The speed at which AI has moved from its first steps to viable business applications has been incredible. Early adopters were quick to report positive results, especially when using large data sets, and advances in hardware performance since mean few industries are not exploring the technology’s potential.
Initially, many feared that AI would lead to staff redundancies, but what has become apparent is that the tech serves better as an ally, changing jobs rather than replacing them. For example, in the financial sector, it’s likely that data science will play an increasing part in training courses going forward. There’s also the argument that data scientists might start to specialize in the financial sector, in order that they can wield the power of AI on vast data sets.
In the short term, however, it’s worth being aware of how some parts of AI technology works.
That knowledge suggests it is industry-specific AI that will make the biggest impact on businesses’ financial operations.
Almost akin to automation-on-steroids, AI can handle repetitive, mundane tasks easily. An ability which, on its own, can have positive effects on organizations’ cost burden. But where accuracy in book-keeping, financial forecasting, and regulatory compliance are concerned, mainstream AIs can prove counterproductive.
The ways that the now household-name LLMs (Large Language Models) are trained means their output is only an approximation of accuracy, one that’s often subjective and always dependent on the body of learning data fed into the AI before it’s available for use. Results, therefore, are designed only to give a semblance of accuracy, which in many contexts such as text summarizing, is perfectly acceptable.
Finance professionals’ obvious need for accuracy, however, means only some aspects of generalized AI are suitable for their working environment. AI-powered anomaly detection, for example, is already highly effective at payment gateways run by credit card companies and clearing houses. Similarly effective are processes that can standardize data ready for subsequent processing, such as intelligent document scanning, or smart data cleaning algorithms.
Give a general AI access to private financial data for modeling or trend prediction, though, and there’s a danger the resulting outputs will be affected by the data the model was originally trained on. Those inaccurate predictions can be far less valuable than a seasoned industry expert’s ‘best hunch.’
In addition to the issue of ‘simulation of accuracy’ there are significant concerns about uploading sensitive financial information to a third party, too. The obvious security implications are joined by dozens of red flags raised around data governance and compliance in what is a highly regulated industry.
What’s needed, then, are specifically-trained AIs that guarantee individual companies’ data integrity, that are able to produce answers based on local information, yet do so with transparency – in effect, showing the user its ‘workings-out’ so the skilled human operator can then make more informed decisions.
In the finance function, industry-specific AI can react and adapt to the immutability of pertinent regulations and legislation. That makes modeling possible outcomes accurate in the ‘real world’ and able to create proposed scenarios which are practical and workable. The same ability means that all transactions and processes are compliant without needing additional detailed work by humans to ensure lines aren’t being crossed.
This brings us back to the concept of technologies that can alleviate some of the overhead of necessary-yet-menial tasks among a finance department’s staff. Meniality may never be fully removed (unfortunately), but the nature of tasks will change with greater adoption of specialist AIs. Where a few decades ago a menial task may have meant double-checking columns of figures by hand, the next generation of finance graduates will (hopefully) be skilled in comparing the efficacy of financial models created by AI, and tuning the algorithm with local considerations. Still relatively menial, but bringing greater value.
AI has indubitably brought its successes from ‘out-of-the-box’ implementations: creating social media posts in ChatGPT from blog articles, for example. But for real value to be extracted from the technology, it’s the specificity of AI models that is important.
Finally, there is also the fact that many end users experimenting with AI at present are, by definition, early adopters. Yet for some suppliers, AI has been an integral part of their offering for many years.
Nuance, for example, has used machine-learning models in its voice transcription software since the early 2010s. In the finance sector, CCH Tagetik implemented AI functionality long before OpenAI and Gemini became household names.
What’s more, the platform’s AI underpinnings were (and continue to be) trained on financial data, so the support it offers to finance professionals is relevant and usable. Financial data is kept private, and the AI only learns from the organisation’s figures.
The platform exists to connect finance and operational teams, automate CPM, handle regulations, and produce analysis and predictions. AI makes it work faster and alleviates some of those more menial tasks.
That leaves highly trained staff to learn with and adapt to the AI-powered finance operations of the future.
CCH Tagetik’s ‘AI in Finance: From skepticism to optimism’ report discusses many of the elements of this article, and is available to download here. If you’d like to discuss the possibilities of the CCH Tagetik financial stack, head here to find out more.
