EducationExecution~16 min readUpdated 30 September 2026
The short answer
A broker either passes your order to a liquidity provider (A-book) or takes the other side of it internally (B-book). Most brokers do both — they decide per client, per instrument, per trade size. The label on the website is marketing. The only way to know which side of your trade your broker is on is to measure your fills over time.
Why this is not a "scam broker" lesson
Every article on this topic splits brokers into two groups: good (ECN, no dealing desk) and bad (market maker, dealing desk). That is wrong. Market makers can offer the best execution in the industry, and ECNs can slip you harder than a market maker during volatile events.
This lesson is not about which model is good. It is about understanding what actually happens between your click and your fill — because that determines how the price you see becomes the price you get.
T
Written by the Trade To The Top team|Reviewed 30 September 2026
Execution model descriptions cross-checked against FCA and ASIC CFD disclosure guidance, ESMA product intervention materials, and Trading and Exchanges (Harris) chapters on dealer markets and internalisation. Internalisation rate context drawn from publicly available broker annual reports.
You clicked buy. The platform said filled. But what actually happened between the click and the fill? That question has a different answer at every broker, and the answer determines everything about how your prices behave. This lesson follows the order through the system.
Key takeaways
A-book: broker routes your order to a liquidity provider. Earns from spread markup and/or commission. Client wins or loses, broker earns either way.
B-book: broker takes the other side. Client loss is broker revenue. Client win is broker cost.
Most brokers do both. Internalisation decisions are made per client, per instrument and per trade size.
Market maker ≠ bad. Internalised execution can produce better fills on small orders than a routed one.
ECN ≠ always better. ECNs can slip harder during news because the order has to travel further.
"No dealing desk" is a marketing phrase, not a guarantee that your order was not internalised.
Requotes are a symptom of a dealing desk that can reject fills. Not all models produce them.
Last look means the LP gets a final chance to reject a fill. It creates a specific kind of slippage.
The only way to measure your broker's behaviour is to log every fill: requested price, filled price, time.
Your execution statistics are the only honest answer. The label is not.
You click buy. The platform sends the order to your broker's server. From there, one of two things happens.
Option 1 — the order is passed on. The broker forwards it to one of several liquidity providers (LPs) — typically large banks, non-bank market makers, or other brokers with deeper books. The LP fills it, the broker keeps a small markup, and the trade shows up on the LP's book. This is called A-book routing. The broker does not hold the other side. The client's profit or loss is between the client and the LP.
Option 2 — the order is kept in-house. The broker takes the other side itself. The trade never touches an external LP. The broker's book now contains a position against the client. This is called B-book internalisation. The broker is now the counterparty. If the client loses, the broker gains. If the client wins, the broker pays.
CLICK TO FILL · THE TWO ROUTES
Every order goes down one of these two paths. Most brokers use both, depending on the trade.
Both paths return a fill. Only one of them puts the broker on the other side of your trade.
Read the two middle boxes. Both paths produce a fill in your platform. Both paths debit your account by the same spread. Both paths return a confirmation. The client cannot see, from the outside, which one their order took.
And that is the whole problem. The path matters — but the platform hides it.
The A-book route
On the A-book route, your order is sent to an external liquidity provider. The broker is a middleman, not a counterparty.
What this means in practice:
The broker earns from markup or commission, not from your losses. A 0.3-pip markup on a 0.1-pip LP spread means the broker's revenue is 0.3 pips on every trade, whether you win or lose.
The broker is indifferent to your outcome. They want you to keep trading so they keep earning the markup. They do not want you to blow up — not because they care, but because a blown-up client stops paying.
Your fills come from the LP's book. If the LP is deep and liquid, fills are clean. If the LP is thin, fills slip.
The broker's only incentive is to route to the best LP for the size. Best execution should be automatic — because the broker's revenue depends on volume, not on adverse selection.
In theory, A-book execution is the cleanest model for the client. In practice, it depends on how the broker selects LPs and how it handles last look. Some A-book brokers route to the fastest LP, not the best-priced one. Some route to whichever LP gives them the highest rebate, not the tightest spread. The conflict of interest does not disappear — it just changes shape.
The B-book route
On the B-book route, the broker keeps the trade in-house. The broker becomes the counterparty.
