Build vs buy: an odds API

A working odds pipeline is not one thing. It is collection (reliably pulling many books, including the hard ones), normalisation (mapping every book and market to one event and selection model), and — if you need matched or arbitrage output — matching, rating and gating on top.

Each layer is a real system with its own failure modes. And none of them is build-once: bookmakers change markets, add and drop offers, and alter how they present data, continuously.

A prototype pulling five books on a quiet afternoon proves nothing about production. Production is the long tail: the book that changes its markup, the fixture that two books name differently, the source that quietly changes how it presents its markets, the bank-holiday card with three times the volume.

This maintenance is permanent and unglamorous. It is also where most build-it-yourself odds projects quietly fail — not at launch, but six months in, when the person who built it has moved on and the coverage is silently rotting.

Building is the right call when the pipeline is your edge. If your product's differentiation is a unique matching model, a proprietary rating, or coverage no one else has and you can sustain — then owning the stack is owning your moat, and a bought feed would constrain you.

It is also reasonable when your needs are genuinely tiny and static: a handful of books, one market, no matching. Just be honest about whether that will stay true.

Buying wins when odds data is an input to your product, not the product itself. If your value is the oddsmatcher UX, the content, the trading model, or the brand — then the months you would spend building and maintaining a pipeline are months not spent on the thing that actually differentiates you.

Buying also wins on the hard parts specifically: covering the difficult books (bet365 is the usual example), maintaining normalisation as the market shifts, and doing matched output well. These are exactly the areas where a specialist feed has structural advantages you cannot match for one product.

It is rarely all-or-nothing. A common, sensible shape is to buy the parts that are hard and undifferentiated — collection, normalisation, even matching — and build the thin layer on top that is genuinely yours.

A feed that offers both a raw and a matched product supports this: buy raw and build your own matching if that is your edge, or buy matched and build only your product surface. You move the build/buy line to exactly where your differentiation starts.