Can agentic AI transform the retail brokerage industry?

Retail brokers may be approaching an inflection point as rising compliance costs, expensive client acquisition and limited product differentiation put increasing pressure on profit margins.

That was the central message from a recent Broker Club discussion on agentic artificial intelligence in financial services, led by Alan Thomas and Alex Vanderlip, partners at, AI consultancy Dailoqa specialising in enterprise agentic Ai applications for banks, brokers and other financial institutions.

Retail trading activity surged during the pandemic, but Dailoqa argued the growth continued well beyond that initial spike. Active CFD trading accounts globally roughly doubled between 2021 and 2025,  and many brokers have struggled to translate that larger customer base into proportionately higher profits.

The industry therefore faces a choice: consolidate through acquisitions, continue outsourcing and white-labelling technology, or use artificial intelligence to redesign how the business operates.

Retail brokers are struggling to differentiate

One of the biggest problems facing trading platforms is that most offer broadly similar products.

Retail traders can access the same major currencies, indices, commodities, shares and cryptocurrencies through numerous providers. Trading platforms may look different, but the underlying services are often difficult for customers to distinguish.

This encourages clients to open accounts with several brokers and move between them depending on which company is offering the best promotion, lowest trading costs or most attractive welcome offer.

Client acquisition has consequently become extremely expensive. During the discussion, the cost of acquiring an active trading customer was estimated at around $1,100, although this can vary significantly by country, product and marketing channel.

Spending more money on advertising may increase account openings, but it does not necessarily produce more profitable or loyal customers.

In contrast, some brokers have concentrated on rapid customer acquisition while others have prioritised extracting more value from existing clients.

The real commercial challenge is therefore not simply winning a client. It is giving them a reason to continue using the platform.

Plus500 presented as an efficiency benchmark

Plus500 was highlighted during the discussion as an example of a broker that has built an unusually efficient operating model.

Rather than relying heavily on manual processes, the company has spent years designing much of its technology and infrastructure internally.

The presentation cited annual revenue of approximately $792 million generated with a workforce of around 700 employees. This equates to more than $1 million of revenue per employee.

This level of efficiency took Plus500 approximately a decade to create through traditional technology development and process redesign.

Agentic AI could potentially allow other financial firms to achieve similar operational improvements much more quickly, although implementing it safely in a regulated environment remains a significant challenge.

Moving an existing process to a lower-cost country may reduce expenses, but it does not necessarily make the process itself more efficient.

True transformation requires the company to reconsider why the process exists, what outcome it is trying to achieve and whether parts of it still need to be performed by people.

How can agentic AI be used in trading platforms?

Most financial companies have already experimented with large language models through products such as Microsoft Copilot, Google Gemini and ChatGPT.

However, these systems are frequently used as enhanced search engines or writing assistants rather than being integrated into the underlying operation of the business.

Agentic AI goes further.

An AI agent can be given an objective, connected to relevant systems and allowed to complete a series of tasks to achieve that objective.

Traditional software generally follows a predetermined sequence. If a process changes or an unexpected exception occurs, the software may stop or require manual intervention.

An AI agent instead focuses on the required outcome and can potentially identify alternative ways to complete the task.

For brokers, this could include checking customer documents, completing onboarding processes, reconciling payments, reviewing corporate actions, updating internal systems or producing customer communications.

One example presented at the event involved a financial company manually processing an upcoming transaction report containing hundreds of pages.

Employees were required to extract information, enter it into spreadsheets, update systems and contact affected clients.

The proposed agentic system could extract the relevant information, interpret it and prepare the required client notifications in minutes.

Hyper-personalisation could improve client retention

The most visible application of agentic AI may be the creation of highly personalised trading tools.

Retail traders do not have the same support network as professional traders working on an institutional trading floor.

• A professional trader can discuss an idea with colleagues, ask someone to challenge their reasoning or draw on the experience of specialists covering related markets.
• A retail trader is more likely to make decisions alone, based on general market commentary, social media, news articles or educational videos.

