Every copier claims to be fast, almost none says what it is measuring. Here is the full path a trade takes, which segment a vendor's number usually covers, and how to measure the only figure that actually reaches your account.
A vendor's landing page says 50 milliseconds. Another says under 100. A third says sub-millisecond, which should make you close the tab. Meanwhile the trader who bought one of them looks at a Follower account that filled 1.4 points worse than the Provider on a gold trade and cannot work out whether that is normal, bad, or what they were sold.
The answer to what "fast" means is that the advertised number and the number that reaches your account are usually measuring different things. A copier's latency is not one figure; it is a chain of six or seven segments, only one or two of which the vendor controls, and the quoted number is almost always the controlled part. What lands in your account is the whole chain, expressed not in milliseconds but in price — the difference between the Provider's fill and the Follower's fill, which is the only unit that has ever cost anyone money.
INFO
Quick answer. Measure slippage, not milliseconds. Compare the Provider's fill price with the Follower's fill price on the same trade, over at least fifty trades, and look at the distribution rather than the average. A vendor's latency figure normally covers only its own processing segment, and that segment is rarely the largest one in the chain.
A copied trade crosses seven boundaries. The illustrative figures below are orders of magnitude for a typical hosted setup, not a specification and not a promise — your own numbers depend on your brokers, your instruments and the hour.
Two observations follow immediately, and they are the whole subject.
Segments two and six are usually the largest, and no copier controls either. How quickly your Provider's broker tells anyone a fill happened, and how quickly your Follower's broker accepts and fills an order, are properties of those brokers. A copier that processes in one millisecond and sends to a broker that takes 400 has produced a 400-millisecond trade.
The vendor's quoted number is segments three to five. That is the honest reading of almost every latency claim in this industry. It is not dishonest to measure your own component — it is the part you can improve — but quoting it without saying so invites the reader to think it is the whole path.
And "sub-millisecond" end-to-end is not a claim about copying. Light takes about five milliseconds to cross the Atlantic in fibre, before any processing. Anyone advertising sub-millisecond copying between retail brokers is either quoting one internal function call or is not being straight with you.
There is also a hop the diagram above hides, and it is platform-specific. A hosted service cannot speak MetaTrader's protocol without a MetaTrader terminal running somewhere, so on MT4 and MT5 there is an execution provider's cloud between the relay and the broker, in whichever region your account is homed. On cTrader, the connection reaches the broker's backend through the platform's own proxy cloud rather than directly. On Tradovate our worker is the broker-facing hop. Three platforms, three different chains — which is the real reason a single advertised latency figure cannot describe your setup.
Here is the part that takes a while to accept: two copiers with identical latency can produce very different results, and a slower copier can produce better fills.
Latency only costs you when price moves during it. That means the damage is the product of two things — time elapsed and how fast the instrument is moving — and the second term varies by more than the first ever will.
An illustrative comparison on a single gold trade:
The same copier, the same latency, three completely different outcomes. This is why the useful metric is measured in price, not time: slippage already contains both terms.
It is also why a slippage limit is a more valuable feature than a fast relay. A limit that declines to fill a copy more than a configured distance from the Provider's price turns an uncontrolled loss into a logged skip. You will miss some trades. You will also not be filled forty points into a spike because a queue backed up.
WARNING
A copier with no slippage limit is fastest in exactly the conditions where being fast does not help and being wrong is most expensive. Ask what a tool does when it is slow, not how fast it is when everything is fine.
This takes an afternoon to set up and a fortnight to produce an answer, and it is worth more than every latency figure you will read.
If a vendor will not let you run that test on a demo account before you pay, that is your answer about their numbers.
Ranked by how much improvement they tend to produce, which is close to the reverse of how much attention they get.
Broker execution quality on the Follower side. By a distance. A Follower at a broker with slow order acknowledgement will be slow regardless of everything upstream. This is the variable most traders never test and can change most easily.
Instrument and hour. Copying gold in the first seconds after a release is a different problem from copying EURUSD at 10:00 London. Much of what gets blamed on a copier is the spread widening.
Order type. A market order takes what is there. A limit entry waits for the price you named and either fills or does not. Where your strategy tolerates it, a watched limit entry removes the slippage question rather than optimising it — which is why signal entries at a price not yet reached are held and watched until they fill or expire rather than being converted into market orders.
Geographic distance between the relay and the brokers. Real, and smaller than people assume next to the two factors above — and on MetaTrader and cTrader it is partly out of any vendor's hands, because a terminal-hosting provider and a platform proxy cloud respectively sit in the middle of the path. It is the reason dedicated regional relays are on our roadmap; they are planned, not shipped, and their honest benefit is on the platforms where our worker is the broker-facing hop.
The copier's own processing. Genuinely the smallest term in a competently built system, and the one every marketing page is about.
Our relay is hosted. The worker that receives a Provider's fill and issues the Follower's order runs on our infrastructure rather than on your machine, which is what lets it keep working when your laptop is shut, and which means the network path is a real part of your chain.
What we do about that is measure it and show it rather than claim a number. Every copied trade records its own relay time, and the Cockpit shows it per trade alongside the fill. So does every trade that was held or skipped, with the reason — slippage limit exceeded, risk check failed, no contract mapping, account halted.
That is a deliberately unglamorous answer to "how fast is it". The alternative — a single averaged figure on a marketing page — tells you nothing about your brokers, your instruments or your hours, and cannot be checked. A per-trade measurement you can export can be.
End to end, typically tens to hundreds of milliseconds between two retail brokers, dominated by the two brokers rather than by the copier. The vendor's advertised figure usually covers only its own processing segment. Any claim of sub-millisecond copying between separate brokers should be treated as a marketing number: network physics alone exceeds it over any real distance.
There is no universal figure, because it depends on your instrument, your session and your Follower's broker. What is meaningful is your own distribution over at least fifty trades: the median, the 90th percentile and the worst case, split between quiet hours and volatile ones. Measure it on demo before you commit an account, and re-measure if you change brokers.
Sometimes, on one segment, and the difference is usually smaller than the broker difference sitting on either side of it. A local copier removes one network hop. It also stops when your machine stops, keeps no independent record, and cannot copy while you sleep unless you are running a VPS — at which point it is a cloud copier you administer yourself.
It should tell you, with a reason, and not replay it later at a price that no longer reflects the decision. Ours records every held and skipped trade in the Cockpit with the cause — slippage limit exceeded, risk check failed, no contract mapping, account at its daily halt. A silently missing trade is how a Follower account quietly becomes a different strategy from the Provider it is supposed to be tracking.
It reduces one segment, and only if the copying itself runs on that VPS. Where the relay is hosted, the relevant distance is between the relay and the brokers, not between you and anything — your browser is looking at a dashboard, not carrying orders. Before paying for proximity, measure whether your slippage is actually dominated by distance or by your Follower broker's execution.
Because time passed and the price moved, and possibly because the two accounts are at different brokers with different spreads and different liquidity. This is expected and it is roughly symmetrical in quiet conditions — Followers fill better about as often as worse. What is not expected is a persistent one-directional gap, which usually points at the Follower's broker rather than at the copier.
For a fuller map of the path a copied trade takes and which segment belongs to whom, /blog/copy-trading-data-flow-execution-speed walks the whole route. And if the question behind your latency question is really what the copier does when something goes wrong mid-trade, the retry and reconciliation behaviour is the subject of /blog/trade-replication-engine-explained.
