Two weeks ago we launched DaMoney.AI. As someone at the forefront of building AI guidance in finance in the UK, We’ve seen firsthand how large language models (LLMs) struggle with mathematics and in particular tax calculations.

Despite this, we’ve successfully taught DaMoney to make these calculations more accurately. But…

…we’ve decided to NOT allow DaMoney to provide tax calculations directly in its answers despite having the ability to do so. Instead, users are pointed to dedicated tax calculators.

It’s a good example of “just because you can, doesn’t mean you should”:
AI is a non-deterministic technology and so how a user asks a question influences how an LLM attempts to answer it.

For example a consumer might ask an LLM… “How much CGT do I owe on my shares that I sold for £50,000 last week?”… and the LLM will attempt to answer from the limited information it’s been given.

If the LLM was to give a correct answer then the AI model should theoretically ask a range of ‘fact-finding’ questions to ascertain further necessary details, such as the purchase price, income tax band etc.. Right now LLM’s as standard don’t ask follow-up questions, and that’s a good thing in a financial environment, otherwise they run the risk of crossing from guidance into advice.

The drawback of allowing users to ask tax questions of an LLM in freeform is that it dramatically increases the chances of errors in the answer.

A dedicated “old-school” tax calculator, however, forces a standardised input which ensures a 100% correct answer.

In our view, if an LLM identifies that a question involves a tax calculation and then suggests an appropriate “old school” calculator to use, that’s the best outcome for consumers.

Our job is to provide accurate and reliable guidance, and sometimes that means knowing when to defer to a tool designed for absolute precision with controlled inputs and outputs.

AI can be used by finance companies to provide compliant and useful guidance to consumers, but it has to be built by people who understand the technology, the financial industry, its customers, regulation and who also have a long history in providing guidance.

We’ve launched DaMoney Business (damoney.ai/business) – our API platform that allows organisations to integrate our powerful guidance engine directly into their own business. Our goal is to help organisations provide accurate, compliant, and personalised financial guidance to their customers and employees, seamlessly within their own websites, apps, or platforms – cost-effectively and ethically.

If you’d like to learn more about how DaMoney can help your business, please get in touch.