An AI system could potentially reproduce some elements of the trading-floor experience.

It could consider the customer’s existing portfolio, previous trading activity, preferred markets, knowledge level and risk profile. It could then combine this information with market data, research, technical analysis and scheduled economic events.

Rather than sending the same generic market email to every customer, a broker could provide information relevant to each individual user.

For example, a customer with significant exposure to US technology shares might receive an explanation of an upcoming inflation announcement and how market volatility could affect their portfolio.

The system could also adjust the complexity of its language depending on the customer’s experience.

A beginner may need an explanation of what a stop loss is, while an experienced trader may prefer a concise summary using industry terminology.

This type of personalisation could make a trading platform more useful and potentially more difficult for the customer to leave.

As one participant observed, users may remain with an AI service because it has accumulated an understanding of their preferences, history and behaviour that would need to be rebuilt elsewhere.

When does a helpful prompt become regulated financial advice?

Dailoqa said their retail-facing systems are designed not to tell customers explicitly to buy or sell a particular investment.

Instead, the system may highlight relevant information, explain possible risks or ask the customer to reconsider an assumption.

For example, rather than telling someone to buy an asset, it might point out that an important economic announcement is due within 12 hours and explain how it could affect market volatility.

The distinction is important but may not always be straightforward.

Brokers would need to define what their systems are permitted to say, taking account of the FCA Handbook, Consumer Duty requirements and the company’s own internal policies.

The discussion suggested that AI could ultimately improve compliance because every interaction can be recorded, time-stamped and reviewed.

A firm could potentially retain a complete record of the customer’s question, the market information available at the time, the rules consulted by the system and the reasoning behind its response.

This could provide a clearer audit trail than a conversation between a client and an employee, particularly if that conversation is reviewed several months later.

Reducing the risk of AI hallucinations

General-purpose large language models can produce convincing answers that are inaccurate, outdated or inconsistent.

That makes them difficult to use in regulated financial services without additional controls.

Large language models should mainly be used to understand natural-language questions and communicate answers.

The calculations, rules and actions behind the answer should rely more heavily on deterministic systems, defined data sources and narrowly focused agents.

Each agent can be given a specific role and prevented from operating outside it.

The system must also monitor communications between agents and detect gradual changes in behaviour.

The risk is not necessarily that an AI system suddenly produces an obviously dangerous response. It may slowly drift away from its original instructions over thousands of interactions.

This requires firms to implement controls around explainability, accountability, human oversight, data protection and the ability to shut down individual agents.

Businesses must also avoid becoming dependent on a single AI model or technology provider. Their infrastructure should allow one model or agent to be replaced without rebuilding the entire system.

The biggest opportunity may be behind the scenes

Hyper-personalised trading tools are likely to attract the most attention, but the greatest financial benefit may come from less visible operational processes.

Customer onboarding, Know Your Customer checks, transaction monitoring, payment reconciliation, corporate actions and regulatory reporting all involve repetitive work and large numbers of exceptions.

These processes are expensive because they require significant numbers of employees while creating little obvious differentiation for customers.

Agentic AI could allow brokers to automate more of this work while escalating complicated or unusual cases to experienced staff.

The potential result is a business that processes more customers without increasing its headcount at the same rate.

Brokers need a clear strategy before adopting AI

The message from the discussion was not that brokers should add AI to every part of their business immediately.

They first need to decide what type of provider they want to be, which customers they want to serve and what experience they want to offer.

Without a clear strategy, firms risk buying a collection of disconnected AI tools that do not communicate with each other and require separate compliance frameworks.

Agentic AI is not simply another chatbot to place on a website. Used properly, it could change how a brokerage acquires customers, supports traders, completes administrative work and meets its regulatory obligations.

The technology may allow brokers to achieve in two years some of the operating improvements that previously required a decade of conventional transformation.

But the firms that benefit will not necessarily be those that deploy AI first. They will be the ones that combine it with reliable data, clearly defined business rules and strong regulatory controls.