TS Imagine is taking its use of artificial intelligence beyond information retrieval and analysis and into trading, portfolio and risk workflows with the launch of TSIQ, an AI platform designed to explain, recommend and – subject to user-defined controls – act.
The launch follows a five-year, $100 million investment by the company in data, infrastructure and financial context, including the development of a proprietary ontology designed to give consistent meaning to the data used by AI models. TSIQ sits across TS Imagine’s trading, risk, portfolio, wealth and prime brokerage platform and is available as an additional licensed service.The release marks a significant progression from the AI use cases Thomas Bodenski, COO and Chief Data & AI Officer at TS Imagine, discussed with TradingTech Insight early last year. At the time, the company was already using AI in production to automate labour-intensive processes, including data management and customer support, while retaining human responsibility for critical decisions. The ambition with TSIQ is to bring AI much further into the workflow.
“I can’t get excited about chat anymore,” Bodenski tells TradingTech Insight. “That was cool when it first came out, but building a solution that can explain things is now the absolute bare minimum. For us, the key was recommending and acting, which are much more difficult pieces to establish. What we found from adopting AI internally was that it worked best when it was seamlessly integrated into the workflow. If you’re able to recommend and act, you can integrate AI seamlessly into those workflows, and that enables much more powerful adoption.”
Building the foundations
Getting to that point has required more than adding large language models to existing applications. That work has since moved through a year-long beta programme involving around 30 clients, giving TS Imagine an opportunity to test TSIQ in a range of institutional workflows before making it generally available.“This is five years in the making,” says Bodenski. “We’ve put more than 50,000 person-days into it. You can’t simply say tomorrow, ‘I want AI’. It requires a lot of work, a lot of thought and a lot of testing, and you need to have your AI delivery lifecycle fully sorted out. You need to be able to version-control those agents, monitor them and trace them, but you also need to evaluate what they’re doing, rank them and score them so that you can intervene. Otherwise, there’s no ability to delegate.”
That infrastructure builds on TS Imagine’s existing unified data layer, which spans its execution, risk and other applications. TSIQ adds a semantic layer designed to give the AI an unambiguous understanding of the financial concepts represented by that data.
Giving financial data meaning
For Bodenski, that semantic layer is central to making AI usable for trading and investment decisions. “If you asked me to pick one thing that’s most important for making this successful, it’s the ontology – an ontology that’s embedded into your ecosystem and fully connected to it, not a disconnected ontology that’s generic and abstract. You can have a number that’s accurate but still wrong for your decision. That’s what the ontology solves. The data needs to be believable, but the insights that the AI generates also need to be believable, and they’re not going to be, unless everything is clearly defined,” he says.
Average daily volume provides a relatively simple example. An ADV figure could refer to a local listing, primary listing or composite volume; it could cover five, 30 or 180 days; and it could include reported liquidity that isn’t actually accessible to the institution. All could produce technically accurate numbers while leading to very different execution decisions.
The same problem applies to concepts such as exposure. A portfolio’s exposure to a company or investment theme could encompass direct holdings, derivatives, ETFs and index positions. Establishing those relationships within the ontology allows TSIQ to interpret questions and recommendations against consistent definitions rather than relying on the LLM to infer what the user means.
Connecting portfolio decisions to execution
Bodenski gives the example of a portfolio manager investigating exposure to an investment theme. TSIQ could identify existing direct and indirect exposures, incorporate external research and propose portfolio changes. Those changes could then be passed through TS Imagine’s existing scenario analysis, stress-testing and value-at-risk capabilities before moving into execution.
At that stage, established trading infrastructure remains responsible for functions such as compliance and limit checks, order routing and execution. The AI effectively provides an orchestration layer across capabilities that already exist within the platform rather than replacing deterministic trading and risk controls. The degree to which that workflow can proceed without intervention is configurable.
“It’s not all or nothing,” notes Bodenski. “You define the appropriate level of delegation for whatever use case you have. Even within a particular use case, you can say that for one set of criteria you want to delegate more, while for another you want to delegate less and be more in the loop. People are at different stages of AI fluency, firms are at different stages, and even departments within firms are at different stages. You have to give people the chance to get comfortable with this.”
TS Imagine characterises those models as human “in”, “on” or “over” the loop, allowing institutions to vary the level of supervision according to the workflow and their confidence in the technology.
Keeping an audit trail
Allowing AI to participate more directly in workflows also raises the question of whether firms can reconstruct why an action occurred.
“You need to be able to understand what has happened and connect the orders back to the conversation that you had – full traceability and full transparency, end to end,” says Bedenski. “TSIQ has to answer the questions that you ask yourself. It shouldn’t require anyone to read anything outside or talk to anyone outside the platform. It must provide all the answers you need as a compliance officer, as an auditor, as the user or as the boss of the user.”
That traceability is closely tied to the broader AI delivery lifecycle Bodenski describes. Agents need to be monitored, evaluated and version-controlled, while the actions they initiate can be linked back through the workflow that produced them. The ability to delegate therefore depends as much on the control framework surrounding the AI as on the capabilities of the underlying models.
TSIQ’s common semantic model also allows it to operate across execution, risk, portfolio management, wealth and prime brokerage rather than as a standalone AI application. Support for Model Context Protocol (MCP) is intended to extend that approach beyond TS Imagine itself. According to Bodenski, TSIQ can be plugged into external workflows through MCP, while capabilities from other providers can also be brought into the TSIQ environment, providing a mechanism for agents and systems to interact without rebuilding the underlying semantic definitions for each connection.
With TSIQ now generally available following its year-long beta programme, TS Imagine is moving from testing the technology with selected clients to commercial deployment. The next stage will show how far institutions choose to move along the path from AI-assisted explanation and recommendation towards controlled delegation of actions within live capital-markets workflows.
Subscribe to our newsletter
