AI – the end of the road for Financial Modellers?
Published by Myles Arnott on Apr 6, 2023, 8:23:49 AM
Talk of the impact of artificial intelligence on jobs, and humanity as a whole, has run to fever-pitch in the last few weeks following the release of the fourth version of Chat Generative Pre-Trained Transformer (ChatGPT).
This culminated in the release of an open letter from the Future of Life Institute calling for “all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4”.
Goldman Sachs has predicted that around 25% (300 million) of jobs in Europe and the US are at risk due to AI. Whilst The Wall Street Journal state that “Accountants are among the professionals whose careers are most exposed to the capabilities of generative artificial intelligence”.
So, does this mean it’s the end of the road for Financial Modellers?
The role of a Financial Modeller is to build a representation of a real-world financial situation. This can be a single asset or investment or a full business model.
There are five main disciplines that all Financial Modellers need to master. These are:
Let’s consider each of the five disciplines of financial modelling in turn and assess the impact of AI on our role as Financial Modellers.
Scoping & Design
The starting point for any financial model is the scoping and design phase. As Financial Modellers, we need to have the soft skills to be able to successfully interact with our project stakeholders so that we understand their requirements.
We then need to understand the required outputs that are needed from the model, what inputs are available and how reliable they are, and finally the business logic needed to transform the inputs into outputs.
This phase requires a strong mix of communication, interpretation and design skills which financial modellers hone over years of modelling.
Of all of the five disciplines of financial modelling this phase seems the least suitable for AI – It really does benefit from the human touch.
Build
The build phase is what most people think of when they think of financial modelling. This is the phase when a financial modeller creates the representation of a real-world financial situation in a model.
I group the tools available to modellers into the following:
– Spreadsheets – E.g. Microsoft Excel and Google Sheets
– Modular – E.g. OpenBox and Modano
There are a number of good practice standards and guidelines (FAST, BPM, SMART, ICAEW) that help financial modellers to structure their spreadsheet-based models in a standardised way.
There are many benefits to standardisation, including improved collaboration, reduced risk and greater build efficiency.
The modular approach to spreadsheet-based builds takes the idea of standardisation to the next level by enabling the financial modeller to simply select the modules and timeline of their model.
The tools then generate the model and all its data flows for the modeller, without the need to build everything at the cell and row levels.
Off-spreadsheet solutions require minimal build effort on the part of the financial modeller. In Brixx, for example, the user simply creates the Profit & Loss structure by selecting pre-built modules.
The balance sheet and cashflow are then automatically generated within Brixx with minimal input from the modeller.
These three levels of model development represent an evolution from spreadsheet-based, cell-level, and ground-up build-through to modular and off-spreadsheet solutions.
This area of financial modelling feels ripe for automation using AI.
And in my opinion, this is no bad thing!
Imagine a future where you can feed the outputs of your scoping and design work into AI and out pops a model, ready for you to review.
Review
It’s well documented that spreadsheet-based financial models are error-prone, so thorough review and testing is absolutely essential. The review phase is, however, an area of financial modelling that is all too often overlooked.
Reviewing a model involves assessing the following key areas:
- – Are the input assumptions valid
- – Is the calculation logic correct
- -Are the formulae correct – both the logic and accurate function usage and cell/range selection
- -Does the model conform to good practice standards
- -Does the output look reasonable
- -How sensitive is the model to changes in the input assumptions
Reviewing is much improved by the use of review tools such as OAK, riskHive, Rainbow and Numeritas’s NxT tool.
These tools can run automatic formula checking in line with defined rules and run sensitivities to identify which inputs have the biggest impact on the outputs.
The use of software already much improves the review process. I would imagine that layering AI into model review tools will make the process even better.
Scenario Analysis
The real value of a financial model is the ability to run through different scenarios and also to test the business case’s sensitivity to changes in assumptions.
Tools such as @Risk, Crystall Ball, riskHive and Visyond make this process much easier to run by managing the scenarios and routines such as Monte Carlo simulations for you.
As with model review, we already rely on software to help us with running scenarios. AI will be able to run scenarios and assess the results much better than humans can. As Financial Modellers we can then review the areas of possible interest highlighted by the software.
Communication
AI has already proven itself highly capable of finding useful (and, let’s be honest, some not-so-useful) insights in vast data sets, and creating great visuals.
Microsoft has recently announced Microsoft 365 CoPilot, which is designed to do just this within MS Excel.
But these insights and visuals will still need to be interpreted and presented to stakeholders. Just like the scoping and design phase, I think the human touch wins over AI here.
In summary
When I reflect on the five disciplines of financial modelling, where do I think AI will be applied?
It seems to me that AI will be most successfully applied to the build, review, scenario and insights disciplines, whilst scope, design and communication will be seen more and more as the core areas where Financial Modellers add value.
How can Financial Modellers stay ahead of the curve?
With this in mind, how do we, as Financial Modellers, make sure that we stay relevant in this ever-changing world?
The World Economic Forum believes that AI will lead to long-term job growth and that reskilling and upskilling will be key. As Financial Modellers, we need to embrace that innovation and the changes that AI will bring.
Those that learn and adapt will continue to thrive.
Comments by C Human – Goalfix Financial Modellers – May 23
Thank you Miles for an informative and enlightening review of AI in the context of financial modelling.
In my own rudimentary examination of ChatGPT as a model building tool, ChatGPT referred to itself a number of times as follows
As an AI language model, I don’t have personal opinions nor beliefs. However, here are some steps that can be taken to identify input assumptions in a financial model:
As an AI language model, I do not have access to the specific context or details of which input model drivers you are referring to. However, I can provide a general outline on how one can allocate appropriate values to input model drivers:
Whilst fully realising that it is early days, I believe that until AI (ChatGPT or other AI platforms) can achieve the ability to think laterally, creatively and critically examine all the elements of a business case, it will not pose a threat to human interventions. Whilst the mechanisation of certain aspects of model building would be clearly advantageous, the true value of a model is dependent on the veracity of the substance underpinning the calculations.