Liquidity
Why latency belongs on the P&L, not the infrastructure ticket
Latency is usually owned by whoever runs the infrastructure, which is why it tends to be discussed in milliseconds rather than in money. Framed that way it always loses to whatever else is on the roadmap. Framed as a recurring cost per fill, it stops being a technical detail.
Where the delay actually goes
The time between a client pressing a button and a fill coming back is spent in several places: the client's own connection, your platform, the network between your servers and the venue, and the venue itself. Only some of that is yours to change.
The portion that is yours is usually dominated by physical distance. Servers in the wrong region add a fixed cost to every single order, and no amount of software optimisation recovers it.
The cost is per fill, which is why it compounds
A small amount of extra delay produces a slightly worse price on each order. On its own that is unnoticeable. Multiplied by every order every day, it becomes a number large enough to compare against the cost of moving infrastructure.
This is the calculation that reframes the conversation. Latency stops being an engineering preference the moment somebody expresses it as an annual figure next to the cost of fixing it.
Rejection is the second-order effect
Delay does not only produce a worse price. Past a certain point it produces no price at all, because the quote has moved beyond tolerance by the time the order arrives. The client sees a rejection and re-sends, usually at a worse level.
Rejections also generate support contact, which has its own cost that never appears in an execution report.
Measure it before optimising it
Round-trip time should be measured continuously rather than tested once during procurement. A single benchmark taken on a quiet afternoon says almost nothing about behaviour during a news event, which is when latency costs the most.
Once it is measured, the fix is often unglamorous: move the servers closer, reduce the number of hops, or change where the connection terminates. These are not sophisticated interventions, but they address the part of the delay that is actually yours.
In short
What to take away
Express latency as an annual cost, not in milliseconds, to get it prioritised
Physical distance dominates; software optimisation rarely recovers it
Past a threshold, delay converts into rejections rather than worse prices
Measure continuously — a quiet-afternoon benchmark proves nothing
Read next
Find out where your delay is
Most of the round trip is not where teams assume it is. We can walk through where yours is going and which parts are worth moving.