Data is everywhere, and it’s critical to the work of treasury and FP&A. Some of the ways treasury and FP&A professionals are trying to manage data more efficiently include automation, generative AI and machine learning.

At an AFP Member Meet-Up, treasury and FP&A professionals shared how they’re using AI and machine learning in their work and what they’ve learned so far.

An FP&A team makes use of generative AI and LLMs

The FP&A team at StreamlineClose needed to efficiently search and query their diverse data, so they implemented a Retrieval-Augmented Generation (RAG) tool. When searching a PDF, Word or Excel file, traditional keyword searches can be limited. Because RAG searches by meaning rather than keyword, you get much more accurate results.

For example, if you're looking for information on “PTO,” but a document contains the word “vacation,” RAG understands that “PTO” and “vacation” mean the same thing and provides you with the information you need.

Another aspect of RAG the team is exploring is the translation of plain English questions into SQL code. “There are so many times when you might find the need to generate answers for ad hoc questions based on your data, and there might not be existing reports or prebuilt and existing SQL codes already out there,” said Matt Osborn, Co-Founder, StreamlineClose.

Previously, finance or accounting would need to generate code for a new report. “What we're looking into is having users ask questions in English,” said Osborn, “and because the LLM is trained on the database structure, it's able to translate those questions into SQL code. Then you could execute that SQL code and return the results back to the user.”

For example, you could ask: What are the sales in California as a percentage of national sales by employee for the last six months? RAG would understand and execute the query. This is fast for finance, and faster still for the business partners who can get answers to their questions without having to ask finance!

A couple of important things the team has learned along the way: “One of the advantages of large language models is that they don't need to have structured data; however, having structured data always leads to improved accuracy,” said Osborn. “And another thing we've learned is that the manual review is still necessary.”

An unexpected application is the use of the RAG model for audits. “If our audits are asking for PBC lists or audit request documents, trial balance, general ledger activity, invoices, contracts, those sorts of things, there might be a solution where the RAG model could pass through those items and pull the relevant data, compile it, and have it ready for review and upload,” said Osborn.

Treasury teams integrate machine learning

The treasury team at Konica Minolta Business Solutions played a crucial role in a machine learning project, which they undertook to improve cash management, specifically cash forecasting. The project took about eight months and was implemented in phases.

In phase one, they automated the inflow and preparation of data. In phase two, they incorporated open AR and open AP into the cash forecasting module, leveraging AI algorithms to produce even more accurate forecasts.

“It significantly improved our forecast accuracy, leading to faster and better decision-making, and enhanced the automation process for cash management,” said Glisson F. Inguito, Director of Treasury at Konica Minolta Business Solutions.

Before this, their daily cash positioning, including gathering balances and transactions from their major banks, took two to three hours each morning — that was time ticking by before funding decisions could be made. “Now, with data being fed automatically, it takes just half an hour at most,” said Inguito.

On the forecasting side, the treasury team can now generate forecasts on demand using various scenario analyses. “We created specific cashflow categories, breaking down revenue from collections and disbursements to better project and analyze trends,” said Inguito. “This allowed us to use AI to precisely forecast these cash flows.”

Similarly, Akriti Oswal, CTP, Treasury Manager, Nestlé, implemented an AI-based tool at a previous company that helped improve their cash forecasting. One lesson she learned from the process — as the project owner and intermediary between IT and the technology vendor — is the importance of ensuring that IT and the vendor are speaking the same language and understand each other.

Oswal also advised spending a lot of time asking your vendor about the rules they are using. “Whenever a new customer gets added, a new vendor gets added, the bank narrative may change, and it might not capture some rules in the backend or the algorithm that the forecasting tool is using,” said Oswal. “It’s a constant refine and tune.”

An FP&A team’s use case for automation

At Uline, the FP&A team is currently using a time series model to forecast where their sales are going to end each day. They are manually feeding historical and recent data into the time series model in order to teach it over time the seasonality of their sales revenues.

“A lot of what we're learning is that this process requires a trial-and-error period,” said Lisa Bartko, Senior Finance Manager at Uline. “It is not as precise as we'd like, so we're still working through that trial-and-error phase and learning along the way.”

Uline has its own data team within finance, the Finance Process Improvement team, that serves as a liaison between finance and IT. Most on the team have worked in finance and learned additional programming tools and languages, so they understand the day-to-day of finance and the vision of what is possible by harnessing the latest tools. As a result, they are a valuable resource in this process. “They've made a world of difference within our organization,” said Bartko.

One piece of advice that Bartko shared was the importance of clear communication and expectations to ensure that everyone — finance, IT, senior management and other stakeholders — is on the same page regarding the timeline, deliverables and challenges. “Getting that buy-in from management before we started, but then also being proactive with regular updates, was really important,” said Bartko.

Read More About AI in Finance

• 5 Insights on AI-Powered FP&A
• Generative AI in Finance: 2 Big Questions, Answered
• 4 Questions About Embracing AI in Finance, Answered
• Ways FP&A Leaders Manage the Costs of AI in Finance
• FP&A Leaders Discuss the Role of AI in Finance Today
• Four Rules for Your AI Journey
• The Role of AI in Forecasting and Where It Falls Short