CFO Connect logo
artificial-intelligence-finance
Event Recap

How to build artificial intelligence into modern finance processes

Luc Hancock
Luc Hancock CFO Connect

Artificial intelligence is yet to really catch on for finance teams. Survey after survey tells us that CFOs are excited about and see the value in AI, but they haven’t integrated it into their processes yet. 

Tooling is part of the issue — there aren’t many out-of-the-box solutions that do for finance teams what ChatGPT does for marketers or sales teams, for example. But that’s no reason to give up. The artificial intelligence available today can already enhance and automate much of a finance team’s work. You just need to know how.

Gabriela Gutierrez led a recent CFO Connect workshop to walk us through it. She’s been using AI to create forecasts, build reports, and analyze trends for years. 

For her, there’s no going back.

Gabriela broke down her process and showed real examples of these AI upgrades in action. Here’s a recap of what was discussed, and you can also watch the replay in full: 

About the expert

Gabriela Gutierrez was CFO at Teads, a cloud-based platform that helps companies manage programmatic digital campaigns. Gabriela joined in March 2023, after previous senior finance roles at Roqad and classified ads provider Kleinanzeigen.

In 2021, Gabriela founded no-code financial forecasting tool Predictful.ai. Having left Teads, she recently founded tabs, an AI planning tool for finance. Gabriela is based between San Francisco and Hamburg. 

The need for AI and finance automation

Every good journey starts at the beginning. And Gabriela’s enthusiasm for AI comes from a problem familiar to most finance professionals: manual data work. 

“I joined a B2C company, and we were using a lot of manual Excel files. We needed to copy/paste and clean up data for hundreds of thousands of rows. Which was not scalable, and our team was understaffed. I think many finance professionals can relate to this situation.

“We really wanted to focus on the strategic part of the business. But our business requirements were increasing over time. Large amounts of data, no accurate forecasting, and not enough time to become a business partner to all of our business unit managers.

“In short, we urgently needed a solution to make our processes more efficient. So I embarked on a journey to find this solution.”

The problem here was scalability. Which is often the catalyst for companies to invest in automation and AI. 

A forecasting model based on “Prophet”

“I found an open source model through Facebook forums called Prophet. This model was designed to predict website traffic. So working in a B2C company, it was a perfect match — our revenues were tied to our number of visitors.

“We asked Prophet to predict 72 periods into the future based on historical data. Prophet gives better results where you have a high amount of historical data from many different periods. The past three or five years would be ideal.

“We created a forecast for the next 12 months, and for one product we had an accuracy of 99%. That obviously helps us to make better decisions. We could better plan our resources, because if we can plan our top line accurately, we can plan our OpEx. And that helps to increase our profitability.”

Enhancing the forecast with AI

Gabriela had a highly accurate forecasting model. But part of her role is to analyze these forecasts and feed back to the business. And AI can help here too.

“Making the forecast is more data science than AI. But we can use language models to interpret forecasts. We were creating these forecasts in the middle of the pandemic, in an unknown situation we’ve never seen before. [With machine learning] we could create multiple scenarios almost automatically. 

“We asked AI to see when trends in our data had changed. It would put red dotted lines into the charts to show changes. And we could go back to the business to try to figure out what happened. Did our servers stop working? Did something else unexpected happen? 

“AI can predict, but it doesn’t have the same experience and business context to answer why. So as finance business partners, we need to be on top of what’s happening in the company. 

“But we can also Chat GPT for a variance analysis. When we see clear variance in trends, we can ask Chat GPT for the possible causes based on historical data. It can look for similar occurrences in the past, very quickly. 

Other opportunities for AI analysis

Gabriela went into Prophet in detail, but that’s just one forecasting model. So how could other businesses - including B2B - use these tools?

“Suppose you want to use AI to look at your unit economics. You’d feed all this data - business units, COGs, and more - to the AI, and ask for both predictions and analysis. It can first give you an overview, and you can also ask it for more in-depth analysis and predictions. 

“Some models work well to analyze and predict your pipeline. Again, it depends on how much data you’re talking about, and there will always be some trial and error. 

“To ensure accuracy requires a lot of testing. Train the model using historical data where you already have the results. So let’s say you have three years’ data. Give the model only the first year, and ask it to predict years two and three. You’re hoping to see accuracy above 95%. And then test out other models the same way. You’ll take the model that’s closest to your actuals after these tests.” 

The challenges for finance professionals

Artificial intelligence is a new frontier for most businesses, and everybody’s still learning - even the engineers developing it. So if you follow Gabriela’s lead, what diificulties can you expect?

Coding skills are required

“The only problem was we needed to code. And as a finance manager that has never done programming before, it was quite a journey even to onboard myself to the world of Python. But after a while I was able to implement the first model.

“Implementation was pretty quick in the end. Prophet took about one week - there are only really five lines of code. But finding this model took some trial and error, and that’s where the human element comes in. You need to use judgment and choose the right metrics to measure. But once you know how to code and can understand the model, it's quite fast.

“There is also a huge job opportunity for people that already know SQL and especially Python to be able to use machine learning models. In the future we will really only be using Python code to make predictions and automate processes.”

Manual processes still exist

“We used AI for predictions and analysis, which was great. But we still needed to manage all those Excel files and ensure the data was accurate. We would load all our data into Google Sheets, and then we could feed it to the model using Google’s Colab notebooks.

“Today we have point solutions, but they’re not solving the workflow as a whole. So I'm not able to directly connect to the machine learning model and put it into my ecosystem. Rather, I have many different steps in order to be able to use it. It still involves a lot of manual work. 

“In fact, incorporating AI properly requires completely rethinking your processes. You can’t just drop it in.”

Expect ongoing maintenance

“We also need to maintain the models. The moment you have new data, you need to update - unless you are directly connected to a database. 

“And that brings you back to coding. In my case, nobody else in the team was able to maintain these models.”

There may be data privacy concerns

“Some companies can’t use tools like ChatGPT for compliance reasons. So how can we ensure data privacy while using AI? 

“This is actually the benefit of open source data models. You don't need to share your data when you use these. It’s the code that’s open source. And you can use this in a closed system. 

“For example, this code that I'm using in Prophet, I'm not sharing with everyone. It's only used within my Google account. Your forecasts and predictions aren’t shared externally. 

“But yes, tools like ChatGPT can be an issue for some companies. So you may not be able to do as much analysis using AI for now.” 

The future for AI in finance

Clearly these are still the early days. So where does Gabriela see these tools moving in the near future?

“For accounting, there are already companies replacing the entire accountant flow. Truewind AI, for example, can book every incoming invoice automatically. The AI knows which general ledger to assign each transaction to, and can create a P&L statement and balance statement from this data. 

“We will reach a future where data cleaning for all the models would be automatically done. That would make a huge difference for the approach I outlined.

“And I would like to be able to easily have our own database. We could access any data just by asking a question: how much was my revenue from the last three years? How are the predictions for the next three months?

“Our chatbot or co-pilot would do all the work for us. The AI would understand the context of the business and remove some variables, add new relevant variables and so on.

“Which leads to automated reporting. We wouldn't need to tell the AI what to look for, it will do it for us, with all the variables and KPIs from the data we have today. For example: costs. If we already have sales and marketing costs, we could ask for the burn rate and runway, and it would automatically create those formulas for you. That’s taking out some manual processes already.” 

Further reading