January 22, 2024
min read

Building a Better Challenger Model: Leveraging AI/ML for Headcount Forecasting

Can a finance challenger model powered by artificial intelligence/machine learning allow finance teams grappling with headcount forecasting work smarter, faster, and most strategically? 

That’s the crux of the conversation between Precanto CEO Sandeep Madduri and The FP&A Guy, Paul Barnhurst, on a recent LinkedIn Live conversation, “How AI/ML can help transform headcount forecasting for FP&A.”

Madduri says that running an AI/ML headcount forecasting tool run alongside current in-house headcount models has the power to demonstrate where current models fail, and where they can be improved. This allows for deeper assessments of headcount planning, and more accurate forecasting.  

Their live conversation broadcast Nov. 13, but you can view the video and read highlights of the transcript, edited for clarity and length, below. 

Watch the Video

From the transcript

Paul Barnhurst: We've all been hearing a ton about AI over the last few years, and there's a few areas that I believe it can help all of us improve forecast accuracy.

We've seen AI be used a lot in fraud detection and anomaly detection, and there's other areas, even in, you know, forecasts, detecting what may go wrong there. We've heard, “I need a lot of data for AI to work.” Not only do you need a lot of data, but you also need clean data, right?

If your data is garbage, if you don't have good data governance, if you don't have master data, it can really create some challenges.

Two [points] summarized really well what CFOs have said about the promise of data. We're seeing two things. We're seeing that AI/ML improves forecast accuracy.

No ERP has cleaned data these days.This is the allocation policies in the real world. Accounting inputs are garbage. So we in FP&A always end up cleaning before reporting. Can you relate to that, Sandeep? 

Sandeep Madduri: Absolutely. You know what we have seen is data sources, such as ERP, HRIS, and even a recruiting/applicant tracking system. So there's data coming from different systems that may or may not be clean. So a prerequisite is to be able to leverage some kind of automation to get a clean data set. We’re hearing 

PB: Talk a little bit about how clean the data needs to be for us to start using AI/ML. How much data do we need? To get started with an AI or an ML approach towards headcount forecasting or predictions, you would need at least two to three years, ideally, of historical data, right?

SM: An insights platform or a predictions platform cannot be built in this day and age without an inbuilt mechanism to clean data. So our platform looks for things such as missing values in your data set and imputes them. If there's anomalies, it detects them and bubbles them up, such as duplicates in your records and your transactional records.

Then we run models on top of your data to produce a prediction and recommendations in terms of what you should be doing.

Before you even get to implementing any kind of AI tool, where do you start with cleaning your data?

SM: You want to look at frequencies at which your systems are getting updated, which ones are your messy data sources, and start working with those first.

PB: When it comes to AI/ML, there's a lot of areas you can forecast. A great area to start from is headcount forecasting. Why do you think that's such a good use case for AI/ML?

SM: Headcount is somewhere between 60% and 85% of operating expenses at most companies. Think about direct headcount costs, base pay, bonus, payroll taxes, benefits, stock based compensation, and indirect costs – facilities, sales compensation, training, and any kind of travel and entertainment and contractor expenses. They're all indirect costs tied to head count.

PB: In my FP&A career, headcount was always a challenge, especially the bigger the company – are people in the right cost center, the right department, do we have the right hiring plan? I've often struggled with headcount. I think anything that can be done to make that easier is definitely good for the business.

So what does the process look like for a company that deploys AI/ML for headcount forecasting?

SM: Essentially, step one is coming up with the clean set of data. After you come up with a clean set of data, how do you auto forecast your head count expense to begin with? Think of this as a daily close for everything headcount, in terms of expenses. 

Step two is engaging or collaborating with business partners. I've seen this in several of my previous roles where I'd get a lot of asks from my budget owners: “I have a director role” or “I have some contractor expenses,” “I want to pull forward certain things or push out certain things.” So the collaborative aspect between FP&A and budget owners is understanding the headcount expenses and making decisions sooner in the quarter.

The last step is predictions and insights powered by AI and ML – What would a challenger model say in terms of where my headcount units and headcount spend dollars would land for the quarter? 

Are there areas where there's a big difference in your forecasting models?  

The delta between those two versions will explain a lot and help you accelerate some of your decision making in the quarter.

PB: So that challenger model is basically saying, “Hey, let me use technology to give you another view to help ensure that you're being as accurate as possible.”

SM: Absolutely. It's another point of view. 

When you're fixing data you also need to make sure the policies are such a way that the data will stay clean, because if you don't, over time, the data becomes messy even with good policies.

PB: This is an ongoing evolutionary thing. 

SM: Yes, it’s an ongoing effort. One of the reasons why we are excited about bringing AI and machine learning to this thing is it's not just an ongoing effort, but there's a continuous learning curve. Your data could change, your imputed values for certain things could change, the way you look at any kind of missing values or duplicates or anomalies could change, as well as your business model.

And that's what makes it exciting to bring machine learning into the mix. 

PB: If you have good tools, it will save time.

Learn more about the future of headcount forecasting with AI/ML – book a Precanto demo today.

Anne Miller
Head of Marketing
Company news
Precanto Company

Transform your headcount forecasting with Precanto

Schedule a demo to learn how Precanto can help your organization.