What this means:
The broker earns from your losses and pays from your wins. The revenue model is inverted relative to A-book.
The broker controls the price you see. Because there is no external LP involved, the broker sets the bid and ask. On non-regulated or weakly regulated brokers, this is a serious conflict.
B-book does not automatically mean worse fills. For small retail positions, internalisation can produce faster fills and fewer requotes than routing to an LP with minimum size requirements.
B-book does mean the broker sees your position and can act on it. This is the real problem. It creates asymmetric information that a well-run A-book broker does not have.
The B-book conflict is only a problem when it is abused. On regulated brokers, it usually is not. The broker still wants to keep you as a client, and driving every client to a loss makes them leave. On unregulated brokers, abuse is more common: stop-hunting, widening spreads precisely when a client's stop is in range, and other practices that have been the subject of regulatory fines.
Why most brokers are hybrids
This is the part that almost no comparison explains. The same broker runs both models simultaneously. They do not choose once. They choose continuously, per client, per instrument, per trade.
The decision logic looks something like this:
The internalisation decision tree — simplified
01
How big is the position?
Small retail positions (most of the flow) are internalised. Large positions are routed. There is no economic reason to route a 0.01-lot trade to a bank that will reject it for being below minimum size.
02
Is the client profitable?
Consistently winning clients are often auto-routed. Consistently losing clients are often internalised. This is not universal, but it is common and legal in most jurisdictions. The broker measures clients like a casino measures card counters.
03
Is the instrument liquid?
Majors like EUR/USD are easy to internalise because pricing is well-known and the broker can hedge cheaply. Exotics and less-liquid CFDs are routed because pricing is harder to synthesise.
04
Is the current volatility abnormal?
During news events, more flow is routed because the broker does not want to be caught on the wrong side of a spike. During quiet hours, more is internalised because the risk is small.
05
What is the broker's net exposure?
If the broker has accumulated net long exposure from internalised clients, it may route incremental flow to offset. If the flow is balanced (equal buys and sells), it may hold in-house and pocket the spread.
None of this is disclosed to the client. Your broker does not tell you which side of this tree you are on today. The label on the website says "ECN" or "STP" or "no dealing desk" regardless of which model is being applied to your specific trades.
The broker does not decide once. It decides on every single order.
Market maker, STP, ECN
Three labels get used constantly. Here is what they actually mean.
Label
What it really means
What it tells you about your fills
Market maker
Broker quotes both bid and ask. May internalise or hedge.
Nothing about whether your trade is A-book or B-book. The label only says they set the price.
STP (Straight-Through Processing)
Broker routes orders to LPs without manual intervention.
Usually more consistent pricing. Does not guarantee best LP selection.
ECN (Electronic Communication Network)
Orders matched against a pool of LP quotes.
Tight spreads, commission charged separately. Not immune to slippage on news.
Dealing desk
Manual or semi-automated intervention on order flow.
Requotes are possible. Fills can be rejected at the broker's discretion.
No dealing desk
Marketing term. Means no manual intervention.
Nothing specific. The broker can still internalise at server level.
Notice the column on the right. Almost none of the labels tell you anything concrete about your fills. "Market maker" is a technical description of how the broker quotes prices. It does not say whether the broker is on the other side of your specific trade. "No dealing desk" says nothing about whether your order is being internalised by the server rather than a human.
The two labels that do carry information:
STP implies systematic routing. There is a technical mechanism that sends every order out. It is not a guarantee of quality, but it is a guarantee of model.
ECN implies matching against multiple LPs. The broker is not setting the price; it is aggregating. This is the closest thing to a structural commitment you will find.
Even those two can be abused. Some brokers call themselves ECN while routing all retail flow to a single LP with a strong markup. The only way to know the truth is to measure.
The conflict of interest question
This is where the theory meets your account. If your broker is B-booking your trades, they profit when you lose. That does not mean they will cheat you. It means the incentive structure is inverted.
What this changes in practice:
A-book broker — aligned incentives
Revenue sourceMarkup + commission
Client wins?Broker unaffected
Client loses?Broker unaffected
Best clientLong-term profitable trader
AlignedBoth want you profitable
B-book broker — conflicting incentives
Revenue sourceSpread ± client P&L
Client wins?Broker pays out
Client loses?Broker keeps the loss
Best clientHigh-volume loser
ConflictingBroker benefits from your loss
Two things soften this conflict on well-run brokers:
Churn destroys the model. A B-book broker still needs clients to keep depositing and trading. Driving every client to blow up is bad business. The best B-book brokers are quiet about their model and generous about the spread.
