Smart Order Routing (SOR): How It Works

What is smart order routing?

Smart order routing (SOR) is the automated process of deciding where to send an order — across exchanges, ECNs, dark pools, and internal liquidity — so it executes at the best available terms. A smart order router continuously reads consolidated market data, evaluates price, displayed and hidden size, fee and rebate schedules, latency, and fill probability at each venue, then splits and sequences the order across those venues. In US equities it is the mechanism firms use to meet their best-execution and Reg NMS order-protection obligations without a human choosing a venue on every order.

In one line: the order says what to buy; the smart order router decides where.

How does smart order routing work?

A smart order router runs the same loop on every order, usually in microseconds:

• 1. Ingest market state. Consolidated quotes plus direct feeds establish the National Best Bid and Offer (NBBO) and depth at each venue.

• 2. Score the venues. Each destination is ranked on price, available size, expected fill probability, historical fill quality, fee or rebate economics, and round-trip latency.

• 3. Slice the order. The parent order is broken into child orders sized to the liquidity actually resting at each venue, so the router does not signal its full intent.

• 4. Sequence and send. Child orders go out in parallel or in a deliberate order — often dark venues first to seek non-displayed liquidity before posting displayed size.

• 5. React. Partial fills, cancels, and quote changes feed straight back into the scoring loop. Unfilled residual gets re-routed, re-priced, or posted.

• 6. Record. Every routing decision, child order, and venue response is captured for transaction cost analysis (TCA), best-execution review, and CAT reporting.

The quality of a router lives almost entirely in steps 2 and 5. Any system can send an order to the venue at the best displayed price; the difference between routers shows up in how they handle hidden liquidity, fleeting quotes, and residual size.

At Gato, that routing loop runs inside gHub, which pairs the router with a broker-agnostic FIX connectivity hub so venue additions are configuration rather than integration work.

Smart order routing vs direct market access (DMA)

These get conflated constantly, and they are not the same layer.

DMA is a pipe. SOR is a decision engine that can use many pipes, including DMA ones. Most firms need both: DMA for strategies that already know where they want to be, SOR for everything else.

Smart order routing strategies

"Which router" matters less than "which routing strategy for which flow." The common ones:

• Sweep (liquidity-seeking).—Hit every venue at or better than the limit price simultaneously. Maximum immediacy, maximum signalling.
• Sequential / ping.—Probe venues in ranked order, dark first. Slower, lower footprint.
• Post-and-wait (passive).—Rest displayed size at the venue with the best queue position and rebate economics, re-pricing as the market moves.
• Dark aggregation.—Route only to non-displayed venues to source size without moving the quote.
• Fee/rebate-aware.—Weight venue choice by the maker-taker economics, which for high-volume flow can matter as much as a fraction of a cent in price.
• Algorithmic parent orders.—VWAP, TWAP, POV, and implementation shortfall algos decide *when* to trade; the SOR decides *where* each slice goes. They stack rather than compete.

A router with one strategy is a router with one opinion about your flow. Look for configurable, per-flow routing tables.

How US market fragmentation shapes routing

There are 16 registered national securities exchanges in US equities plus dozens of alternative trading systems and a large share of volume executed off-exchange by wholesalers and internalizers. No single venue holds a durable majority of liquidity in most names, and displayed size at the top of book is frequently a fraction of the size a firm actually wants to trade.

That structure produces three routing problems a naive "send to the best displayed price" rule cannot solve:

• Quote fading.—The size you saw may not be there when your order arrives. Routers compensate with fill-probability scoring per venue rather than treating the quote as a promise.
• Hidden liquidity.—Midpoint and non-displayed size in ATSs and conditional venues never appears in the consolidated quote at all, so a displayed-only router systematically overpays on larger orders.
• Odd-lot and sub-penny effects.—Odd-lot quotes are not protected under Rule 611 but still represent real liquidity, and price improvement often arrives in sub-penny increments off-exchange.

The practical read: routing quality is mostly a function of how the router models the liquidity it *cannot* see, not how quickly it reacts to the liquidity everyone can see.

Measuring routing quality: the metrics that matter

Fill price alone tells you almost nothing without a benchmark. A defensible routing review looks at:

Two disciplines separate firms that improve from firms that only report. First, segment by flow: retail marketable orders, institutional parent orders, and options hedging flow have different optimal venue rankings, and a blended average hides all three. Second, close the loop — TCA output should change routing tables on a defined cadence, and each change should be dated so the next best-execution review can attribute the improvement.

Routing beyond US equities

The routing loop generalizes, but the constraints change by asset class:

• Listed options.—Sixteen-plus options exchanges, a linkage plan rather than Reg NMS, complex-order books, and maker-taker economics that vary sharply by class. Routing has to reason about multi-leg net prices, not per-leg quotes. Our gTrader Suite handles multi-leg construction ahead of the router.
• Futures.—Single-venue per contract in most cases, so the interesting decision is order type and timing rather than destination — which is why DMA dominates.
• Fixed income.—Quote-driven and relationship-driven; "routing" means RFQ dealer selection and list-trading logic, scored on response rate and historical spread rather than posted depth.
• Crypto and FX.—Fragmented, no consolidated tape, and venue credit and settlement risk enter the scoring function alongside price.

