AI can now do most of the boring, mundane, repetitive tasks such as data entry, transactional coding, and AP/AR. This raises the question of what exactly a human accountant does. According to several leading software vendors featured at the Institute of Management Accounting’s technology showcase, the answer is clearly to revisit this work to see if the AI is making mistakes.
Processing details
This was a point emphasized many times throughout the event, which demonstrated a range of solutions for corporate accounting professionals. This is largely because many of the routine processes in this field are highly rules and formula-based, allowing for consistent automation on a task-by-task basis.
Dana Alhasawi, senior manager of midmarket and commercial solutions at payments solutions provider Ramp, addressed this point in her presentation, saying the finance function is the area that will benefit most from AI more than any other part of the business. Financial processes are built on learnable, enforceable patterns based on data-rich financial workflows generated during repetitive daily operational tasks such as coding transactions, reconciling invoices with colleagues, tracking receipts, and closing the books each month. He said this is particularly important because the stakes are high for the finance department.
Muhammad – Stock.adobe.com
“We know that finance is a zero-error culture. We can’t rely on AI agents to make decisions in the dark. That’s why everything we build, every AI-powered workflow, has a full audit trail, fully explaining what the system did and why. There will be full visibility and human control at every meaningful decision point. AI will be applied selectively. You will decide where the AI will operate autonomously and where it will escalate. Compliance requirements governing the function will not go away.
Specifically in the finance space, she said the accounts payable process is one of the most concrete examples of how finance and AI work together. Laura Van Lenten, principal consultant for midmarket solutions at Ramp, demonstrated the solution’s ability to read invoices with optical character recognition, capture things like invoice number and date, automatically encode authorizations, and optimize payments according to users’ specifications. This can be done not only for invoices that the customer currently has, but also for new invoices that the customer has never seen before. The overall goal is to create as contactless an experience as possible, she said.
“Our overall goal with everything at Ramp, whether it’s the AP side or the card side, is that the whole experience should be essentially touchless for the user. The typical entry process for teams today should actually be more like a review process, and it takes away that manual work and obviously saves teams hours of input work,” she said.
Once most of the day-to-day tasks are completed, the user will primarily be looking at output and approving payments. And here again, AI is already flagging invoices that indicate a risk of fraud, and also providing recommendations on whether something should be approved based on user activity and data. Nowadays, it’s common for people to do automated reviews, but Ramp can automate even parts of the reviews themselves, so even this process can be done with little scrutiny.
“And as individuals became more and more comfortable with these recommendations, we started to see teams come in and say, yes, this is good. We don’t have to dive in anymore and scrutinize line by line. And because it provided that level of detail, teams gained confidence in the solution very quickly and were able to spend hours going from approved time to payment really quickly,” she said.
Finally, a similar message emerged from Asaf Gover, CEO and co-founder of financial workflow automation solution Apprentice. He began by pointing out that while all businesses have similar needs, such as document processes, the details vary too much for a truly universal solution. As a result, automating finance functions can actually be very complex, he said. Even if there is an AI that can replace humans, people still need to explain the details of their individual businesses, including processes, data structures, controls, etc.
However, just because it’s difficult to automate doesn’t mean it’s impossible. The governor said a different approach is just needed. Instead of following a set of rules and procedures, programs learn by doing. When users start a job, they record themselves doing their job and provide a step-by-step audio explanation of what they’re doing. The system then synthesizes that description and builds its own automation to match it from there. You can then fine-tune your automations and provide feedback to the system if anything goes wrong. He said that when a process in the system goes down, it is run by a deterministic system that delivers consistent results every time, rather than a probabilistic AI.
As such, although Apprentice does not rely on specific defined rules from the start, it is still intended to automate many routine tasks. Again, once this is established, the human role is primarily to review the output and monitor the system.
“The process can be very long or very short, but the user can see it step by step. The user has the peace of mind that the system is actually capturing its full complexity and can trust it,” he said. “And the users who review the automations are the same users who run the processes, who know the details and who know if the automations were done correctly.”
Noel English, senior solutions consultant at payments and finance platform Bill, similarly pointed out that with so many processes automated, humans will primarily become reviewers. Documents, whether physically scanned or digitally accessed, are processed instantly and automatically checked for duplicates, incomplete information, etc. The system also combines AI and OCR to automatically capture header-level details and populate information such as vendor name, invoice number, due date, total amount, and payment terms. We do something similar with the purchase order workflow, where the system automatically retrieves the purchase order and fills in the details based on the user’s actions. Once that’s all done, humans mainly need to look at the information and see if everything is okay.
“As you can see, almost 90%, probably 95% of the work is already done here. For me, it’s kind of a review process at this point,” he said.
Mark Fisher, senior vice president of marketing at Vic.ai, an AP automation solutions provider, echoed Alhasawi’s point. AP is in a unique position in AI automation because it is a very low-risk, high-reward field where there is already a lot of repetitive manual work that AI is good at, which is why so many companies are focusing on it right now.
“So we think AI for AP is clearly one of the best entry points in accounting,” he said.
Steph Hartnett, solutions engineer at Vic.ai, said the program quickly predicts header data commonly found on invoices, such as vendor and total amount, as well as less common information such as department and location on purchase order backing invoices, with 97% accuracy. He said the program also looks at a user’s email to find any invoices or communications from vendors that need attention. Like other solutions, many of them are automated and primarily reviewed by humans.
But Vic.AI also does a lot to automate even the review process. Invoices marked green are highly reliable and can be configured to bypass AP review completely and automatically send invoices to the approval stage (a final decision must be made by a human). Ultimately, humans won’t need to review everything that AI generates, only those that require their attention first. This is not as much as people think.
“So imagine your AP team is working on a subset of invoices that they actually receive. 50% are processed autonomously by the AI, and the remaining 50% ends up in a queue. Only 20% require editing the AI prediction fields. The team can easily navigate to that review queue directly from the dashboard,” she said.
And even if the system is not AI-native, it may have powerful automation capabilities that still require human review. Jonathan Grimes, a technical accounting consultant at financial operations platform FinQuery, said his platform automatically scans all the information from an invoice that someone needs from an organizational perspective.
“That’s why this system has been built once again to ensure this process is easy and streamlined for your company. Not to mention, if you’re talking about advances and invoices, we also have all the additional details to get more context about your portfolio,” he said.
The webcast was held on March 12th. This was the first technology showcase hosted by IMA.