Regulation constrains abuse. In Tier-1 jurisdictions, sharp practices like stop-hunting and last-minute spread widening are heavily fined. In Tier-3 jurisdictions, they are not.
The general rule: on a well-regulated broker, A-book is safer. On an unregulated broker, you cannot trust either model. This is why Lesson 58 places such emphasis on the regulation tier.
Common mistake
Assuming a B-book broker is automatically cheating you. Some of the cleanest-filling brokers in the industry run B-book for retail flow. Their pricing is tighter than most STP brokers because they do not pay LP fees on every small trade. The conflict exists, but it is not always acted on.
Why the labels lie
The marketing labels on broker websites are not regulated in most jurisdictions. A broker can call itself an ECN while running a dealing desk. It can call itself an STP while internalising 90% of client flow. The label describes a capability, not the current model being applied to your trades.
Specific abuses to watch for:
"We are ECN" but there is no commission line. True ECNs charge commission because they do not mark up the spread. If you see tight advertised spreads and no commission, the broker is either marking up on the LP side (fine) or running a B-book without saying so (problematic).
"No dealing desk" but requotes happen. A requote is a fill rejection. If your broker requotes, there is a decision being made about your order — automated or not.
"Tight spreads" during quiet hours, "spread widening" during news. Every broker widens during news. The question is how wide. A broker that widens from 0.5 to 5.0 pips during a release is behaving very differently from one that widens from 0.5 to 1.5.
"Average execution" statistics with no methodology. Average execution speed of 30ms is meaningless unless you know what percentage of orders were internalised, rejected, or filled with slippage.
Last look — the invisible rejection
Last look is a pricing mechanism where a liquidity provider gets a final chance to accept or reject an incoming order at the quoted price. It exists to protect LPs from latency arbitrage — traders exploiting tiny price differences between venues.
It also creates a specific kind of slippage. If the market moves in your favour during the last-look window, the LP rejects the fill and requotes at a worse price. If the market moves against you, the fill is accepted. The result is a systematic asymmetry where you get the worse outcome slightly more often than random.
Last look is standard in interbank FX. It is not illegal. It is not unfair. But it does mean that "0.0 pip spread" is not literally what you always get. Any broker using LPs with last-look provisions has a small, structural, invisible cost baked into your fills.
Worked example — one trade, three models
Same trade, three brokers running different models. EUR/USD long, 1 lot, entry at 1.0853, stop at 1.0841, target 1.0877. The market makes a 5-pip spike against the position 12 seconds after entry, then rallies to target.
Worked example — same trade, three execution models
Trade
Buy EUR/USD at 1.0853, stop 1.0841, target 1.0877
Event
12 seconds after entry: 5-pip adverse spike
Outcome
Target hit 4 minutes later
Broker A — clean A-book routing.
Fill at 1.0853. Spread 0.3. Spike triggers a temporary 5-pip widening but the position is not stopped.
Target hit at 1.0877. Net: $240 − $3 = +$237.
Broker B — B-book with tight internal pricing.
Fill at 1.0853. Spread 0.2 (internal pricing). Spike is internal — the broker widens to 6 pips for 3 seconds, briefly touching the stop level on the broker's own quote.
Stop hit at 1.0841 on the widened quote. Position closed at a loss.
Result: −$120. Market then goes to target.
Broker C — ECN with last look.
Fill at 1.0853.5 (0.5-pip slippage on entry due to last look rejecting the tightest quote).
Spike causes a re-quote on the exit. Target filled at 1.0876.5 instead of 1.0877.
Net: $230 − $6 = +$224.
SAME SETUP. +$237 · −$120 · +$224.
Three things happened:
Broker A behaved as expected. Clean fill, clean exit, small cost.
Broker B stopped out the trade on a widened internal quote. Nothing illegal happened. But the client's outcome was determined by the broker's own spread behaviour, not by the underlying market. The market never traded at 1.0841. The broker's quote did.
