Best use of machine learning/AI
In the fast-paced world of capital markets, precision, speed and reliability are essential. CompatibL – a leader in risk and analytics technology – has taken an innovative approach to automating trade entry for what-if analysis, addressing one of the industry’s most persistent challenges. By harnessing the power of large language models (LLMs) and designing workflows that emulate human problem-solving, CompatibL’s solution has transformed this process. This ground-breaking work, embedded within the CompatibL Platform v10.0, earned the company the prestigious Best use of machine learning/AI award at the 2025 Risk Markets Technology Awards.
CompatibL’s innovation goes beyond technological achievement. It tackles a critical, time-sensitive pain point in financial risk management – manual trade entry. This process is not only tedious but also error-prone, with potentially serious consequences. This feature explores the motivation behind CompatibL’s journey, the challenges the company faced and how its innovations are setting new standards in the financial industry.
Revolutionising trade entry
What-if analysis plays a crucial role in risk management by assessing the potential impact of a trade on market and credit risk, as well as trading limits. Before a trade is executed, its details must be captured in the bank’s risk systems so it can be evaluated within the context of the existing portfolio. Historically, this task relied on analysts manually interpreting free-form trade descriptions – often from email chains or relayed verbally – and entering the data into structured formats.
Alexander Sokol, executive chairman and head of quant research at CompatibL, describes the inefficiencies of the traditional process: “On a busy trading desk, entering trades under time pressure is a complex task nobody wants to do manually. It’s tedious, error-prone and can cost banks valuable opportunities.”
CompatibL’s artificial intelligence (AI) solution was designed to automate this process, utilising LLMs, which excel at interpreting unstructured data. However, the team realised that existing LLMs were insufficient because of their well-known limitations, such as inconsistency and ‘hallucinations’. Unlike traditional software, LLMs behave more like humans – they excel at interpreting data but can also misinterpret details or produce inconsistent results.
“The key insight for us was recognising that, like humans, LLMs succeed and fail,” Sokol explains. To address this, CompatibL developed workflows that replicate human strategies for ensuring reliability.
Input markup to mimic copy and paste
A key component of CompatibL’s innovation is its use of input markup to replicate the accuracy and reliability of a simple human behaviour: copying and pasting. Sokol emphasised how this deceptively simple approach can have a revolutionary impact: “Rather than typing out or memorising complex data from a trade description, humans simply copy, clean it up and paste it. It’s reliable because there’s minimal room for error,” says Sokol.
Traditional LLM workflows typically involve extracting data directly from a trade description and reassembling it into a structured format. However, this method is prone to errors, as LLMs can misinterpret or alter details in the process. CompatibL’s approach diverges significantly. Instead of asking the LLM to extract data outright, the system marks up the input data, tagging relevant sections such as tables, fields or trade parameters. These tagged sections are then processed by traditional code, extracting the data without any risk of alteration or omission.
The result is a system that performs with a level of reliability exceeding manual methods. “By replicating the copy-and-paste approach, we’ve eliminated one of the major sources of error in LLM workflows,” Sokol says. “It’s a small innovation, but its impact on accuracy is enormous.”
Checklists to tackle complexity
CompatibL’s system shines when handling structured notes and other complex trades. The AI’s ability to capture these complexities accurately is enhanced by the same copy-and-paste-inspired approach. By marking up data fields and breaking down trade descriptions into manageable chunks, CompatibL’s workflow ensures every attribute is processed correctly. The system also uses a checklist mechanism to ensure no critical details are overlooked.
Checklists are particularly valuable when processing structured trades. These trades often include complex attributes such as embedded options, custom coupon schedules or non-standard parameters. The increased complexity of these features makes the process more prone to oversight, especially when analysts are under pressure.
CompatibL’s checklist system plays a pivotal role in ensuring no steps are skipped. It mirrors processes used in high-stakes industries such as aviation, where checklists are crucial to verifying each step in complex operations. “Even highly skilled humans can miss important details under pressure,” Sokol says. “The checklist guarantees every step is reviewed before the trade is finalised.” This attention to detail ensures that even highly technical trade features, such as embedded options or overrides, are correctly captured.
Markets Technology Awards judges highlighted this approach, with one noting that the system “improves accuracy significantly for highly technical trade features that even seasoned analysts might overlook”.
Escalation mechanisms to flag ambiguities
CompatibL also addressed broader challenges associated with LLMs, such as variability and hallucinations, by incorporating escalation mechanisms into its workflows. When the AI encounters ambiguities – such as a trade description with multiple interpretations – it flags the issue for human review rather than making a risky assumption. This ensures that even the most complex or unclear trades are handled correctly, without compromising accuracy.
Another challenge CompatibL overcame is the inconsistent way LLMs handle defaults and conventions, which can vary widely across asset classes and counterparties. To address this, the AI dynamically interacts with reference systems, ensuring defaults are applied accurately while still allowing for manual overrides when necessary. The escalation mechanism integrates seamlessly with the checklist system, ensuring that any unprocessed or ambiguous elements are flagged for human review. “Instead of rushing through and guessing, the model escalates doubts, allowing the human to focus only on unclear elements, not recheck everything,” Sokol explains.
