Trade surveillance turns to AI to broaden data capture, but firms should proceed with caution

Financial firms are being driven to adopt ever more high-tech trade surveillance systems by tightening market regulations, according to a report from Juniper Research.¹ The firm expects spending on third-party trade surveillance systems to grow by a sizeable 82% globally by 2029, from $2.7 billion in 2025.

Regulations to protect against market abuse and insider trading are growing and trade surveillance is at the heart of how financial firms can comply. MiFID II, the second iteration of the EU’s Markets in Financial Instruments Directive, requires that companies gather as much data in as broad a scope as possible so that an increasingly complex market can remain transparent for regulators. Fully implemented in 2018, this directive imposed a number of new requirements on compliance teams.

MiFID I had already created multilateral trading facilities as alternative trading venues to increase competition, and a significant proportion of fixed-income, derivatives and OTC securities moved to these new facilities. But that inadvertently decreased transparency and price discovery because of market fragmentation.

With MiFID II, regulators were trying to bring all organised trading back to regulated trading venues or organised trading facilities and required compliance teams to integrate trading data across these venues and markets. MiFID II also required compliance teams to have a thorough understanding of how their algorithmic trading systems work and monitor that trading in real-time.

Best execution for clients also changed, with 65 data points to be recorded for trades, including details like price, costs, speed, execution and settlement, and size. Banks and brokers have a much higher bar to proving that they got their customer the best possible price.

Algorithmic solutions for algorithmic surveillance

In addition to new regulations, enforcement and regulatory actions worldwide have shown increased expectations for trade surveillance. In August 2022 for example, the UK’s Financial Conduct Authority imposed a fine of £12.6m on Citigroup’s international broker-dealer for failures relating to the detection of market abuse under the Market Abuse Regulation (MAR).²

In the US, JP Morgan faced $350 million in fines in 2024 for reporting incomplete trading data to surveillance platforms. Goldman Sachs had in 2023 paid a small fine to the Financial Industry Regulatory Authority (FINRA) for failing to include relevant securities in automated surveillance. And the US Commodity Futures Trading Commission (CFTC) penalised Goldman $3 million in September for surveillance and control failures.³

Regulators increasingly expect financial institutions to be able to produce complete and accurate data across a broad range of regulatory scenarios, incorporating data from multiple sources and channels. This includes not just multiple markets across a complex trading environment, but also diverse channels such as news reports and internal company communications.

As more firms use algorithmic and high-speed trading, and the scope of data grows, firms need to use algorithmic surveillance to keep up.

“To capitalise on a shifting regulatory environment, we urge vendors to leverage AI at the core of their operations. Vendors who fail to implement robust, proactive AI models will lose out to more agile competitors,” said Juniper Research analyst Daniel Bedford.⁴

According to Deloitte, AI-powered methodologies can transform the effectiveness and scalability of trade surveillance. Network and behavioural analysis techniques can be used to surface hidden patterns and identify market manipulation, AI can help to monitor and cross-reference trading data and some systems may incorporate distributed ledger technology with AI.⁵

“Trade surveillance involves data from various sources, such as exchanges, trading platforms, and news feeds. While integrating this data with communication surveillance data can be complex, deploying different AI models, each excelling in various aspects of surveillance, can vastly improve the analysis of suspicious activity. For example, specific models can be trained for speech-to-text translation, analysing news and market sentiment, detecting trade patterns, analysing relationships in the market, and flagging unusual trading behaviour,” Deloitte said.⁶

But the consultancy also made it clear that using AI requires clear objectives, relevant human expertise and oversight, and the right data to really be effective. If AI-powered trade surveillance systems don’t have adequate historical data and a human team that understands its uses and implementations, turning to AI will not improve the efficiency and efficacy of trade surveillance – or meet regulatory requirements.