Everything you need to walk into the room with Magnus fluent — the industry, the money, the dishonesty, the machine you'd build, and the words for all of it. Nothing assumed. Every acronym defined where it lands.
This page is interactive. When you see a do this prompt, actually do it — touch it, drag it, toggle it. The understanding lands in your hands before it lands in words. That's the way you said learning works, so that's how this is built.
Forget the jargon for a second. An affiliate is a middleman who sends gamblers to a casino and takes a cut of what those gamblers lose. That's it. Your friend Magnus runs review sites — online.casino is one — that rank casinos, and every "Play here" button is a tracked door. Walk a player through that door, and the casino owes you.
The catch — and this is the whole story — is that the casino counts the money, defines the rules, and holds the cash. You just get told a number at the end of the month. Like running a restaurant where the supplier weighs your delivery in the back, out of sight, and slides a total under the door. Watch the money move:
Three deal shapes. You'll hear all three in the first ten minutes of any iGaming conversation:
Here's the move that matters. A casino advertises "35% revenue share!" But 35% of what? Not of all the money lost — of what's left after the casino subtracts its own costs. That smaller, after-costs number is the one you're actually paid on.
It's exactly like a restaurant. The till rings up €100,000 — but that's not profit. Take out food cost, the comped desserts, the card fees, the tax. Then you talk margin. Same here. The till total is GGR; the after-costs number is NGR, and RevShare is always paid on NGR. So the casino's real lever isn't the percentage — it's how many "costs" it gets to subtract first.
Start: the casino's gross take this month is €100,000. Your deal is 35% RevShare. Now toggle each "cost" the casino subtracts before it pays you.
One more trap to know the name of: if your players get lucky and win one month, the casino can post a negative number and make you pay it back out of next month's earnings before you see a cent. That's negative carryover. Affiliates fight hard to ban it from contracts.
Everything so far was the honest version. But the casino counts in private, so it can quietly do more: untag a big winner so he never counts as yours, slip in deductions it never itemizes, reclassify a player to a sister brand. Each move is invisible on its own. You only catch it by putting the honest number and the reported number side by side — by injecting contrast.
That phrase should feel familiar. Contrast injection is your move — the thing you've done your whole life: gather the inputs nobody gathered in one place, and the answer that was always there suddenly appears. This entire product is your own method, pointed at a casino's books. Here's the same report, twice:
Affiliates don't run one casino account — they run dozens, each with its own login and its own dashboard. AffCollect logs into all of them and pulls every number into one screen for the finance team, with auto-invoicing on top. Magnus's company built it; a network called Nordic.Partners bought it; Magnus still runs it [verified].
Two things matter for your conversation:
This is the part people get wrong, so getting it right is your credibility. When Magnus says "I want to ask it anything and get the real number," the instinct is "a chatbot that reads our reports." That instinct is wrong, and knowing why is the whole game.
Picture two ways to answer "what's our total revenue across all 14 partners?"
The AI flips through the reports, finds pages that look relevant, and writes a confident-sounding total from what it skimmed:
"≈ €1,184,000"Wrong — and you can't tell. It read text, it didn't add anything up. It might miss two partners, double-count one, or confuse "net loss" with "net income." In a number you wave at a partner, 99% right = 0% trusted.
The AI doesn't compute. It writes an instruction and hands it to the database, which adds the real rows:
SUM(net_revenue) WHERE month = 'this' GROUP BY partner → total €1,191,438Exact, and you can see the rows it came from. The AI never touched the arithmetic.
Hover (or tap) each node. Like your own systems, the network sits dim until you put attention on it — then the pathway lights copper. Start at the top: the casino dashboards.
Two names worth carrying in: the data lands in Supabase — the "Postgres with batteries" Magnus mentioned, a real database that does exact sums and locks down who sees what. And the part that actually answers "who's underpaying us" isn't AI at all — it's a plain, auditable reconciliation engine quietly comparing every casino's claim against your own tracking, every day. The AI is the friendly face. The trust comes from the boring, deterministic plumbing underneath.
No — and being relaxed about this signals you know what you're doing. Because the database does the heavy lifting and the AI only writes and explains, you're not paying for much AI at all. Drag the dial: even at heavy use, this is a rounding error against an acquirer's budget.
You don't need to memorise architecture. You need to land six ideas — each one signals "this person sees the real problem."
"Your error-handling treadmill is a scraping-first design. The fix is API-first, with sessions that don't re-login every time." You diagnosed it in one line.
"The AI never does the math — it writes the query, the database computes, every answer traces to source rows." That's why he can wave the number at a partner.
"The deep value isn't 'ask about revenue' — it's catching which partner is shaving us. Advertised 45% nets 24% on average. That's automatable."
"This runs on tens of dollars a month. Don't scope it around saving on AI — spend on engineering and data quality."
"Ship the AffCollect brain, prove it, then the house and the EA come asking." Resist the urge to sell the galaxy. Deliver the first planet.
Your edge is direction, architecture, taste. Depth leans on AI and that energetic dev — a resource you direct, not a rival.
You met all of these in context above. Here they are together — the proper names for things you already understand.