What can artificial intelligence do to ease the workloads of finance staff and the CFO? What is possible now, and what applications will we see down the road? Will finance chiefs have to discard old habits and adjust their thinking to enable AI tools to permeate organizations?
Getting clear answers to those questions can be difficult. Last week, we met with Sarah Spoja, CFO of accounts payable and payments company Tipalti, and asked her how AI will reduce workloads, automate manual tasks, and even improve accuracy in the finance function. Below are some of the highlights of the conversation.
Sarah Spoja
Tipalti
First CFO Position: 2018
Notable Previous Companies:
- KKR Capstone
- Bain
This interview has been edited for brevity and clarity.
VINCENT RYAN: How do you think about artificial intelligence and its potential applications inside the finance department?
SARAH SPOJA: At small businesses and many mid-market organizations, CFOs lack the resources to build custom AI applications. I talked to the woman who runs AI for the office of the CFO at Microsoft, which has a 1,000-person finance organization. They have a whole team of people that are just thinking about AI. But I don't have a systems engineer or an engineering team for finance.
"Can I find all the data and provide one view of cash flow quickly, work that might take an analyst three or four hours to dig into? That capability is on the development roadmap of every FP&A tool out there."
So, inside the finance organization, we’re looking at our tech stack and how vendors are innovating to bring AI technologies forward. They’re using [AI technology] in one of two broad categories: one, to automate manual tasks, do the matching and automation that previously couldn't be done without these large language models [LLMs]. Or two, which is further down the line — I haven't seen many good applications of it — how does AI change the insights that we're able to pull from our data and financials? How does AI actually make us smarter about the data?
What might be an example of a finance application in that second category?
SPOJA: I'm really interested to see the development of it in something like cash-flow forecasts. To get a good cash-flow forecast, finance has to pull the data from many disparate sources, like existing bank accounts, receivables, payables, and payroll. FP&A tools can give you a cash forecasting view that is really additive. But I think AI has a chance of really turning data into insights in a very ‘easy-to-do way.’ … Here's the budget, and here are the actuals; where was the miss? Can I find all the data and provide one view of cash flow quickly, work that might take an analyst three or four hours to dig into? That capability is on the development roadmap of every FP&A tool out there.
How will AI improve automation and make workflows even smoother?
SPOJA: Where finance will see early returns is in manual tasks, having an AI perform standard workflows quicker and with fewer errors. At Tipalti, our AP automation platform pulls all the data off the invoice so an employee doesn't have to hard-code an invoice into the system to make it payable. When you add [LLMs] and AI on top of something like OCR [optical character recognition], we get much higher success rates on matching invoices [to purchase orders] and having those invoices auto-coded through the platform.
"I’m telling my team that we have to go experiment. Not everything’s going to work. But if we see an interesting use case or implementation of AI, let’s go try it."
An AI can disaggregate large tasks and workloads, automating what’s standard and flagging what’s non-standard — in the second instance, an invoice that a human needs to look at. Over time, the percentage of standard items increases. The best applications in the office of the CFO automate a manual task but do it with more smarts around how much can be considered standard. … How will this play out in finance, when precision and accuracy are considered table stakes? Over time, CFOs will trust the AI more when they sample the standard [items], and the system gets them right 100% of the time.
Can midsize and smaller organizations sit back and wait for others to bring AI applications to them?
SPOJA: We’re in this mode with new technologies where if you don’t experiment, you’ll never get any results. Experimenting is kind of costly, not in money in the short term, but in time and energy and other intangibles. So hearing public companies talking about the ROI from AI experiments in fourth-quarter earnings calls was great. I’m telling my team that we have to go experiment. Not everything’s going to work. But if we see an interesting use case or implementation of AI, let’s go try it.
As an example of a real-life use case, we’re starting to draft our technical accounting memos using a [large language model] that’s been trained on IFRS and GAAP. Now they’re not going to be perfect. But it saves us from having to go through all the code, and it tells us what to include in the memo. … I’ll also do the opposite — put the info back in and say, ‘summarize this for me’ or ‘let me know what’s missing.’ Using an LLM can save us a couple of days per memo.
What barriers lay ahead for finance departments trying to adopt LLMs like ChatGPT or other AI tools?
SPOJA: The team that runs our AI projects just sent around a cheat sheet on all the best prompts, educating our entire company on how to improve at AI. There are a lot of use cases, and massive technology changes are at the heart of it. But how do people respond to those technology changes? And how does the organization build the skills and capabilities? How do you get people over the initial hump of, ‘I don't know what I'm doing here.’
A lot of it is getting people involved. … And CFOs can't be concerned about having previously always looked at every single journal entry. A few years from now, we'll be looking back, and all this will be in the rear-view mirror. And we'll feel as comfortable with AI [technology and tools] as we are with Excel.