FMTT · Love1Ticket
for Uzair
August 2026
A trading business, run by agents.
We trade tickets to live events. We are replacing every spreadsheet, group chat and manual job in that business with one system — operated by AI agents, watched by people. Most of it is already running.
club email → read by AI → stock → sale lands → auto-assigned → need to buy → priced → listed → delivered on WhatsApp
The business
FMTT trades tickets for football across the UK and Europe, NFL, major tournaments, darts and live events — and is now moving into the American market: any live event that's profitable, we work it.
Inventory originates from an estate of memberships and accounts across more than 25 clubs and events. Sales go out through every major marketplace — there is no single "sell button" in this industry, which is exactly why the system matters. A small team runs the whole operation. It's a real business with real volume; the numbers come when you're inside.
The problem with this industry is that everyone runs it on spreadsheets, inboxes and memory. Buying confirmations arrive as emails. Sales arrive as emails. Passes get forwarded by hand. Prices get checked by eye. Every one of those is a flow a machine should carry — and that's the whole thesis.
The end state
An agent on every flow. Humans on judgement only. Ninety percent of the day-to-day automated, and the whole book marked to market, live.
Stock knows where it came from, down to the email that proves it. Every sale finds its stock by itself, or tells us exactly what to go and buy. Prices track the market across every platform. Passes reach buyers without a person forwarding anything. The desk shows one honest picture of the whole business at any second, on a phone.
Already running
This isn't a pitch deck of intentions. These are systems in production today, used by the team daily — built in months, largely by directing AI agents rather than typing every line.
LIVE
The mail estate
Over a thousand mailboxes across the club estate feed one gateway in real time — webhook-driven, durable-queued, idempotent. Every email is stored, classified and searchable. Passwords, codes and card numbers are stripped the moment mail arrives — nothing that grants account access ever reaches the database, by design, from day one.
LIVE
AI reading every email
A Claude-based extraction agent turns club mail into typed data: purchase confirmations, ticket deliveries, ballot results, renewals, fixture changes — each with per-session line items and confidence, never guessed, always traceable to the email it came from. Per-sender templates handle the highest-volume formats to the penny.
LIVE
Inventory that builds itself
Stock appears in the book minutes after a confirmation email lands — batch by batch, seat by seat, with cost, currency and provenance. Nobody types stock in. A typed entry is allowed, but a proven email always outranks it.
LIVE
The sales hub
Sale notifications from seven marketplaces parse straight into one ledger — order, quantity, money in its true currency (the desk never converts silently), event, buyer. Cancellations release stock by rule. The business is mid-shift to US dollars, and the parsing, ledger and FX handling moved with it this week.
LIVE
The allocation flow
Every sale is matched to stock automatically — but only when the match is unambiguous; anything uncertain waits for a human. Where there's no stock, the shortfall becomes a live NEED TO BUY list, derived fresh every look and never stored, so it clears itself the moment stock arrives.
LIVE
The desk
A mobile-first console — the book, orders, inventory, market, mail, the estate, risk, season, intel — with per-account access control and a design rule the whole system obeys: unknown renders as an em-dash, never a fabricated zero.
BUILT
WhatsApp delivery agent
A conversational Claude agent on WhatsApp's official API that hands passes to buyers, answers them, records receipt — and is hard-scoped to discuss only the orders attached to the number that's talking. Deployed, switches on the moment the business number goes live.
The arc from here
The order of battle is deliberate — each stage feeds the next:
1 close the delivery loop — pass in from mail, pass out on WhatsApp
2 one event identity — every system naming the same game the same way
3 pricing — see the market's floor, propose our price, with reasons
4 listing — one pool of stock live on every platform at that price
5 buying — the need list turned into per-account buy plans
6 the American book — the same machine, pointed at US events
The lanes — in detail
Each lane below is a real piece of the arc, waiting for an owner. For each: what exists today, what's genuinely needed, what you would build, and a first ship that's achievable in a few focused sessions — live, in production, used by the team.