A router built strictly around Reg NMS assumptions rarely ports cleanly. Ask any vendor which asset classes share the same routing engine and which are separate code paths.

Common smart order routing mistakes

Patterns we see repeatedly in routing reviews:

• 1. Optimizing rebates instead of execution. Passive routing to the top-rebate venue looks free until reversion analysis shows the fills you get there are the ones you did not want.

• 2. Over-slicing. Too many small child orders raises message rates, invites detection, and can worsen the residual it was meant to protect.

• 3. Static venue tables. Fee schedules and venue behaviour change quarterly; routing tables reviewed annually are stale by definition.

• 4. Risk checks outside the routing path. A pre-trade control that runs before the parent order but not on each child order does not satisfy the spirit of SEC Rule 15c3-5.

• 5. Reconstructed routing events. Rebuilding routes for CAT from fills rather than the router's own log produces timestamp and sequencing rejections.

• 6. No kill switch at the routing layer. When a strategy misbehaves, the fastest safe stop is at the router, not at the venue session.

Smart order routing and best execution obligations

Routing is not just an execution-quality question — it is a regulatory one.

• Reg NMS Rule 611 (order protection)—prohibits trade-throughs of protected quotations. Your router is the control that enforces this on every order.
• FINRA Rule 5310 (best execution)—requires reasonable diligence to obtain the most favourable terms, and a regular and rigorous review of execution quality. That review needs routing data, not just fill prices.
• SEC Rule 15c3-5 (market access)—requires pre-trade risk controls on every order reaching an exchange or ATS. Those checks sit in front of the router and must run at routing speed, not in a nightly batch. We covered this in detail in the SEC Rule 15c3-5 compliance guide.
• Rule 605/606 disclosures—publish your execution quality and routing behaviour. They are generated from the same decision log the router writes.
• CAT reporting—requires every route, modification, and cancel as a reportable event with accurate venue and timestamp data.

The practical consequence: your routing log is a compliance artifact. If the router cannot explain, per order, why it chose a venue, your best-execution review is an assertion rather than an audit. Gato pairs the router with real-time pre-trade risk ahead of the routing decision and automated regulatory reporting behind it, off the same event store.

How to evaluate a smart order router

Ask vendors these, in this order:

• 1. Which venues, and how are they added? Count is less interesting than the marginal cost of venue number twenty-one. If it is a release cycle, you are carrying implementation risk on every new market.

• 2. Can I see and change the routing logic? Venue ranking, dark-first sequencing, fee weighting, minimum fill sizes. A black box cannot be tuned to your flow or defended in an exam.

• 3. Where do the pre-trade risk checks run? In the router's path, or bolted on upstream? Two risk engines on one flow eventually disagree.

• 4. What does the decision log capture? You want the venue scores considered, not just the venue chosen.

• 5. Is TCA fed back into routing? Measurement that never changes behaviour is reporting, not optimization.

• 6. How does it behave on the residual? Handling of the last unfilled 10% is where slippage concentrates.

• 7. What is the latency budget, end to end? Risk check plus routing decision plus wire time — measured at the 99th percentile, not the mean.

• 8. How do routing events reach CAT? From the router's own log, or a reconstruction? Reconstructions produce rejections.

Where the router sits in the trading stack

Smart order routing is a component, not a product category. In a working stack it sits between the execution surface and the venues:

Trader or client order → OMS (capture, allocation, compliance) → EMS / algo (when and how to trade) → pre-trade risk (15c3-5) → smart order router (where) → venues → fills back up the chain → TCA, books-and-records, CAT

If you are still deciding how the layers above the router should be organized, the OMS vs EMS guide walks through where to draw that line — and the FIX connectivity guide covers the transport layer underneath it.

The build-vs-buy question

Building a router is tractable. Keeping it current is the expensive part: venue rule changes, new order types, fee schedule revisions, market structure amendments, and the execution-quality analytics needed to prove the thing still works. Firms that build usually do it because routing logic *is* their edge. Firms whose edge is elsewhere are better off buying a router they can configure, and spending their engineering on the strategy that generates the flow.

How Gato handles smart order routing

Gato's routing sits on one order state machine rather than a chain of integrations:

• gHub — order management, broker-agnostic FIX connectivity, and configurable smart order routing with a full per-order decision log.
• gTrader Suite — the execution surface: consolidated market data, DMA, algos, and multi-leg options, feeding the same router.
• gPrecision — market-aware pre-trade risk in the routing path, aligned with SEC Rule 15c3-5.
• gReg — CAT and CAIS reporting generated from the routing events themselves, not reconstructed after the fact.

If you are evaluating routers — or trying to work out why your fill quality drifts on the residual — request a demo and we will walk through the decision log on live flow.