Aida CTO Warns: Why Correct Arithmetic Isn't Enough for Financial AI

Tom Richards, CTO of Aida by MoneyMeans, has released a White Paper titled ‘Where the Model Stops’, addressing the UK’s financial advice gap. For fintech leaders, the report outlines a critical framework for deploying customer-facing AI that prioritises controlled decision layers over raw language models to ensure regulatory compliance and mathematical accuracy in financial planning.

What was announced

The White Paper argues that generic AI models often provide incomplete or "wrong-shaped" answers when applied to complex financial scenarios. To bridge the advice gap effectively, Richards proposes a system where a language model acts as the "mouth" while a separate, controlled financial-planning logic serves as the "brain." This architecture ensures that AI-driven advice is based on fixed, version-controlled rules rather than the unpredictable outputs of a Large Language Model (LLM).

The paper specifically targets the deployment of Aida, a digital financial adviser currently in development by MoneyMeans. Aida is designed to examine a user's actual financial data and provide plain-language guidance. Under this model, the system must collect specific information, consider financial alternatives—such as comparing mortgage overpayments against pension contributions—and maintain a reproducible audit trail of inputs and assumptions. The system is built and supervised by regulated financial planners to ensure the logic remains sound even if the conversational model changes.

Key technical requirements highlighted include the need for a "stop" mechanism where the system refers customers to human review if it lacks sufficient data or moves outside its permitted scope. Testing protocols must evaluate not just the conversational fluency of the AI, but the underlying calculations and the ultimate regulatory outcome. The goal is to make high-quality, regulated financial advice accessible to those currently priced out of traditional planning services in the UK market.

"Overpaying a mortgage and contributing to a pension can compete for the same money over roughly the same horizon, and for many people the pension wins. A system that only answers the mortgage question can be mathematically right and still fail to assess the decision in front of the customer."

Tom Richards, CTO of Aida by MoneyMeans.

The companies involved

MoneyMeans is a UK-based financial services firm focused on making high-quality financial advice more accessible through technology. The company is currently developing Aida, which is positioned as an AI-driven financial adviser. Aida is designed to work from a structured fact-find and connected financial information, using human-approved financial-planning logic to determine outputs that are then explained in plain language by a language model.

The project, led by CTO Tom Richards, seeks to solve the "advice gap" in the UK market—a long-standing issue where a significant portion of the population has financial needs that are too complex for simple execution-only platforms but are not wealthy enough to afford bespoke human advisory services. By separating the conversational interface from the logic engine, MoneyMeans intends to create a scalable, compliant solution that can withstand regulatory scrutiny while remaining user-friendly. The firm emphasizes that while the language model handles the explanation, the core financial outputs are determined by code that applies relevant planning rules and requires the system to compare material alternatives.

What this means

This announcement highlights a pivotal shift in how fintechs approach generative AI. The industry is moving past the "chatbot" phase toward highly structured, "logic-first" architectures. By advocating for a controlled decision layer, MoneyMeans is directly addressing the "hallucination" risks and lack of transparency that have made regulators wary of AI in financial services. This puts pressure on traditional wealth managers to either automate their own logic or risk being undercut by these more agile, AI-augmented platforms. The real test for the sector will be whether these hybrid systems can truly replicate the holistic "fact-find" process of a human advisor without losing the trust of the consumer.