Week one is usually discovery: map order flow, venues, risk checks and regulatory obligations. By week six, a venue connection can often be in integration testing, while a proprietary matching engine is still being specified. Those are typical timelines, not delivery guarantees. The choice is whether to operate the mechanism that creates executions, or connect to one operated by an exchange, multilateral trading facility (MTF) or liquidity provider.
The decision is about control, not just latency
A matching engine receives orders, validates them, assigns sequence, applies matching rules and publishes executions. A central limit order book might use price-time priority. Other models use pro-rata allocation, auctions or request-for-quote workflows. The rules determine which orders can interact and how partial fills, cancellations and rejects behave.
Using a venue means accepting its rulebook and liquidity, then building the systems around access: order entry, market data, pre-trade risk, routing, drop copy and reconciliation. A venue does not remove integration work. It moves the boundary. The firm still needs to control what leaves its systems, understand what came back and recover cleanly after a disconnect.
Owning an engine can suit a broker or prop firm that needs bespoke matching rules, control over an internal market or a venue-independent execution layer. It does not create liquidity by itself. A thin book can mean wide spreads and unpredictable fills. Firms handling client orders also need to establish whether internalisation is permitted, how conflicts are managed and how execution obligations apply in each jurisdiction.
Compare the operating models
The figures below are typical market planning ranges, not quotes, service-level commitments or audited results. Latency depends on geography, network path, order type and where the measurement starts.
| Factor | Own matching engine, typical range | Venue connection, typical range | What the range means |
|---|---|---|---|
| Initial production delivery | 6-18 months | 4-12 weeks for a basic connection | Custom rules, controls and testing extend an engine build. |
| Order-to-acknowledgement latency | 0.1-2 ms inside a co-located matching stack | 1-10 ms co-located; 10-50 ms on a more distant route | These are typical technical ranges, not end-to-end client latency. |
| Core engineering and operations team | 5-12 people | 2-5 people for a small venue estate | Production coverage, change control and support affect staffing. |
| Ongoing resilience work | Engine, state recovery and matching tests | Gateway, session recovery and venue failover tests | Both models need tested recovery, monitoring and reconciliation. |
Ranges vary with asset class, order flow, venue count and service model. A firm should compare total operating cost, not only software licensing.
An engine needs continuous development, release management, monitoring and recovery procedures. Venue access can add connectivity, market-data and transaction fees, as well as exposure to venue outages, rule changes and capacity limits.
A hybrid model is common: route orders to external venues while crossing eligible flow internally under defined controls. It can reduce dependence on one source of liquidity, but adds routing decisions and more complex execution reporting. For broader trading infrastructure coverage, compare how platforms handle these boundaries, not just their headline feature lists.
Build the decision in six steps
Start with the required execution behaviour, then select the smallest operating model that can deliver it. A useful sequence is:
- Map every order path. Record where orders originate, which checks run, where they are routed and what produces the execution report. Include rejects, cancels, amendments and partial fills.
- Write down the matching rules. Specify priority, eligible participants, order types, self-trade prevention and auction behaviour. If the rules are not distinctive, a proprietary engine may add little value.
- Measure liquidity needs. Identify the instruments, trading hours and expected activity. Check venue depth and fee schedules against the firm's actual order profile. Do not treat a venue's advertised capacity as guaranteed liquidity.
- Set risk boundaries. Decide where credit, position, price-band and kill-switch checks run. Define what happens if a risk service is unavailable. A reject should be predictable; a bypass should not be an informal recovery plan.
- Model failure and recovery. Test lost sessions, stale market data, duplicate messages, sequence gaps and restart from persisted state. Reconcile order and execution identifiers across the engine, gateway, venue and back office.
- Confirm obligations and ownership. Ask compliance and legal teams to check licensing, best execution, record retention and reporting requirements for the relevant jurisdiction. Assign an owner for rule changes, venue certification and incident communications.
For a venue connection, also establish session limits, heartbeat behaviour, throttles, supported order types and market-data entitlements before implementation. For an engine, define the authoritative clock, event sequencing, persistence model and client-visible status transitions. Those details affect incident handling as much as the matching algorithm.
Treat the engine as a live risk system
A matching engine is not a component that can be handed to operations after launch. Its state determines which orders are live, which have filled and what positions the firm may hold.
A restart that restores a book but loses sequence or execution history can create duplicate orders, disputed fills or incorrect exposure. Recovery tests should cover state rebuild and reconciliation, not just process uptime.
Set service indicators around the full path: order acceptance, acknowledgement, execution publication, market-data freshness and reconciliation breaks. Keep separate measures for internal processing and external network time, so a venue delay is not misdiagnosed as an engine fault. Alert thresholds need to reflect normal variation by venue and instrument, with a clear escalation route when the source of delay is uncertain.
Operations teams should maintain a short, tested runbook that covers:
- How to pause order entry and cancel or retain resting orders;
- How to switch routes or venues without duplicating live orders;
- Who can change risk limits, matching parameters and session settings;
- How to reconcile fills after a disconnect, including late reports;
- Which records and timestamps are preserved for incident review.
Change control matters in both models. A venue can change a protocol or trading schedule; an engine release can alter priority or rejection behaviour. Use certification environments where available, replay captured messages and test rollback before production changes. Keep a versioned record of configuration, rule sets and venue specifications. These controls make it possible to explain not only that an order failed, but what the system believed at the time.
The practical split is clear: venues sell access to an established market, while an engine makes the firm responsible for the market mechanism. Trading infrastructure rarely removes complexity. It decides where the complexity sits.
Frequently Asked Questions
What does a matching engine do?
A matching engine receives orders, applies validation and matching rules, and generates executions when eligible buy and sell interest meets. It also manages order state, sequence and events such as amendments, cancellations and rejects.
Is it cheaper to use a venue or build a matching engine?
Venue access usually has a shorter initial delivery path, but can involve connectivity, market-data and transaction fees. An engine adds development, support, resilience and regulatory work, so the lower-cost option depends on flow, rules and operating scope.
How long does it take to connect to a trading venue?
A basic venue connection commonly takes 4-12 weeks as a typical market planning range. Production delivery can take longer if the work includes multiple order types, market-data feeds, risk controls, certification or complex back-office integration.
When should a broker build its own matching engine?
A broker may consider an engine when it needs specific matching rules, an internal market or a controlled execution layer that available venues do not provide. The case depends on having sufficient liquidity and the people and controls to operate the system continuously.
Can a broker use both an internal engine and external venues?
Yes. A hybrid design can match eligible flow internally and route other orders to external venues. It needs explicit routing rules, controls for conflicts and self-trading, and consistent reporting across all execution paths.
Platforms · Software · Providers · Markets