Dispatch automation
Close the loop the business feels most: pass arrives by email, buyer receives it on WhatsApp, receipt confirmed, order closed — no human forwarding anything.
- TODAY
- Passes and mobile tickets already land in the mail estate as attachments and wallet links, parsed and typed. The WhatsApp agent is deployed with media-upload plumbing ready. The allocation flow already knows which order each sale belongs to.
- NEEDED
- The middle of the pipe: matching an arriving pass to its order (which delivery email belongs to which sale), a send pipeline that uploads the pass and hands it to the right buyer's number, a receipt-confirmation loop that closes the order, and an exceptions queue for everything that doesn't match cleanly — wrong name on the pass, partial deliveries, replacements.
- YOU'D BUILD
- The pass↔order matcher (same discipline as allocation: auto only when unambiguous), the send worker with full audit trail, the order-closing state machine, and the desk view where a human clears the exceptions in one look.
- FIRST SHIP
- One club's mobile passes flowing mail → WhatsApp with a human approving each send from a queue. Automation of the approval comes after the matcher has earned trust — that's how everything here ships.
Event identity
The quiet lane that unlocks three others: one canonical answer to "which game is this?" across every system.
- TODAY
- A marketplace names a game one way, the club's confirmation another, the market data a third. Matching is done by date-anchored name comparison — good, conservative, and tested, but it means fixtures on the orders page can't yet link to their market pages, and every join between systems re-derives the same answer.
- NEEDED
- An event catalogue: one id per real-world event, an alias table that learns every vocabulary that names it, and a tiny review surface where a human resolves an ambiguous name once and the system remembers forever.
- YOU'D BUILD
- The catalogue schema, the alias learner (seeded from the matcher's existing judgements), the resolver UI, and then the satisfying part — walking through the codebase replacing per-system name matching with one lookup, watching links light up across the desk.
- FIRST SHIP
- Sales and stock joined through the catalogue for one competition, with the orders page's fixture names becoming real links to their market pages.
Pricing agents
The desk already sees the market's floor per event. The next step is an agent that proposes what our price should be — with its reasoning shown, never a silent repricer.
- TODAY
- Market pages show per-event competitor floors with strict honesty gates (a stale capture is aged visibly, absence of data reads as unknown, never as "we're the floor"). Our own listings' prices are captured and shown beside them.
- NEEDED
- The pricing rules encoded — margin guards against the batch's real cost, never-below floors, event-proximity curves, how to react when a competitor undercuts — and a proposal loop: the agent watches, computes, and posts a suggested change with the reason ("floor moved from £212 to £189, we're now 4th cheapest; suggest £195, margin holds at 38%").
- YOU'D BUILD
- The rules engine (pure, tested against synthetic market states), the proposal queue on the desk with one-tap approve, and the audit trail that records every proposal, decision and outcome — the dataset that eventually justifies letting it act alone within bounds.
- FIRST SHIP
- Reprice proposals for one competition's events, human-approved, with a week's audit trail proving the agent's judgement against what the team actually chose.
Listing automation
One pool of stock, live on every platform, without a person creating the same listing five times — and the moment one platform sells, everywhere else delists.
- TODAY
- Stock lives at batch and seat grain with split rules understood (a pair, a run of four). Listings are made by hand on each marketplace; the system deliberately holds no false picture of them. Official API access exists for the first target platform.
- NEEDED
- A listing state per channel per batch (what's live where, at what price and split), create/update/delist through official APIs only, and the crown jewel: cross-delist on sale — a sale landing from one platform pulls the same seats everywhere else within a minute, which is the single biggest risk-killer in this industry (double-selling the same seat).
- YOU'D BUILD
- The channel-listing model, the first platform integration end to end, and the delist-on-sale worker driven by the sales hub's ledger — idempotent, audited, alarmed when a delist fails.
- FIRST SHIP
- One platform: stock listed from the desk, and automatic delisting there when the sale arrives from anywhere. Each further platform is then a template, not a rebuild.