Broker C lost value on both sides of the trade — slippage on entry due to last look, slippage on exit due to a re-quote. Total cost was 0.9 pips more than Broker A, on the same trade.
The result: Broker A paid the client $237. Broker B cost the client $120. Broker C paid the client $224. Same trade, same target, different execution models, $357 swing between best and worst outcomes.
The trade you place and the trade you get are two different things. The broker decides how different.
When this fails
Where broker mechanics break down
Assuming the label on the website describes your execution. It does not. Every broker uses multiple models simultaneously. The label is a marketing position, not a technical commitment. The only way to know what is happening to your specific orders is to log them and analyse the pattern.
Choosing a broker based on the model, not the fills. An "ECN" with poor last-look handling can cost you more than a "market maker" with tight internal pricing. The model does not determine the fill. The infrastructure does. Rank brokers on your own measured slippage, not on their marketing label.
Ignoring requotes as a signal. A broker that requotes is a broker that can reject your fills. Some requotes are legitimate (order to buy at a price that no longer exists). Some are adversarial (order was priced correctly but the broker decided not to fill). If you see frequent requotes, treat it as a data point on your broker's model.
Trusting "average execution" statistics. Every broker publishes something like "30ms average execution". These numbers are selectively reported and often omit rejections and requotes from the average. The only honest statistic is your own fill log. Measure yours, not theirs.
Forgetting that the model changes. A broker that A-booked your flow for two years can switch you to B-book if your account becomes profitable enough to matter. Do not assume the model you started with is the model you still have. Review your slippage patterns quarterly.
None of this means brokers are fraudulent. It means the industry structure is opaque and the marketing is misleading. The framework is not "find the honest broker." The framework is "measure the actual fills."
How to measure your broker's behaviour
Log requested vs filled price on every trade.Average the slippage. Track it weekly.
Track requote frequency.How many orders were rejected and repriced?
Track spread at entry vs spread at 15:00 EST.Is the tight advertised spread achievable at your trading hours?
Compare fills to the market price at the moment of your click.If you consistently fill worse than the current market, execution is deteriorating.
Review quarterly.A broker that was fine six months ago may not be fine today. Your data tells you.
In one box
A-book: broker routes to LPs, earns markup.
B-book: broker takes the other side, earns spread ± client P&L.
Most brokers do both. The decision is made per client, per instrument, per trade.
Market maker ≠ scam. Internalised execution can be cleaner for small orders.
ECN ≠ always better. Last look can cause structural slippage.
"No dealing desk" means nothing specific. It is a marketing label.
Requotes are a signal. They indicate fill rejection is possible.
Last look creates systematic asymmetry. You get the worse side slightly more often.
Your broker may switch your model. Profitable clients are often routed to A-book.
Measure your own fills. Not the broker's published numbers. Not the website's label.
Log requested vs filled price on every trade. Our free journal has a slippage field and a requote flag. After 30 trades you have a personal measurement of your broker's execution quality — not the label on their website.
5 questions · immediate feedback · retake any time
Question 01 of 05
What is the difference between A-book and B-book?
Correct: B. A-book means the broker passes the trade to an external LP and earns from markup or commission. B-book means the broker internalises the trade and is on the other side. Both paths return a fill in your platform.
Question 02 of 05
What does "no dealing desk" actually guarantee?
Correct: C. "No dealing desk" is a marketing term. It means no human intervention on order flow. The broker can still run automated internalisation at the server level, and the term says nothing about which model is being applied to your specific trades.
Question 03 of 05
What is "last look" and why does it matter?
Correct: A. Last look is a pricing mechanism where an LP can accept or reject a fill within a small time window. If the market moves in your favour, the LP can reject and requote at a worse price. If it moves against you, the fill is accepted. The result is a structural asymmetry baked into your fills.
Question 04 of 05
Most brokers run:
Correct: D. Most brokers run both models simultaneously. The internalisation decision is made continuously — per client (profitable or not), per instrument (liquid or exotic) and per trade size. Small retail orders are usually internalised; large orders are routed.
Question 05 of 05
What is the only reliable way to know which side of your trade your broker is on?
Correct: B. The broker's model is not disclosed per-trade. Marketing labels are not regulated. Support staff may not know. Regulatory filings rarely disclose internalisation rates. Your own fill log — requested price, filled price, time, requote or not — is the only reliable measurement.