FP&A Educator Says AI Is Changing The Skills Finance Teams Need

Before founding my online course platform, Inside FP&A, I spent over a decade at Procter & Gamble and Unilever, leading FP&A and accounting teams across global markets. And before that, I joined fast-growing tech companies, including helping take Squarespace public and preparing the company’s first financial story for Wall Street analysts.

How AI Should Be Used In FP&A

Christian Shares

You no longer need a data science team or expensive software to build an algorithmic forecast. ChatGPT, Claude, and Gemini are fantastic at building statistical forecasts.

AI should be used for your first draft of analysis and model building.

"First draft" is the operative term there. People tell me they thought AI would take over 95% of their modeling tasks, but building strong financial models requires more time reviewing the output, which eliminates much of the time savings.

It's also good for forecasting based on time-series analysis. You no longer need a data science team or expensive software to build an algorithmic forecast. ChatGPT, Claude, and Gemini are fantastic at building statistical forecasts. All you have to do is feed them historical data and ask them to create Python code that runs an ARIMA model in Excel.

And here are other ways I use AI in FP&A every day:

• Variance analysis: Build driver trees that explain variance.
• Business partner prep: Walk into meetings prepared to gather the right information from partners.
• Budget reviews: Identifying risks and opportunities in a business partner's budget without damaging the relationship.
• Report polishing: Telling the story behind the numbers
• Board presentations: Anticipating tough questions before they're asked

For prompting examples on each of these use cases, head here.

Some tasks, however, must remain human:

• Deep analysis
• Business partnering
• Influencing
• Weighing trade-offs

Why Variance In AI Forecasting Is Dictated By The Inputs

A few years ago, when I led an FP&A team at a large consumer goods company, I maintained the working capital forecast. It was a beast of a forecast, with dozens of sub-forecasts for line items and divisions that laddered up to the full company number. Because it was so complex and time-consuming, we chose it to test a machine learning approach for the first time.

We outsourced the AI model's development to a third party. When we received the new model's first results and compared them a month later to actuals, we were excited. The forecast accuracy was significantly better than our previous results. I was sure this project would be a resounding success until the leadership team wanted an explanation for the remaining variance between the machine learning model's predictions and actual results.

I had no idea how to answer that question because I wasn’t involved in building the model. I contacted the data scientist responsible for the model at the third party. I approached it the same way I would analyze a variance from a traditional Excel forecast model: I asked him to explain how the model works so I could identify what might cause the variance. He said, "We use an ensemble of different machine learning algorithms. They compete against each other every month, and the best-performing model forecasts. Specifically, we use machine learning algorithms like DeepAR from Amazon, Temporal Fusion Transformer from Google, and traditional algorithms like Holt-Winters and Autoregressive Integrated Moving Average models. In total, the ensemble includes 12 different algorithms that compete against each other. We backtest each, and then choose the one with the highest accuracy.”

I didn't understand more than half of what he said, so I followed up with, “OK, why do you think such a significant variance still exists between actual results and the model's predictions?" He answered that he had no idea because he didn't understand our business.

The situation was that I didn’t understand how the machine learning model worked, and the data scientist didn’t understand how the business worked. How could we explain a variance here? I brainstormed different solutions to this problem. Should we invite the data scientist to our office so we can explain how the business works? Should we ask someone on our team to take a course on machine learning and data science? Was it even possible to learn this without getting a degree in math or computer science?

It took a while, but I finally realized I was looking at the problem all wrong. In fact, I was trying to solve the wrong problem entirely. With machine learning, you don’t explain variances by pointing to flaws in the model itself. The most advanced algorithms are called “black box models” for a reason — even the scientists who built them sometimes don’t understand the conclusions they make. I realized it’s all about the inputs. Garbage in, garbage out.

To explain variances with machine learning, we must investigate whether the data points we feed the model are right and sufficient. The best way to do this is to combine experimentation with an approach called “backtesting”. With AI models, you can pretend the last year hasn’t happened yet and ask the model to create a forecast. For example, if it’s May 2026, I can ask the AI model to forecast 2025 by feeding it historical data from 2020 to 2024. Then, you can immediately compare its closeness to actual results, adjust the data points, and test it again.

In the end, we realized forecast accuracy significantly improved when we fed the model additional data from our inventory systems, data points we had never considered before because doing so would have made our traditional spreadsheet-based models even more complex.

With machine learning, you don’t explain variances by pointing to flaws in the model itself. The most advanced algorithms are called “black box models” for a reason — even the scientists who built them sometimes don’t understand the conclusions they make.

Share This Quote on:

• Share on Twitter
• Share on LinkedIn
• Share on Facebook

Christian WattigOpens new window

Founder of Inside FP&A

Why CFOs Must Standardize Processes Before Implementing AI

AI doesn't fix broken finance processes. It scales them.

I've watched enough finance teams adopt new tech to know how this goes. The tooling amplifies whatever's already in place, good or bad.

If your variance commentary is a mess today, e.g., every analyst writes it differently, there are no agreed thresholds, there is no consistent structure — AI will produce that same mess across every cost center, fluent and fast. If your model has no documentation or naming standards, AI will help you break it more confidently. If your forecast swings 20% from one month to the next with no clear driver narrative, AI will keep it swinging, just with better formatting.

The finance teams I see getting actual value from this stuff aren't the ones with the fanciest setup. They're the ones who already knew what good looked like. They had templates. They had a working definition of "useful." They'd done the unglamorous work of standardizing before they ever asked AI to do anything.

Why Finance Teams Must Use Greg Brockman's Prompting Method

Many finance teams struggle to get good output from their AI tools because their prompts are too short or not specific enough. They treat AI like it's a Google search.

I always suggest Greg Brockman's method for prompting reasoning models:

1. Goal
2. Output Format
3. Warnings
4. Context

With this approach, teams invariably get better results.

How CFOs Should Redesign Hiring For The AI Era

Christian Shares

Watch for overconfidence, tools-first thinking, weak communication, and the absence of trade-offs. These three exercises will help you evaluate the attributes I mentioned above.

CFOs need to redesign their job-applicant screening processes. The AI era requires different skills.

Traditionally, leaders focus on spreadsheet speed, Excel proficiency, and report-writing quality. But AI is automating all of these tasks. Instead, look for the following:

• Judgment under uncertainty
• Business communication
• Systems thinking
• Learning velocity
• Data fluency

When you're interviewing candidates for any role, try the following:

• Give candidates a short, plausible, AI-generated variance explanation or performance summary. Ask them what's questionable and what needs to be verified. Then, ask them to rewrite it into a few bullets for an executive.
• Ask the candidate to give you the first five questions they would ask if they needed to explain why revenue forecasting was off — and why.
• Tell the candidate that you expect margin pressure and demand uncertainty next quarter. Ask the candidate to define three scenarios, name the key drivers, and recommend what to do now.

Watch for overconfidence, tools-first thinking, weak communication, and the absence of trade-offs. These three exercises will help you evaluate the attributes I mentioned above.

Why Strategic Skills And Deep Analysis Are What Matter Now

Finance teams need to know what they are optimizing for.

Once they free up all that time with AI, senior leaders will expect them to add value differently — by becoming strategic partners and providing deep analysis. They'll be expected to connect the dots between operational metrics and financial results to make concrete recommendations on how the business should capture opportunities or mitigate risks.

In other words, they need to upskill in the higher-level strategic finance skills that AI can't address. That's the only way to earn a seat at the table in an AI-enabled future.

Follow Along

Follow Christian Wattig's work on LinkedIn. And learn from him on his YouTube or check out his online course platform, Inside FP&A.

More expert interviews to come on The CFO Club!