Financial modelling is the practice of inputting financial data into a spreadsheet to forecast a company’s future performance. Their historical performance and future projections will then become available for review after collating their financial, business and accounting information into one place. The company can then use the data for various purposes, including:
Financial modelling is integral to the standard skillset for finance, banking and private equity analysts. It’s also a skill that requires sound judgement, firm knowledge of accounting principles and experience.
But, as m.stanfield – the author of this viral Tweet – suggests, building a financial model in Excel is time-consuming. At its most straightforward – with accessible data and a simple outcome – financial modelling can take an hour. At its most complex, it can take months.
A drawback of highly accurate financial models is the increased amount of data they require. The more data required, the more likely manual data errors will enter the mix, distorting the output. Striking a balance between data quality and quantity is an endemic machine learning problem that researchers are still finding ways around.
Though there are algorithms that can attempt to validate the consistency of data, the capacity for humans to unwittingly introduce errors into their datasets is immense. In a Blackline survey of finance professionals, when asked about concerns regarding data accuracy, 41% of respondents pointed to manual input as the biggest contributor to human errors. Ultimately, AI-powered modelling tools should introduce mechanisms to maintain data accuracy by eliminating the possibility of human error.
AI-driven financial modelling approaches use traditional methods of machine learning analysis combined with natural language processing (NLP) techniques bound together by AI’s decision-making prowess.
Here’s how AI financial modelling tools generally work:
Some offer real-time data analysis. Simply put, as you update the data, the tool adjusts the financial forecasts and scrapes key ratios and other insights.
Often, users will integrate these tools into general financial planning and analysis (FP&A) tools that offer features like data visualisation and budget planning.
FP&A tools like Oracle BI are difficult to learn yet can be incredibly useful for viewing data from different angles, such as visualisation and scenario modelling. Such tools will likely be highly attractive for big companies looking to crunch data.
Of course, a better question than ‘Do AI tools for financial modelling work is, does financial modelling work better than humans? The answer is yes, insofar as it can process more data faster and more accurately depending on the tool. Of course, nothing comes for free, so the downsides of AI-powered financial modelling tools include:
There are more lightweight methods to automate financial modelling that do not necessitate migrating all of your data over to a new platform. By automating parts of the financial modelling process, you can harness AI’s benefits without committing to using a new technology. For example, three-stage financial models often rely on copying the data from a financial statement PDF into Excel. Sophisticated data management tools can extract the data, leverage it to calculate financial ratios and structure it in Excel.
In other words, AI-powered tools can skip the manual work involved in the first stage of collecting the data. For those who value control over their financial modelling, data management tools offer a time-efficient solution.
AI financial modelling tools are good at certain functions but are also under constant development. These tools offer several advantages to the individual analyst and the company if successfully implemented into analytical workflows.
If you work with enormous datasets, working with Excel will be cumbersome and, at worst, impossible. Imagine plugging in millions of data points from decades of financials.
AI financial modelling tools increase the scope of the data you’re working with by cleaning it up and picking out the highlights.
This viral Tweet by Joanna Maciejewska makes a good point – that AI should be giving people more scope to reflect by automating the mundane in exchange for richer, in-depth thinking. The same principle applies here – financial modelling is not a mundane skill but repetitive and tedious for experienced professionals.
The initial promise of AI was that it would automate back-office functions and allow more time for strategising. But in 2024, this doesn’t seem to be happening. Read our take on this here.
Nonetheless, making more accurate data readily available to analysts isn’t a bad thing. Some ideas for the types of activities that analysts should be undertaking instead of financial modelling include:
Financial analysts are generally multi-skilled. So, confining them to repeatedly manipulating data represents a waste of talent and experience.
Despite promising advances in AI financial modelling tools, developers face challenges in improving their models’ performances without compromising data quality or their models’ explainability. Let’s break this down further.
As aforementioned, more data makes AI financial modelling tools more accurate – the same goes for their training, too. However, finding enough legal and accessible financial data to train these models is extremely challenging. Most datasets are proprietary. And the ones that aren’t are likely to be limited in scope, contain errors, etc. When training our models, we were lucky to access a proprietary document store of 25 million documents.
Of course, developers can always create synthetic data to train the model. Making synthetic data that matches the real-life patterns of financial data would be a lengthy process and, ironically, may necessitate financial modelling. Also, quality-checking the synthetic data is time-consuming and can result in distorted outputs if completed carelessly.
Speaking first-hand, the risk of developing AI models for a particular function is that adjusting the context windows, playing with the prompts and altering the training datasets can push the AI model into becoming unexplainable.
The inexplicability of AI is evoked by the ‘Black Box’ model. If a source provides useful information, but the deep learning decisions aren’t available, how far can we trust the data? When dealing with sensitive financial data, that’s not a question many analysts would feel comfortable addressing.
Accordingly, developers must realise the importance of making AI-powered financial models extremely accessible (e.g. if you click a data point, it tells you its origin). Otherwise, there’s no way for analysts to maintain accountability.
Current AI-powered financial models are made accessible via price, democratising access. They’ve become even more accurate along the way, meaning manual financial modelling is completely obsolete, like paper-based record systems or teletype machines.
In this scenario, AI financial model developers are constantly battling to see whose financial modelling systems are faster, most accurate, etc. New, AI-literate graduates entering the financial services industry would learn to operate and question AI-generated financial models rather than poring over Excel.
Given how AI in financial services is manifesting, it seems more likely that larger companies will continue to develop their own specialised financial modelling products, complete with extra features like investment recommendations. For example, JP Morgan is releasing a tool, Moneyball, that makes investment recommendations by identifying market patterns.
As for the drive towards pouring money and resources into AI, JP Morgan is projected to spend $17 billion in 2024. Such an investment reduces fair access to sophisticated AI financial modelling tools, widening the divide between data accessibility and decision-making in financial services firms. To many, the technological imbalance might seem unfortunate or even unjust.
–
Of course, both of these scenarios are speculations about the future of financial modelling, which (in parallel with the development of AI) remains largely unknown. What is knowable is that AI developers must pair functionality with explainability.
Another key consideration for all AI financial modelling developers is that their platforms do not become ‘feature factories’. Many AI financial modelling platforms offer various ways of presenting and manipulating the data. For people looking to expedite financial modelling processes, the easiest way to do so is not to engage these platforms but to use a tool that collates and presents data for financial modelling.
We’ve created a tool that assists financial modelling by releasing key financial data (and ratios) from financial statements and annual and quarterly reports. Financial Statements AI enables analysts to extract data from financial statements. Here’s how it works:
Seeing is believing, so here is a video demonstration of Financial Statements AI in action:
Interested in learning more? Book a call with our financial data project team or email us at hello@evolution.ai.
