Hype or ROI? AI’s impact on finance organizations

Not a day goes by that you don’t hear about artificial intelligence or generative AI and its effects on culture, society, and business. Building a company based on AI models and applications for the last 11 years has given us a credible perspective on the changes that have taken place, the direction the market is heading, and how much of what we’re hearing is hype versus actual ROI.

This perspective is based on the real-life experiences of our customers across thousands of companies. I recently had the opportunity to share thoughts on the accelerated pace of technological change with customers and prospects at our AppZen on Tour events in Frankfurt, Germany and Dallas, Texas. The question on everyone’s minds is, “What will work look like for all of us, in a world where AI is handling tasks? How do I equip myself, my team, and my Finance organization? What does the future of business look like?”

At the start of the recent AI race, The Wall Street Journal posted an article titled, “Generative AI Is Already Changing White-Collar Work as We Know It.” The reason for such a profound change is in the numbers. From 63 different business use cases, McKinsey & Company estimated that generative AI could add between $2.6 trillion and $4.4 trillion annually to the global economy. That’s equivalent to adding a nation the size of the UK or Germany to the world economy. When they analyzed additional use cases integrating generative AI capabilities into existing software applications, the impact nearly doubled.

AI’s greatest challenge: Accuracy

So how did we get here? Large language models (LLMs) like those used to build ChatGPT are extremely powerful. They can handle complex tasks that might take a person years to master, such as creative activities like writing. When I asked to summarize a timeline from the early development of AI to today, it did a decent job, saving me hours of research and writing. It’s incredibly useful for creatively synthesizing knowledge.

Next, I used an image generator that also uses an LLM to create an image for this blog, one depicting the AI technology journey from predictions and anomaly detection to generative AI. It created an interesting, colorful picture of technological evolution that looked legitimate. Perhaps it was inspired by biblical characters and landscapes, or someone achieving a zen state or Nirvana, and all the changes that might happen in between.

While it had easily researched and summarized text and terms for me, when I looked closely, I realized the image generator had been unable to put any coherent writing into the image. It made up words and added meaningless, misspelled gibberish. These errors happen when a language model is trained to make a prediction based on context or probability but not accuracy, or lacks a fact-checking mechanism.

People make errors all the time. But our threshold for accepting AI errors is half what we would tolerate from a person. A 2022 Gartner survey found that finance leaders were willing to accept a 10% margin of error from humans, but only 5% from AI. And yet, in 2023, Gartner data revealed that 18% of accountants reported making financial errors at least daily due to increasing workload demands.

This level of error is potentially disastrous for us, our teams, and our companies. Finance work requires 100% accuracy. How can we use generative AI to help us with our day-to-day finance tasks while avoiding both AI hallucinations and human error?

For finance, the technology build matters

At AppZen, we have been creating AI models since our founding. It’s hard for new competitors jumping into the game to catch up to the deep body of domain knowledge on which our AI models rely. First, they were trained with a narrowed scope of finance expertise for enterprise spend—specifically around expenses, cards, and invoices—then on millions of receipts per day. They’re smaller than the large language models (LLMs) used by organizations like OpenAI for building ChatGPT. Our models are better at balancing precision (how often they’re correct) and recall (how much information they draw from to get the right answer) because the data they need to access is smaller and more focused.

Another very important distinction is our AI’s ability to recognize when it’s not confident in its prediction. It doesn’t make up reasonable-sounding, false information. Instead, it asks for human feedback. It learns from user responses and processing outcomes. And it becomes more accurate and efficient over time, just like a real coworker.

Our newest language models are hosted in the AppZen environment, making them more secure than solutions relying on outside LLMs. This allows us to provide enterprise-level security and 100% accurate results.

Highly accurate reinforcement learning through human feedback
Varied data across 1000s of sources and 50+ countries
Fully anonymized, verified, cleaned, and labeled data
Privacy and guaranteed IP protection
Seamless integration into your everyday workflow

Can AI automate your SOPs?

However, accurate AI models are not enough. The technology has to be relevant to your everyday workflow. The Mastermind AI platform is the only AI tool purpose-built for finance to automate tasks that are normally part of your standard operating procedures, or SOPs.

SOPs make up the gigantic, step-by-step handbook for performing your day-to-day operations. It’s the set of rules that govern “the way we’ve always done it.” They detail exactly how to do each task, who is involved, and anything else you need to know to do your job. They use workers’ knowledge, training, intellect, and available information, to guide decisions. If a new person joins the team, SOPs ensure the process remains consistent, like reminding an AP clerk that, when an invoice comes in, they must read the invoice line description and match it to the right PO line. Everything is documented in deep detail.

Normally, these tasks rely on human knowledge and decision-making. Essentially, SOPs are instructions with triggers, checks, and actions. So the question becomes: What if you could automate your SOP to meet your exact business needs?

Finance operations for the way you work

Our AI platform does just this. It absorbs your organization’s institutional knowledge, autonomously performing tasks, helping you make informed decisions, and allowing you to adapt to new scenarios. It thrives on complexity and change. It takes overwhelming, complex SOPs and simplifies them, making them easier to understand, adjust, and work with. This is finance operations for the way you work.

Incorporating generative AI across finance workflows has allowed us to create amazing new features: AppZen Inbox, which automates your AP inbox; AppZen Coach, which offers insights for expense audit policies; and Team Insights, with interactive dashboards that provide managers with valuable employee spend behavior analysis.

Generative AI is not something to be scared of. It's a tool to help you do your job more effectively. As a leader in the industry, we’re here to help you learn how to work with it so it becomes a valued member of your team, a digital coworker expanding your abilities, equipping you with information to help you make better business decisions.

I hope you’ll connect with us online or join us at an upcoming AppZen on Tour event to hear powerful customer stories and build an empowered mindset around AI and its abilities. It’s time to “change white-collared work as we know it” for the better.