May 12, 2021
Startup Finance

Adding AI to Financial Modeling

Financial modeling is a very interesting discipline, because even if you are an expert or you have the perfect template, you still need to do some guessing. Isn't there a better alternative? Honestly I think that the disciplines where there isn't a "correct" answer or where the "correct" answer isn't known until you try it (basically in the future) are the most fascinating and hardest ones. Jobs like these are researchers, entrepreneurs or financial modelers.
Filippo Burattini

Where is the guessing?

A financial model is basically comprised of two parts: assumptions and projections. Assumptions are all the data that your enter (basically the inputs) while the projections are all the data that the financial model computes on its own and shows to you (the outputs).
The assumptions are where the guessing occurs. When you create a financial model you have to input the data specific to your business, like average revenue per customer, this is data that you know in advance, but sometimes you have to input data that you may only estimate, like growth rate, customer acquisition cost, etc.

How it is handled now

First of all, the assumptions are usually kept all together and are easily editable, so you can quickly change them if you want to try different assumptions. Then the assumptions are usually taken from historical data, sector data or from experience: you know that “usually” the churn rate is around 10% so you estimate that for your startup it will also be around 10%.
This is the best way of doing things mostly because there isn’t really a better way, historical + sector + experience usually gives the best result.

Growing entropy

As you can imagine, every time we add a new input to the financial model that is an assumption, we increase the “error range” of the financial model, basically how much the projections will differ from the reality.
It is easy to understand with an example: if you only have one assumption, you only need to get that right, but with ten assumptions, you need to get all of them right, making it exponentially harder and generating exponentially bigger errors.

bad financial model

The problem

Unfortunately there isn’t (for now) a better way of doing things, this is the uncertainty of starting a business, regardless of how well you plan for it there is no real way of knowing in advance if it will be profitable or not.
Unfortunately many founders have a financial model (you have one right?) for their business, but they are too optimistic (very dangerous) or too pessimistic, this is because coming up with good assumptions before launching your startup is very hard.
This is where an AI could help.

Giving a new tool to financial modelers and founders

Artificial intelligence is being applied everywhere and it is helping a lot of processes already. It has already been applied to the world of finance, but never before to financial modeling. An AI could help make better predictions or better assumptions, which will result in better projections.
The AI that we envisioned is used to check the assumptions that you insert and tell you if your assumptions are too optimistic, slightly optimistic, realistic, slightly pessimistic or too pessimistic. This would prove really useful for everyone who is trying to create a new financial model. Of course it wouldn’t solve every possible problem, but it would be a very useful tool never before seen.
This AI could be trained with the public available information of existing businesses and with the data of every financial model that it checks, so it can constantly evolve.

Our Solution

We thought we could tackle the problem and so we did. We successfully created the AI and it is running on our platform, Sturppy. On Sturppy you can create a financial model for your business super easily and quickly and get an estimate of how realistic your projections are, helping founders eliminate uncertainty in their financial models.
We also believe that this tool is super useful for investors too, in fact, an investor can quickly check if the assumption that have been made on a financial model sent to him are optimistic or pessimistic. This could be an amazing tool for investors that need to check many different businesses or that want to really dive in a specific startup.

financial model realism score

The state of the art

As of now the AI (which we called "Paolo" ) has been trained to give estimates only for SaaS companies since they are the most common and the training data for them is best available. We would like to include our other business models (mobile app, e-commerce) and we are working on it.
The estimates are in beta and you also receive a confidence score associated to them, telling you how confident we are of our estimate (es. You financial model is “Slightly Optimistic” with a confidence of 86%).

The “Realism Score”

We want you to understand that the estimate is just a way for you to know if you are being too optimistic or pessimistic in your assumptions, IT IS NOT a metric that you should optimize.
For instance if you insert into Sturppy the assumptions for the financial model of Facebook, it will tell you “Very Optimistic” this does not mean that your financial model is wrong, but that it is optimistic and will probably be harder to execute, but definitely not impossible.

In Conclusion:

If you already have a financial model made with Sturppy, just go to the Projections tab and you will see the estimate. If you don’t have a financial model yet or have made one outside of Sturppy, make a free account, create your financial model and see the realism of your projections to let you know if you are being realistic or too optimistic.

If you have any question please, feel free to contact us, we are always happy to talk to like-minded people. Also, if you are a startup founder or are interested in running a startup in the future you might like what we have built here at Sturppy, a financial modeling tool for startups that is fast and easy.

Good luck on your journey.

If you like this article, subscribe to the newsletter!


This article has been featured on the page on Medium.

article featured on medium