By Daniel Shepherd, CEO, BTON
The evolution from quantitative trading to machine learning isn’t a revolution—it’s a logical progression. For two decades, quantitative methods transformed execution from intuition-led art into structured science. But as markets fragment and regulation tightens, static models no longer meet the demands of modern institutional workflows.
At BTON, we’ve taken that next step. Using explainable AI, we’ve built a broker selection engine that learns continuously from execution data—turning the conventional algo wheel into an intelligent decision assistant.
Static Wheels Are Not Enough
Most trading desks still use version 1.0 of the algo wheel—rotating broker algos randomly which the intent to achieve a specific percentage of flow to each broker, and reviewing performance quarterly. The intent is fairness. The reality can be inefficiency. These wheels ignore the specific context of each trade and rely on backward-looking opinion rather than forward-looking evidence.
BTON sharpens this. Our AI-driven model evaluates broker algos in real-time using order characteristics, and selects the optimal broker. Recommendations are explainable and auditable, with full transparency and trader override options.
It’s not just about automation. It’s about intelligent automation—bringing the same kind of tailored decision support you get from your phone’s AI assistant to multi-million-dollar execution decisions.
Collaborative Data, Not Compromised Control
Unlike firms that rely only on their own TCA, BTON securely aggregates anonymised execution outcomes across clients. This collaborative data approach improves statistical relevance while preserving confidentiality. It enables us to build a benchmark for broker performance that is both neutral and industry-wide. Our goal is to fairly raise the bar for execution quality across the industry.
The more data we see, the better our model performs. This is what makes machine learning so compelling: it doesn’t just scale—it compounds.
A Better Alternative to Outsourcing
Some buy-side firms are turning to outsourced trading to manage costs and scale. But this often means less control, less transparency, and more complexity. BTON offers a smarter alternative: retain your desk, but enhance it with infrastructure that thinks, learns, and adapts.
You don’t lose control—you gain capability. Traders can set rules, apply automation only where appropriate (e.g., <5% ADV)..
The Industry is Already Moving
This is no longer an experimental idea—it’s becoming the norm. Norges Bank, one of the world’s largest and most respected asset managers, recently overhauled its internal workflow with AI. As CEO Nicolai Tangen put it: “It can’t be voluntary. If you don’t use AI, you will never be promoted.”
At Norges, AI wasn’t just a cost-saver. It identified behavioural biases and inefficiencies that humans overlooked. Crucially, it didn’t replace analysts—it became their analytical partner.
That’s exactly how we see BTON: not as a black box, but as an assistant that learns with you, helping you make faster, more consistent execution decisions.
Built by Practitioners
We have a really great team that reflects this approach. I’ve spent my career building trading infrastructure and sitting on sell-side desks. Adam Mingos ran trading desks at Morgan Stanley and was Head of Electronic Trading at Fidelity. Emmett Cawley began as a physicist at CERN before becoming a quant strategist at global banks and hedge funds.
We’ve seen the problems first hand and have found the solution using technology and data. And we believe now is the moment to move.
Conclusion
Machine learning in execution isn’t speculative—it’s operational. The cost of continuing with quarterly reviews and static allocations is no longer just opportunity loss—it’s strategic risk. AI isn’t replacing the trader, but traders who embrace it will outperform those who don’t.
Version 1.0 served us well. But version 2.0 is here—and it learns.
