BMLL and Features Analytics have partnered to develop new trade surveillance benchmarking and market integrity analytics built on reconstructed historical order book data, signalling a shift towards more measurable, performance-driven surveillance frameworks.
Under the agreement, Features Analytics will build and commercialise surveillance benchmarking products on top of BMLL’s harmonised historical Level 3, 2 and 1 data. The objective is to enable firms to independently assess how effectively their existing surveillance systems detect potential market abuse, using reconstructed order book behaviour as the analytical foundation.From Calibration to Measurable Performance
Trade surveillance has traditionally relied on rule-based, parameter-driven systems requiring continual calibration. As markets have fragmented and message volumes increased, firms have faced rising operational burdens associated with high alert volumes and false positives.
“Effective trade surveillance and market integrity analytics depend on high-quality historical data,” explains Cristina Soviany, CEO and Co-founder of Features Analytics, in conversation with TradingTech Insight. “To reconstruct market behaviour and measure risk properly, firms need deep, harmonised datasets – exactly what BMLL provides. Many institutions still rely on parameter-heavy legacy systems that produce large volumes of false positives and demand constant recalibration. As the industry shifts towards measurable coverage benchmarking and regulator-ready evidence trails, our collaboration addresses this gap by combining our native AI detection technology with BMLL’s high-quality Level 3 historical data to accelerate the development of proprietary benchmarking and risk measurement solutions for trade surveillance.”
The partnership is positioned as a move beyond alert generation towards quantifiable performance measurement, allowing firms to assess detection quality more systematically across varying market conditions.
Why Level 3 Data Matters
At the core of the collaboration is the use of granular historical order book data. Level 3 data, which captures order-by-order message events, enables detailed reconstruction of market microstructure and trading behaviour that is not available through top-of-book or aggregated depth data alone.
“Many existing surveillance systems simply do not ingest the granular data required to detect sophisticated behaviour in modern markets,” comments Paul Humphrey, CEO of BMLL. “Take spoofing or layering. Without full order-level data, how can you identify it? A Level 3 metric such as cancelled orders – placing and then cancelling an order – is fundamental. In fragmented markets, a participant might submit multiple buy orders on one venue and cancel them, while simultaneously placing a sell order on another. Detecting that pattern requires complete, cross-venue order-level data. Many surveillance incumbents do not process data at that depth, but they do expect these behaviours to be detected. Cristina’s approach recognises that firms may not immediately replace existing infrastructure; instead, they can augment it. That is what makes this compelling – materially improving detection capability without requiring wholesale system change.”By combining Level 3 reconstruction with AI-driven analytics, the firms aim to provide an additional layer of detection insight alongside incumbent systems.
Introducing Surveillance Benchmarking
A central element of the partnership is the introduction of what the companies describe as a new category of surveillance benchmarking data. The intention is to allow firms to measure the performance of their incumbent surveillance stack over time and across real market conditions.
For compliance and surveillance teams, objective performance measurement has historically been difficult. While firms calibrate models to manage alert volumes and satisfy regulatory expectations, statistically grounded measures of detection quality are less consistently applied across institutions.
“Most firms run complex surveillance stacks that require continual calibration, yet lack an independent way to measure performance over time and across real market conditions using detailed order-level data,” notes Soviany. “Our benchmarking analytics introduces a structured, like-for-like approach to measuring detection performance, grounded in harmonised historical data, particularly Level 3, enabling firms to produce regulator-ready evidence trails based on reconstructed market activity.”
The approach is designed to support like-for-like benchmarking of detection rates and provide explainable evidence trails grounded in reconstructed order book behaviour.
Enabling Innovation Through Data Access
The collaboration is anchored in Features Analytics’ participation in the BMLL Activate: Data Credits Program, which provides selected partners with time-limited access to BMLL’s historical order book data and research environment during product development, alongside a defined path to commercial deployment.
The structure of the programme reflects a broader strategy by BMLL to encourage application-layer innovation on top of its data infrastructure. Historically associated with market quality analysis, back-testing and pre- and post-trade analytics, BMLL is increasingly extending its use cases into market integrity and surveillance benchmarking.
“The Data Credits Programme enables us to partner with firms like Features Analytics and provide global, full-depth data access during development – access that might otherwise be prohibitively expensive,” says Humphrey. “Two years ago, we would not have been in a position to support this. Our expanded global coverage and our position following the Nordic Capital acquisition now allow us to back innovation in this way. Historically, commercial barriers have constrained this kind of development. By removing them and enabling differentiated products to reach the market, we create value for our partners, for BMLL and for the wider industry.”
From a broader market structure perspective, the initiative aligns with increasing regulatory scrutiny of surveillance controls, growing emphasis on explainability, and pressure on firms to reduce operational costs. Should benchmarking frameworks gain traction, surveillance may evolve from a predominantly calibration-driven function to one assessed through measurable performance indicators grounded in granular market data.
While the companies have not disclosed initial client deployments or performance metrics, the positioning of surveillance benchmarking as an independent measurement layer suggests potential relevance across banks, brokers, exchanges and regulatory bodies seeking greater transparency into the effectiveness of their surveillance stacks.
Subscribe to our newsletter
