Microsoft Finance Suite’s AI/ML Journey: Part 1 – AN FP&A Podcast

On the latest FP&A Today podcast, sponsored by Datarails, the financial planning and analysis platform for Excel users, host Paul Barnhurst interviews Microsoft data scientist Daniel Sousa-Lenox.

In this, Part 1 of the podcast overview, Sousa-Lenox discusses how to evangelize finance teams into adopting AI and how Microsoft is transforming its forecasting and FP&A through new machine learning and AI technologies.

Barnhurst: What are some of the things you guys are doing with AI and machine learning?

Daniel Sousa-Lennox: When I joined Microsoft, which was right before Chat GPT came out, generative AI came out with the first answer but without including FP&A answers.

Since then, we’ve been building this standardized framework and we’re leaning into putting machine learning in the hands of finance professionals. We’re doing so without them having to go through a data science machine learning course and without them having to know the coding Python R or any other software language.

We want financial professionals to take advantage of machine learning models to get highly accurate forecasts at a fraction of the time that they used to be. At Microsoft, my team built this tool and we constantly keep adding to it, adding new models, and adding new features to make it more accurate or faster. We also go through the process of evangelizing the tool throughout the organization, especially throughout the finance organization, which is our main target, and then also with our customers.

Sometimes the team and our customers want to hear about what we’re doing in finance to incorporate those tools. Our answer is that we basically build the solutions. Then, we teach people how to use it, how to trust or hopefully get them to trust it both internally within the company and externally to some of our big accounts.

Barnhurst: Can you give us more detail – maybe an example of how your department has helped Microsoft make a better decision within that process? Our listeners would like to hear how you’ve supported them, improved a forecast, or helped them get insight, especially how you’re using machine learning and what you’re doing to improve the overall quality.

Daniel Sousa-Lennox: At Microsoft, we have a few teams that have completely adopted the use of machine learning and specifically our tool called the Microsoft Finance Time Series Forecasting Framework (FINN), an automated forecasting framework for producing financial forecasts.

Our team has completely adopted this tool and has replaced its current process and what that has meant for them. One specific team example is their forecasting process, which they update on a monthly basis.

It took them basically about a week of coming up with an initial number, going to leadership, getting feedback, changing the number – generally issues like that.

At first, the team experienced trouble keeping track of accuracy, which is something that we’ve noticed across the board. With the manual process, the team hasn’t been tracking accuracy consistently through each step of the forecast updating period.

Barnhurst: What was the specific issue in that scenario?

Daniel Sousa-Lennox: The issue with that is the scenario cannot improve if you don’t measure for accuracy.

If you have no idea how accurate your data is, that’s because you don’t get to go back three months later to check what happened and see if our predictions were accurate. It’s hard to know if what you’re doing is actually helping at the time, even though it feels like it does.

But in the end, how do we know we’re making the right decisions based on the numbers that we collect, if we’re not going to go back and check?

In that regard, our machine-learning tool has helped.

For example, one of our teams was trying to forecast revenues with 99% accuracy. Now, I cannot say that ML is magic and it’ll lead to 99% in every single forecast based on how a subscription service behaves, but that’s what our team found.

All in all, machine learning is very well suited for what FP&A is trying to achieve – that’s a big part of it.

The other part is with FINN, we’re seeing the great work it accomplished testing several different models and testing combinations of several different models.

That has led to them getting ridiculously great accuracy metrics. Plus, the time savings on our forecasts are amazing – it takes on average about 90 minutes to get the right forecast using our tool.

That’s significantly less than a week, and it’s also an hour and a half where you don’t have to be actively doing anything.

In Part 2 of the FP&A Today podcast, Barnhurst and Sousa-Lenox examine why AI is not a “magic bullet” for finance, despite its success, along with two of the most powerful lessons in presenting AI to finance teams.

For this and all of the FP&A podcasts, visit the website FP&A Today podcast at Podcast Bean and at the Apple Podcast website.

Recent Posts

Leave a Reply

Your email address will not be published. Required fields are marked *