By working together, the checklist and escalation mechanisms create a safety net that ensures the accuracy of trade entry, even when dealing with the most complex structured trades. This minimises delays and, in effect, eliminates the risk of critical errors.
Going beyond standardisation with interoperability
Sokol also highlighted one of the industry’s long-standing challenges: the lack of standardisation in trade formats. Financial products Markup Language (FpML), a widely recognised initiative to create a standardised trade representation, has seen limited adoption because of the significant costs of implementation.
“The idea behind FpML was excellent,” Sokol remarks. “But many institutions found it prohibitively expensive to overhaul existing systems to align with a common standard. Without universal adoption, seamless trade communication remained out of reach.”
CompatibL’s solution offers a practical alternative by enabling interoperability without requiring complete standardisation. The AI is capable of processing and converting trades between proprietary formats, bridging the gap between systems and enabling institutions to exchange complex trade data without the need for costly infrastructure changes. This innovation reduces operational overhead costs, while making the system accessible to a wider range of financial institutions.
Open-source collaboration at scale
One of the standout aspects of CompatibL’s approach is its decision to make the trade entry solution open source. While many firms treat AI advancements as proprietary, CompatibL has embraced transparency, inviting industry collaboration to refine and expand its technology. This open-source approach not only highlights the company’s confidence in its innovations but also accelerates adoption across the financial ecosystem.
Sokol explains: “We’re building a community where contributors can use and enhance our tools, creating solutions that go beyond what a single firm could achieve alone.”The response from the industry has been overwhelmingly positive, with financial institutions and developers engaging directly with CompatibL’s technology. A key driver for this engagement is the company’s user-friendly online playground, where firms can test the system using their own trade data. This platform allows users to integrate their preferred LLM models, test workflows and suggest improvements through collaborative forums such as GitHub.
This open-source model has catalysed innovation, with participants contributing new use cases, adapting technology for niche requirements and refining existing workflows.
Flexibility in model selection drives adoption
Another key strength of CompatibL’s system is its model-agnostic design. The platform supports LLMs from providers such as OpenAI, Meta (Llama), Anthropic, Google and Mistral, enabling seamless integration into existing bank infrastructures. This flexibility is essential for financial institutions, which often face strict compliance and security requirements.
“Banks typically approve only one or two LLMs due to regulatory constraints,” Sokol explains. “Our system works across a wide range of models, ensuring compatibility regardless of the bank’s choice.”
Clients have praised this adaptability. One institution, for instance, used the solution to process sensitive data securely on-premises using an open-source LLM. This flexibility has been instrumental in driving adoption across diverse financial environments.
Unlocking future opportunities
CompatibL’s success with trade entry automation is only the beginning. The company envisions expanding its technology to unlock the value of unstructured data across financial institutions.
“While generative AI often focuses on content creation, we see immense potential in its ability to comprehend data,” Sokol says. CompatibL is exploring how its technology could be used to extract value from historical transaction records, term sheets and alternative data sources.
CompatibL’s Convince library exemplifies this vision. It enhances AI’s ability to process and analyse unstructured data, opening up new applications – from mining insights from historical transactions to analysing alternative data sources for more informed decision-making.
Conclusion
CompatibL’s AI solution for trade entry demonstrates how simple, human-inspired innovations can have a transformative impact. By replicating intuitive behaviours such as copy and paste, it addresses the limitations of LLMs and delivers improved accuracy, efficiency and reliability. CompatibL’s achievements at the Markets Technology Awards reflect the company’s ability to blend cutting-edge AI with practical applications.
By automating trade entry for what-if analysis, CompatibL has addressed a long-standing industry challenge. This focus on solving real-world problems has earned strong praise, with judges highlighting its practical approach and transformative impact. One judge noted the solution’s clear effort to overcome LLM limitations, while another emphasised its relevance to industry pain points. With a commitment to open collaboration and practical innovation, CompatibL is paving the way for the next era of AI in financial services.
Judges’ comments
CompatibL
“Clear effort to overcome LLM limitations, delivering significant benefits in analysing trade impacts.”
“Addressing inherent variability in LLMs is compelling.”
“The solution stands out with its speed advantage, making it a great example of innovation in what-if analytics.”
“Improves accuracy significantly for highly technical trade features that even seasoned analysts might overlook.”
“The accuracy of CompatibL’s LLM-based what-if trade entry has been highly rated by our clients. TradeEntry AI was able to understand highly sophisticated aspects of trades that may not have been grasped by anyone other than analysts with deep experience in a narrow market segment or a complex asset class.”
Alexander Sokol, executive chairman and head of quant research at CompatibL, says:
Markets Technology Awards 2025: all the winners
Sponsored content