By Daniel Davis, Chief Revenue Officer, Connamara Technologies.
Ask most people how a new trading venue should approach its technology, and the answer comes back as a binary: build it yourself, or buy something off the shelf. Build, and you own everything, but you spend years getting to your first trade. Buy, and you launch fast, but you’re stuck with someone else’s product decisions for the life of the platform.
It’s a comfortable way to frame the choice, but for prediction markets, it’s mostly wrong. Very few operators sit at either pole. The decision that actually matters happens in the middle ground, and it’s really a decision about risk: how much integration risk you’re willing to carry in exchange for the freedom to create a defensible moat.
Two approaches live in that middle ground. They’re easy to mix up because they share a vocabulary and a fair amount of architecture. But the gap between them shapes how fast a new venue gets to market, how big a team is needed to run the platform, and whether the venue still stands apart from its competitors a year after launch. So it’s worth being precise about what separates them.
What a modular approach actually means
A modular approach is mix-and-match. Instead of buying a single platform or building one from nothing, you assemble the venue from best-in-class components: an exchange matching engine, a clearing system, a surveillance tool, maybe with a few pieces built in-house. Each component gets chosen on its own merits.
That approach makes sense. Pick the strongest option for each function rather than accepting whatever a single vendor happens to bundle together, and spread the platform across several providers so that you’re not dependent on any one of them. Because you’ve selected the components rather than inherited them, there’s room to differentiate.
However, the true cost usually only shows up later. Every component has to talk to every other component, and those connections are yours to build and maintain. When something breaks, working out which system is at fault can take longer than the fix itself. Multiple vendors mean multiple support relationships, multiple release cycles, and multiple roadmaps that don’t coordinate with one another. The flexibility that looked so attractive during selection can turn into an integration and management nightmare once your venue is live.
Where extensible modular changes the equation
An extensible modular approach keeps what’s good about modularity while removing those future costs. The core functions of the venue – exchange, clearing, and surveillance – come integrated, from a single source, already built to work together. As the venue operator, you’re not stitching those pieces into each other, and you’re not worried when the seams get tested under load.
What makes it modular rather than just all-in-one is optionality at the core and openness at the edges. You can take the exchange engine on its own, or the exchange plus clearing, or all three together, depending on what your venue needs and what you’ve already got. And around that integrated core sits a set of open, developer-friendly APIs. Your differentiation comes from building on top of a proven foundation, not from gluing separate foundations together.
It comes down to where the engineering effort goes. In a purely modular build, a lot of it goes into making the components cooperate. In an extensible modular build, that cooperation is a given, and the engineering goes into the things your customers actually see and care about: the contract types, the trading experience, the data products, the commercial model. The plumbing is sorted, so you can spend your finite resources on what makes your venue distinctive.
Why the distinction bites harder in prediction markets
None of this is unique to prediction markets. Operators in any asset class weigh integration against differentiation. But prediction markets rely on that trade-off in ways that make the choice of architecture unusually consequential.
Start with the pace of product change. A prediction market lives or dies by how quickly it can design, list, and resolve contracts around real-world events, and those events don’t wait for a lengthy development cycle. A platform that needs structural rework every time a new contract type or a new settlement rule comes along will always be a step behind the news. Rapid contract launch has to be built into the architecture, not run as a project each time.
Then there’s the risk profile. Event contracts often have correlated outcomes, and a platform that can’t calculate collateral relief across correlated positions either leaves the operator exposed to risk it hasn’t priced or forces customers to over-collateralise. That logic has to live in the core, working hand-in-hand with clearing, in real time. It’s exactly the kind of requirement that suffers when exchange and clearing are sourced as separate systems that negotiate with each other across an integration layer.
And prediction markets don’t keep market hours. Contracts tied to elections, sporting fixtures, and global events draw their heaviest activity at unpredictable moments, often well outside any conventional trading day. Availability expectations look less like an equities exchange and more like an always-on digital venue: no maintenance windows, no scheduled downtime, just continuous operation. Holding that standard together across a set of loosely coupled components is a lot harder than holding it across an integrated core.
Put those three pressures together – constant product change, correlated-outcome risk, and round-the-clock availability – and the integration fragility that comes with a purely modular approach stops being a manageable overhead and starts threatening the things that make the venue work in the first place.
The question worth asking
The useful way to frame the decision isn’t “build or buy”, and it isn’t even “modular or extensible modular” in the abstract.
It’s a question about where you want to focus your engineering efforts.
Spend them on making separately chosen components cooperate, and that’s effort you’re not putting into the contract design and customer experience that actually set a prediction market apart. Keep the core integrated and the extension points open, and the same effort goes into the things your competitors will struggle to copy.
That’s the thinking behind EP3 by Connamara Technologies. It brings exchange, clearing, and surveillance together as an integrated platform, with the option to take the exchange on its own or add clearing and surveillance as you need them, plus a set of extensible APIs for building differentiated functionality on top. EP3 runs 24/7/365 with no maintenance-window shutdowns, and it’s in production across a range of regulated venues, among them Interactive Brokers’ ForecastEx, the CFTC-regulated prediction exchange (DCM) and clearinghouse (DCO), Railbird, the DCM-licensed prediction market acquired by DraftKings in 2025, and other well-known names. As one of the most widely adopted technologies among US prediction markets, EP3 shows plainly enough through its production record that the extensible modular model holds up in live, regulated trading environments.
In a market that’s still taking shape, the strongest place to be is to live on already-proven infrastructure, with the freedom to keep differentiating as the sector matures. Get the architecture right at the outset, and that’s the position you’re in.
To discuss how EP3 fits your prediction market requirements, visit www.connamara.tech.
Subscribe to our newsletter
