Heritage and innovation: How MillTechFX is utilising generative AI

In a previous blog, I discussed how the tech team here at MillTechFX were exploring generative artificial intelligence (AI) for a number of use cases. I discussed our early exploration, checking out ChatGPT’s code writing abilities, testing if it could draft strong policy boilerplates and, most importantly, asking it how to dress for a smart casual-themed party.

In the six months since, our exploration has turned into implementation. We continued to tinker with the technology and are implementing tools which are accelerating our work, ultimately benefiting our customers.

Building on our AI heritage

The Millennium Group has a long track record of using AI to deliver business value. Systematic (or quantitative) investment processes have been a core part of the service offering at Millennium for many years now so it’s safe to say we’re comfortable with the technology.

We applied the same institutional approach to generative AI, that we do with all emerging technology – analysing the risks as we explore the opportunities of the new technology.

Data loss, hallucinations and entrenching bias are just some of the very real risks of generative AI. To manage these, we set out a policy that allows us to identify and mitigate risks that come with a specific use case or generative AI technology choice.

 Key areas of this policy are:

• Input information: Scrutinising the quality and source of input data to ensure its reliability.
• Algorithm choice: Carefully selecting the appropriate algorithm for each use case.
• Output information: Evaluating the output for accuracy.
• Legal and regulatory compliance: Staying informed about evolving regulations and adapting their practices accordingly.
• Human oversight: Maintaining a level of human oversight to ensure responsible AI use.

Rolling out generative AI tools

Our AI heritage combined with an institutional approach to the technology enabled us to rapidly explore and adopt generative AI in a safe, compliant manner. So far, we have focused our initial efforts on two internal use cases:

Speeding up compliance – Part of our offering is to deliver express onboarding with our 15 counterparty banks. We have a unique advantage in that we have significant experience in KYC and AML in foreign exchange (FX). However, even with this experience, it’s still a laborious process. It typically involves reviewing complicated legal documents and finding specific information from within these documents. It can be a bit like finding a needle in a haystack. To improve this process, we built a generative AI tool that allows us to rapidly find these answers and further accelerate the onboarding process.

Creating development efficiencies - As our business grows our code base grows too. As it grows, an individual developer might need to create a feature that requires them to understand a part of the system they have not worked with before. Businesses typically overcome this problem by writing and maintaining documentation, though these efforts create an overhead and are rarely accurate. Instead, we feed parts of our code base to a Generative AI model to allow developers to ask questions to the code and return the true answers rather than searching through documentation, saving valuable time.

Given it offers affordable access, generative AI can solve a whole new class of problem. As a high growth business, there is always much more to do than can be done. Generative AI is enabling our team members to automate time-consuming tasks and accelerate their work, freeing them up to focus on higher value activities.

What’s next for generative AI at MillTechFX?

The pace of change in generative AI is so fast. Open source models that performed poorly when first released, take only days to overtake their competitors. Keeping up with this pace of change is only possible if you get hands-on and work with real use cases.

Regulators are working hard to keep up but, in the meantime, the focus must be on organisations using the technology, acting responsibly and managing the risks.

Beyond operational efficiencies, we are also in the earlier stages of analysing how we can incorporate generative AI into decision-making enhancement tools for our clients. We are using certain machine learning (ML) algorithms to combine macroeconomics, market dynamics, price changes and alternative data sources to generate signals that our clients could use to help them with when and how they execute for their transactional or hedging FX requirements.

We plan to continue exploring the technology as the more we work with it, the more ideas it sparks for applications within our business. We run hackathons to rapidly allow us to explore different use cases and technology variants.

The genie is well and truly out of the bottle with generative AI and any organisation that’s not thinking about how this technology can enhance their offering risks being left behind.