Buying agents
The NEED TO BUY list already says what we owe the market. The next agent turns it into a plan: which account buys it, by when, at what price.
- TODAY
- The shortfall per event is derived live — sold minus held, past events ignored by rule. The estate register knows the accounts, their memberships and their status.
- NEEDED
- The eligibility join — which accounts can buy for a given event (membership tier, sale windows, per-account limits), turned into per-account shopping lists with deadlines, tracked to execution: when the confirmation email lands, the buy plan ticks itself off because the stock pipeline already read it.
- YOU'D BUILD
- The eligibility model over the estate register, the plan generator, and the tracking loop that closes plans from the same mail pipeline that builds stock — the whole thing self-verifying because both ends are already machine-read.
- FIRST SHIP
- Buy plans generated for one club's on-sales, with the plan-to-confirmation loop closing automatically.
The American book
The business is moving to USD and US events. Same machine, new territory — the closest thing here to a greenfield lane.
- TODAY
- Banking has moved to dollars. The sale parsers, ledger and desk already handle USD as first-class money, with indicative FX at the ECB daily rate shown as a labelled estimate, never written into the books.
- NEEDED
- US marketplace integrations (sales feeds first, listings later), the US event taxonomy — NFL, NBA, MLB, concerts, with venues, timezones and date conventions that differ from football — and dollar-native views of the book as USD becomes the majority currency.
- YOU'D BUILD
- The first US sales feed end to end into the ledger, the event-taxonomy work in tandem with the event-identity lane, and the desk's dollar-first presentation as the volume shifts.
- FIRST SHIP
- One US platform's sales landing in the ledger with the same honesty guarantees the existing seven have — parsed to the penny, in dollars, cancellations handled.
How a side project works here
- Pick a lane. Talk it through with Adam — you get the full context that doesn't belong on a public page: the numbers, the platforms, the estate, the real constraints.
- Own a scoped slice. Every lane above is cut so its first ship is a few focused sessions, not a quarter. You own it end to end: design, build, tests, deploy.
- Build by directing agents. This system is largely built by pointing Claude-based agents at well-defined problems and reviewing hard. You'll write code too — but the leverage skill you'll practise is specifying, directing and verifying agents. That's the working style this whole company is betting on.
- Ship behind the guard rails. Everything lands with tests, behind the honesty rules, with auto-action only where the system is certain and a human queue for everything else. Your work goes live against real stock, real sales, real buyers — safely.
- Iterate weekly. The team uses what you ship within days and tells you what's wrong with it. There is no better feedback loop.
How we build
- Honesty over completeness. A number the system can't prove is shown as unknown with the reason — never a plausible guess. The books never lie to look finished.
- Security is absolute. Credentials are stripped at ingest, sealed where they must exist, and can never reach the user-facing desk by any route. These rules have no exceptions and no shortcuts.
- Guards fire, never suppressed. Arithmetic, uniqueness and timing checks run on everything; an exception is data to resolve, not a log line to ignore.
- Everything ships tested. Two-thousand-plus tests across the system; untested code isn't done. Small commits, deployed the same day, used by the team the same week.
- Agents do the work; people make the calls. The system auto-acts only where it's certain, and queues the judgement calls for a named human. That line is designed, not accidental.
The stack, so you know what you'd touch: TypeScript end to end, Postgres, Next.js on Vercel, a Node gateway, the Claude API for every agent, official marketplace APIs only. Nothing exotic — the ambition is in what it does, not the tools.
The invitation
Come build this with us as a side project — pick a lane, own it end to end, and ship it into a production system a real trading team uses every day.
You'd be early on something with a clear destination — a fully agent-operated trading desk in an industry still run on spreadsheets — learning the one skill that's about to matter everywhere: making AI agents do real, verified, production work. And you'd be doing it on the journey with us, not on a toy.
Adam · FMTT / Love1Ticket · August 2026
Talk to Adam about where you'd want to start — the full detail of the business and the system opens up from inside.