Beyond A/B Book: Mathematical Routing for the Modern Broker
Every broker with a hybrid model faces the same decision hundreds of times a day: Does this flow go to the book, or does it go to the street? A-book or B-book. Internalize or externalize. Keep the risk or pass it on. For most of the industry’s history, this decision has been made by one […]
Every broker with a hybrid model faces the same decision hundreds of times a day: Does this flow go to the book, or does it go to the street?
A-book or B-book. Internalize or externalize. Keep the risk or pass it on.
For most of the industry’s history, this decision has been made by one of three methods: gut feel, PnL history, or blanket rules. All three methods share the same flaw: they are backward-looking, one-dimensional, and blind to intent.
The result is predictable: Skilled traders get B-booked because their recent PnL was modest—and they proceed to extract value from your book. Churners get A-booked because their volume looks impressive—and you pay LP commissions on flow that carries zero market risk.
The Problem with PnL-Based Routing
PnL is the default routing signal because it is easy to measure. “This account made $50,000 last month, so we A-book them.” The logic feels sound, but it ignores the nature of the profit.
PnL conflates three completely different types of profitability:
- Profitable because Skilled: Genuine predictive ability. Their markout curve shows sustained directional accuracy. They belong on the A-book.
- Profitable because Toxic: Exploitation of timing gaps or feed delays. They are not predicting the market; they are observing a faster version of it. Routing this to an LP just exports the cost—it doesn’t solve the problem.
- Profitable because Lucky: Simple variance. Their PnL will mean-revert. A-booking this flow is wasteful because you are paying LP commissions to externalize revenue-generating flow.
The 3D Framework: Skill, Scale, and Toxicity
Data-driven routing decisions require decomposing each account into three independent dimensions:
1. Skill
Does this account have genuine predictive ability? This is measured by markout persistence—the degree to which the market continues to move in the trader’s favor after execution, sustained over minutes and hours.
2. Scale
What is the potential impact on your book? This is measured by volume, position size relative to equity, and concentration. Scale determines the magnitude of the routing decision.
3. Toxicity
Is this account’s profitability coming at the expense of your infrastructure? Measured by the convergence of latency signals, front-running indicators, and coordination evidence.
The Verdict: A Mathematical Trade Routing Decision
When you measure skill, scale, and toxicity independently, the routing decision becomes a mathematical function rather than a guess.
| Combination | Routing Verdict | Rationale |
| High Skill + High Scale + Low Toxicity | A-Book | Informed flow belongs with the LP to protect your book. |
| Low Skill + Any Scale + Low Toxicity | B-Book | No consistent edge; PnL will mean-revert. Maximize revenue. |
| Any Skill + Any Scale + High Toxicity | Investigate | Flagged flow indicates structural exploitation. Do not route. |
| High Skill + Low Scale + Low Toxicity | Monitor | Genuine ability, but low risk. Useful for market analysis. |
The Economics of Getting It Wrong
The cost of misrouting is not theoretical; it is a measurable drain on your margin:
- A-booking a Churner: You pay LP commissions (e.g., $3 per lot) on flow that carries zero risk. At 500 lots a month, that’s **$18,000 per year** per churner left on the table.
- B-booking a Skilled Trader: A trader extracting $10,000 per month from your B-book is a **$120,000 annual cost** that could have been zeroed out by routing to the LP.
- A-booking a “Lucky” Trader: This is the quietest waste. Every commission paid on flow that would have been profitable to internalize is margin sacrificed for no reason.
From Human Judgment to Mathematical Function
This is not an argument against the dealing desk. The best chief dealers have an intuition for flow quality that no algorithm can replicate. However, intuition does not scale.
When you have 5,000 accounts and 50,000 trades per day, the chief dealer cannot personally evaluate every routing decision. Mathematical routing encodes that intuition into a system that is:
- Transparent: Every verdict is traceable to raw data.
- Auditable: Explainable to compliance and regulators.
- Consistent: Operates without fatigue or bias.
When the COO asks why an account was A-booked, the answer shouldn’t be “I had a feeling.” It should be: “Because the markout persistence is 0.73, the toxicity convergence is below threshold, and the scale-adjusted risk exceeds our internalization limit.”
Trust the Math. Protect the House.
See how intelligent verdicts work →