Ivan Kunyankin, Data Science Team Lead at Devexperts.
AI may have started out its brokerage career in back-office, enhancing operational efficiency by providing human teams with actionable client insights, but it’s now being promoted to more sensitive client-facing roles.
As AI tools continue to evolve and become normalized in more areas of daily life, financial services providers are starting to recognize AI’s value as a front-facing agent capable of shaping user experiences for end traders. AI agents are starting to resemble the personable virtual assistants we all hoped the information age would yield.
User-facing features
For beginners, AI agents are able to aid in developing platform familiarity and confidence in taking those initial first steps to placing a trade. They can share educational content that’s relevant to the client’s current level or prompts regarding what they appear to be struggling with.
If given access to the broker’s knowledge base, they can also answer specific queries, which enhances user experience while leaving human support teams free to address more urgent issues.
For experienced market practitioners, AI can help filter out the noise of the trading session. Signal to noise is something that traders struggle with on a daily basis. The best of them are able to disregard the abundance of available information, homing in on only what serves their strategy. AI can help by curating news, upcoming events, and analyses that are relevant to individual traders, as well as suggestions for further research, allowing them to discover new assets that share characteristics with their current favourites.
These capabilities serve traders and brokers alike. A bespoke experience improves the client journey and increases satisfaction with the broker’s offering. This leads to increased engagement, improved retention, and a higher lifetime value per trader.
Broker-facing features
Traders produce vast amounts of data as they interact with trading platforms, favouring certain assets, trade intervals, and risk management practices over others.
This data has always been viewed as valuable, but brokers lacked the tools to utilize it productively in real-time. Current machine learning techniques allow brokers to leverage client behavioural data in order to better understand their interests, sentiment, and risk appetites.
This allows brokers to offer pre-emptive support, for example when a trader is attempting to fund their account and the transaction is repeatedly failing to go through, or when they would benefit from an educational primer on a certain aspect of trading. Brokers can also use the above data to send clients a personalized report of their trading activities over the past week or month.
AI tools can also be trained to predict drop-offs, intervening at those crucial moments to solve a client problem or prompting a human colleague to reach out.
Today, the big difference is the immediacy of the interventions that can be made. Rather than analysing data after the fact and using it to inform broader sales and retention strategies, AI agents can operate proactively, responding to what’s taking place on the platform in real-time in order to offer timely suggestions and support.
All the above are possible even without the collection of personal data as it’s the in-platform behavioural patterns of traders that offer predictive power, rather than anything that could be used to identify individuals.
Getting the most out of assistants
We’ve found that an efficient way of using machine learning approaches in combination with Large Language Models, is to try and push LLM operations as downstream as possible. This is due to how resource intensive LLM queries currently are.
If the analytical heavy lifting is performed via more efficient means, the output of these processes can then be plugged into an LLM in order to produce the final customized communications. This is especially relevant now, until efficiency and latency become less of an issue with LLMs. In our experience it’s possible to knock average response times down to under a second in this way.
On the broker side, the effectiveness of these assistants is only limited by what data and tools each venue makes available to them. This means that the same third-party AI assistant deployed by two different brokers may possess entirely different qualities and capabilities. This allows thoughtful integrations and on-going customization initiatives to stand out as these systems become more widespread.
To conclude
We believe that the industry is just getting started when it comes to AI assistants. As the technology becomes more accessible and easier to experiment with, brokers are moving beyond the low hanging fruit of FAQ chatbot use cases to more advanced proactive support provided in real-time.
As these systems improve and become more embedded in different trading infrastructures, we expect them to become the standard method for traders to interact with trading platforms and conduct market research.
Subscribe to our newsletter
