Some members of NeuGroup for Heads of Financial Planning and Analysis are making real progress in using algorithms, machine learning and AI to produce automated forecasts that are frequently as accurate or better than current methods in a fraction of the time. Participants at the H1 peer group meeting described processes that generate financial projections with less and less input from business managers or finance staff. - Instead of relying on human estimates, algorithmic forecasting leverages historical data, statistical models and predefined business drivers. Members advancing on this automated path are using combination of ERP data, Python, Power BI and the AI tool Claude (combined with some homegrown coding).
- By taking this approach, one member said, “We have proven that we can get between a +/- 2% [accuracy] range”—and in far less time than it takes business leaders meeting frequently to discuss changes to the forecast.
Blurred lines. This member has incorporated algorithms into all aspects of the P&L with the exception of cost of goods sold, a work in progress. It’s a cross-functional effort at this company, with resources from tech, finance and some blurring of the lines separating them.
- Also blurring is the delineation between top-down and bottoms-up forecasting in the annual planning process. “Rather than framing top versus bottom for me, it’s about the algorithmic point,” the member said.
Cultural obstacles and solutions. FP&A’s biggest obstacle to implementing algorithmic forecasting, members said, is convincing business leaders to accept that an automated system is reliable, often superior in terms of accuracy and far more efficient. “It's still relatively early days, one member said. “Part of this is myth-busting and culture. But some is just basic understanding. Like, wow, it's already here today.”
- Companies on the leading edge of this transformation are taking a gradual approach, rolling out the process in stages in different regions, with some business units continuing to do some forecasts with hundreds of human touch points for now. And members acknowledge the absolute need for human intervention and supervision of the process.
- But by FP&A presenting the automated forecasts for several quarters alongside forecasts done the traditional way, business leaders “will see the data, and they can compare, and say, ‘OK, yeah, maybe this is the right way,” a member said.
The flexibility to go granular. The power of algorithmic forecasts in part reflects a model’s ability to weigh and focus on what drivers matter most and to give far less weight to less material factors that business leaders may include in their plans. The good news, one member said, is the automated systems can also supply the granular info.
- “The costs of processing large amounts of data at scale have fallen dramatically, so you can have both granular automated projections and high-level adjustments,” one member said.
- Another member observed that there is likely a cultural payoff in “starting with the level of drivers that people are comfortable with and used to even if you don’t necessarily need all of them, ultimately, to be more accurate. I start with a level of trust to some degree.”
Other challenges, advice. For some companies not as far along the automated forecasting journey, the biggest remaining challenges include having the resources to get all the needed data in one place and searchable. Peer group leader Andy Podolsky observed, “Completeness of data is the key, not ‘cleanliness’ of data, which appears to have been the historical focus. Get all the data and organize it correctly and your algorithm will figure it out. This can take significant investment.” - Mr. Podolsky noted that members leading the way recommend that others frame success as “does this algorithm outperform what you’re currently using?” instead of ‘does this algorithm perfectly predict the future’ or ‘does the algorithm meet my aspirational performance goal?’”
- Looking ahead, members said FP&A teams will still need humans with experience for more strategic projects. And automation will mean they and business leaders can spend more time on intellectually challenging work and far less time on assembling or synthesizing huge spreadsheets of numbers to generate accurate forecasts.
