A-Book vs B-Book vs Hybrid: How to Choose a Risk Model
2 November 2026 · 9 min read
A-book means you pass client trades to external liquidity and earn the markup; B-book means you internalise the trades and carry the risk yourself; hybrid means you do both, routing each client or flow type to the book that fits it. Nearly every retail broker above a certain size runs hybrid, because pure A-book leaves revenue on the table and pure B-book concentrates risk you cannot always afford.
The right question is not which model is best but which flow belongs in which book, and who is watching the boundary. This article lays out how each model works, what it earns and risks, and how to run the hybrid boundary well. For the cost of getting the B-book side wrong, see what toxic flow actually costs.
How each model works
A-book: client orders are hedged one-for-one with a liquidity provider, usually through a bridge. Your revenue is the spread markup and commission; your market risk is near zero; your costs are LP spreads, bridge fees and the operational overhead of managing LP relationships.
B-book: client orders are internalised. You are the counterparty. Client losses are your profit and client profits are your cost. Your revenue is the spread plus the statistical edge that retail flow, in aggregate, loses money; your risk is that the aggregate is not the individual, and a single large or informed client can hurt you badly.
Hybrid: a routing layer classifies flow and sends each trade, group or symbol to the appropriate book. Small, unsophisticated flow is internalised; large, sharp or correlated flow is hedged. The routing rules, and their continuous review, are the entire game.
It is worth being precise about what hybrid is not. It is not a fixed percentage split set once at launch, and it is not a dealer's discretion about which trades to hedge today. Those are both ungoverned B-book with extra steps. A real hybrid model is a classification system with written rules, automatic routing where possible, and a review loop that checks whether the classification is still earning what it should.
Revenue and risk profile
A-book revenue is predictable and thin: a fixed markup per lot, minus LP and bridge costs. It scales linearly with volume and carries no market tail risk, which is why regulators and risk-averse owners like it and why the margins alone rarely sustain a small broker.
B-book revenue is fat and volatile. In normal conditions the internalised book earns the spread plus client net losses, which is why most retail brokers depend on it. The tail risk is real: a trending market full of one-sided client positioning, or a cluster of sharp accounts, can produce a month that wipes out a quarter. Illustratively, desks that measure it find a small share of accounts, often under five percent, driving a disproportionate share of B-book losses.
Hybrid tries to keep the B-book's economics while capping its tail. Done well, it earns more than pure A-book with materially less risk than pure B-book. Done badly, it inherits the worst of both: LP costs on flow you could have internalised, and tail risk from flow that should have been hedged.
Client classification and routing rules
Hybrid routing lives or dies on classification. The inputs are behavioural: hold times, win rates by tenure, timing patterns around news and price updates, instrument concentration, and network signals like coordinated entries across accounts. The output is a routing profile per account or group, executed on the platform and bridge.
Two disciplines matter. First, classification must be continuous, not a one-time onboarding decision: accounts change behaviour, and last quarter's safe flow can be this quarter's toxic cluster. Second, routing changes must be documented and consistent. If two dealers would classify the same account differently, you do not have a routing policy; you have two routing policies.
Running hybrid well
The operational requirements are unglamorous: exposure limits per book with written responses when breached, automatic rerouting rules for flagged clusters, markout analysis to verify the classification is working, and a regular review cadence where routing decisions are audited against outcomes.
The bridge is where policy becomes reality, and where it quietly stops being reality. Symbols added with default routing, LP changes applied partially, temporary rules never rolled back: bridge configuration drift turns a considered hybrid policy into an accidental one. The fix is treating bridge configuration as an audited artifact, which is the core of our liquidity and bridge management work.
Hedging discipline is the other half. When the B-book breaches limits, the hedge has to happen at the defined threshold, at the defined size, through the defined LP, on every shift. Desks that hedge by feel discover their actual risk appetite only in the worst month. Our hedging and exposure management service exists for exactly this discipline.
Conflict-of-interest and regulatory points
B-book creates a structural conflict: you profit when clients lose. Regulators in most serious jurisdictions know this and expect you to manage it: documented best-execution policy, conflicts disclosure, and evidence that routing decisions follow rules rather than P&L convenience. Even where regulation is light, the commercial version of the same risk applies: clients who suspect the desk trades against them unfairly leave, loudly.
The practical defence is process. Written routing rules, consistent application, and an audit trail showing decisions were made by policy rather than by position. A broker that can produce that trail has an answer to both the regulator and the unhappy client. A broker that cannot has a vulnerability.
There is also a commercial conflict inside hybrid books that deserves honesty: the desk that benefits from B-book P&L is the same desk deciding what gets internalised. The standard control is separation of incentives: routing rules set by policy and reviewed by someone whose bonus is not the book's P&L, with markout data published to management so the classification's quality is visible. Without that separation, the boundary drifts toward whatever internalises more, and the tail risk grows quietly.
Choosing a model by broker stage
Startup brokers usually begin heavily A-booked: the book is small, a single sharp client is an existential risk, and capital reserves are thin. As volume and reserves grow, internalising clearly unsophisticated flow improves unit economics, and the hybrid boundary expands with the data to classify confidently.
Established brokers run mature hybrids with the boundary tuned by evidence. The failure mode at this stage is not the model but the monitoring: classification going stale, drift in the bridge, and coverage gaps in the sessions nobody is watching. If you are unsure where your boundary actually sits today, as opposed to where the policy document says it sits, that is what a Health Check measures.
One stage-specific note for prop firms: the model question looks different because the firm is not intermediating client trades at all. The analogous decision is how much funded-account exposure to hedge externally, and the same logic applies: classify by evidence, hedge the flow you cannot afford to carry, and write the rules down. The firms that get this wrong discover it at payout scale, which is the most expensive classroom.
Frequently asked questions
Is B-book legal?
Yes, in most jurisdictions, under a dealing-on-own-account or market-making permission with appropriate disclosure and capital requirements. The regulatory burden is real: conflicts management, best execution and capital adequacy. Check your specific licence conditions, and treat the compliance work as part of the model's cost.
What percentage of flow should a hybrid broker internalise?
There is no universal number; it depends on your client mix, capital reserves and risk appetite. The right approach is classification-driven: internalise the flow your evidence says is unsophisticated, hedge the rest, and let the percentage be an output of the classification rather than a target.
How do we know if our classification is working?
Markout analysis: where price trades seconds and minutes after fills from each classification bucket. If your internalised bucket shows consistently adverse markouts, sharp flow is leaking into the B-book. If your hedged bucket shows benign markouts, you are paying LP costs on flow you could have kept.
Can a small broker run a B-book safely?
Only with tight limits, small maximum exposure per symbol, and disciplined hedging at defined thresholds. The danger for small brokers is concentration: one large client or one trending day can exceed what the balance sheet absorbs. Many small brokers run A-book by default and internalise only capped, classified flow.
Does the model choice affect which liquidity providers we need?
Yes. Pure B-book needs minimal LP relationships; hybrid needs reliable hedging LPs sized to your worst-case hedge flow, not your average. A common mistake is provisioning LP capacity for normal conditions and discovering, during a volatile session, that hedge fills degrade exactly when you need them most.
Deltar designs and operates routing and hedging policy for brokers on MT4, MT5, cTrader, Match-Trader, TradeLocker and DXtrade: liquidity and bridge management, hedging and exposure management and the outsourced dealing desk that runs it all. Start with a Health Check of your current book.