Straight-through processing, internalization, and the hybrid model in between — the honest economics and ethics of each execution model.
BridgingFX editorial · 2026-08-22
Few topics in retail FX generate more heat and less light than A-book versus B-book. Critics paint B-book as a casino; defenders paint A-book as the only honest model; most real brokerages run something in between and say very little about it. Strip away the moralizing and you're left with a risk-management decision with genuine trade-offs. Here's how the models actually work, what each one costs, and how serious operators think about the choice.
A-book (STP): client orders are passed to liquidity providers. The broker earns the markup or commission and carries no market risk on the trade. The client's loss is not the broker's gain — the broker profits from volume, not from client failure.
B-book (market making): the broker takes the other side of client orders internally. The broker's P&L is the mirror of the clients' — profitable when clients lose, exposed when they win. Hybrid: flow is segmented — typically, toxic or informed flow goes to the market (A-book) while uninformed retail flow is internalized (B-book), decided by real-time risk rules.
Pure A-book is clean but thin: revenue is markup times volume, and volume requires serious scale to cover fixed costs. It's also honest in the way clients and regulators prefer — no conflict of interest to explain.
Pure B-book is lucrative but volatile: most retail flow loses over time, so internalization captures that as revenue — but a single unidentified toxic trader, or a market gap through stops, can erase months of profit in minutes. The B-book is a risk business that looks like a revenue business until the day it isn't.
Hybrid is where the industry actually lives: it captures B-book margin on genuinely uninformed flow while A-booking the flow that would hurt. Done well, it's the best risk-adjusted model. Done badly — with crude segmentation — it's the worst of both worlds.
The hybrid model's entire edge is classification: which clients are toxic? The signals are well understood — latency arbitrage patterns, news-spike trading, consistently positive short-term expectancy, HFT-style order-to-trade ratios, correlated accounts. The technology to detect these in real time is mature.
The discipline is what fails. Segmentation rules need constant tuning as traders adapt, and they need human oversight — fully automated classification creates false positives that punish good clients, while fully manual review doesn't scale. The operators who do this well treat it as a living system, reviewed weekly, not a configuration set once at launch.
Regulators increasingly want to know exactly how client orders are handled — best-execution obligations, conflict-of-interest disclosures, and order-handling transparency are tightening across jurisdictions. Whatever model you run, document it, disclose what's required, and make sure your marketing never promises something your execution model contradicts.
Reputation matters too. The brokers with the strongest brands in 2026 tend to be transparent about their model — many now openly describe hybrid execution in their legal documents and even their marketing. In an industry with a trust deficit, 'here's exactly how we handle your orders' is a competitive advantage, not a liability.
New brokers often default to B-book for the margins, then discover the risk management they skipped. Our guidance: start hybrid with conservative segmentation — A-book by default for anything your system flags, B-book only for clearly uninformed flow, with tight per-client and per-symbol exposure limits.
As volume and data grow, your segmentation gets smarter and your internalization can expand safely. The model should evolve with your data, not your ambitions. And whatever you choose, the dealing desk — staffed or outsourced — is not optional. Unmonitored B-book isn't a business model; it's an unhedged bet.
A-book is clean volume economics, B-book is lucrative but volatile, and hybrid — with disciplined, continuously tuned flow segmentation — is where most serious brokerages land. Choose the model your risk capability can actually operate, disclose it honestly, and never let marketing promise what execution can't deliver.
Let's scope it — honestly, and in writing.
