. Reconcile credit on the merchant side. Report credit for compliance and accounting. Every other feature — partnerships, analytics, marketing tooling — is an extension of those five primitives, not a replacement for them. The temptation to add scope is the single biggest risk in a small-team fintech, and the discipline to resist it is the single biggest leverage.
The lessons below are the ones I'd want a founder starting today to hear. Not novel for novelty's sake — practical, specific to running a regulated, AI-accelerated, small-team product.
Write it before you ship. Update it when the product changes. It's the thing a regulator, an investor, and a future you will all eventually read. The product can change weekly; the position paper is the spine.
Postgres on Supabase, payments on Stripe, edge on Vercel, DNS and security on Cloudflare. None of that is the moat. The moat is the position paper, the merchant relationships, and the specific way credit is modelled in the ledger. Spend your hours there.
The biggest unlock isn't completing your next line — it's the architecture conversation, the schema review, the "is there a simpler way to model this?" question. The conversation surface is the leverage. The autocomplete is a side effect.
The R&D Tax Incentive forces this — what you tried, why, what worked, what didn't. Most early-stage teams don't keep one and regret it. The artefact is more valuable than the rebate.
Schema decisions, ledger design, regulatory framing, terms of service. Get those right and the rest is reversible. Get those wrong and every subsequent decision is constrained by them.
A longer post — "What 18 months of an AI-built fintech taught me" — is in draft. It goes deeper on the specific Claude + Lovable + Supabase workflow, the things that still need a human, and where the AI-first delivery method meaningfully changes what a small-team fintech can ship. Find it on the main site when it goes live.
If you're starting an AI-accelerated, regulated, small-team product in Australia and you'd like a coffee, the door is open. PreCredits is operating, R&D-certified, and I'm happy to compare notes on the regulatory frame, the build stack, or the mechanics of running a fintech at small-team scale. Email's the easiest way in: sam.keil@gmail.com.