Thought Leadership

New technology has long been at the center of major change in the world. However, it often takes time and a whole new way of thinking to fully utilize that technology. Take for example, the harnessing of electricity when the world was still powered mostly by steam. While Thomas Edison created the first power station in 1882 (said to power up to 5,000 lights) and many other inventions made electricity accessible to the masses, only 5% of factories were using this new technology near the turn of the century.

One reason many manufacturers weren’t using electricity at the start of the 20th century was because it didn’t fit the way they had their factory set up. Factory owners thought they could use electricity the same way they had used steam to power their factories: one central engine and lots of belts and shafts to power large machines. What they failed to realize was that electricity was not only less expensive, it could be more easily distributed to individual work stations, rather than being powered in one central location.

It took factories 30 years to realize that they needed to redesign the way they worked with this new technology. They needed to adopt both a new floor plan and a new mindset. By 1910, Henry Ford fully utilized the power of electricity to create a steady flow of power for his conveyor belts and large machines in his Highland Park Plant. Eventually Ford engineered a way for cars to be built more efficiently. His assembly line for the Model T went from taking 12 hours to about 93 minutes to assemble, all using electricity.

Applying Industrial Age Lessons in AI Adoption for Audit

What does any of this have to do with the accounting industry? In a recent Suralink webinar called The Factory Floor Fallacy, Dimitar Stanimiroff, Suralink's general manager for EMEA posited that this history lesson can inform the way that accounting firms utilize agentic AI. Rather than expecting this technology to just do the same work as humans do, a shift in the way firms use the technology needs to be made.

"The most dangerous moment in any transformation isn't when the technology is bad," Stanimiroff said. "It's when the technology is just about good enough to feel like you're making progress while the systems around it remain completely unchanged."

Most firms use a collection of tools to gather, store, sort, and review files from clients, including email, spreadsheets, and software. These different places and several people involved in the process can often cause errors in data or things to slip through the cracks. In fact, survey data from Suralink’s Inside the Client Experience Report shows that 89% of firms report consistent challenges with client data and more than half of client requests are not fulfilled completely the first time. Want more proof that the current model of running engagements needs attention? The same report showed that more than 70% of engagements miss time or budget. And, 82% of clients say information requests are unclear or time-consuming.

Adding AI to the mix doesn’t always make things more efficient, much like manufacturers from the early 1900s trying to use electricity in the same way steam was used. Oftentimes bolting on AI to an existing process can make things more complex.

Ensuring AI agents are used to make the engagement run smoothly, according to Stanimiroff, means getting AI involved at the right stage of the engagement with accurate information. Firms know all too well the pattern of receiving inaccurate data from the client, reviewing it, sending feedback and asking the client for accurate data. That cycle can repeat throughout each step of an audit engagement. Think of it this way: if AI has incorrect data, it will still take a human’s judgment to find and correct it. If AI could find the incorrect information from a PBC and make an automatic request for correct information before a reviewer at the firm sees it, that would prevent much of the rework that happens with most engagements.

How to incorporate AI in Audit

If AI can be present throughout the engagement, it naturally seems the first right place for it to be is at the beginning, when the client receives a request and data is uploaded into the system. Most engagements still run like a steam-era factory, with everything flowing to one central point (the reviewer), where problems are detected. Placing validation at the point of upload is the equivalent of installing a motor at each workstation. If AI can be present throughout the engagement, the logical place to start is the beginning, when the client receives a request and uploads files. A redesigned flow runs through five steps: an intelligent request, a real-time validation of materials, all-clean data in, AI-powered testing, and a confident delivery. This approach is an integrated system from the first request to the final report, not a piecemeal solution that swaps one tool for another.

Incorporating agentic AI technology without a working knowledge of where and how to use it is like swapping out a steam engine for an electric one. AI is not going away any time soon. The choice firms have now is whether to redesign the way they run engagements or just plug in the new motor of AI. Redesigning the way firms use this technology means getting the right data the first time and the compounding efficiencies throughout the rest of the engagement.

The idea of redesigning a firm's entire engagement workflow seems daunting, so where do firms start? Stanimiroff suggests running an experiment. Choose one engagement that is already frustrating the team and run that engagement through an intake process built to catch the problems at the start. Change nothing else. Select the intake process because an incomplete or inaccurate document at the start of an engagement sets up a rework loop. Track hours saved, requests fulfilled correctly the first time and client response time, then let the numbers make the case for further changes.

Industrial Age factory owners needed 30 years to learn that their floor plan mattered more than the motor they swapped out. Today's accounting firms don’t have 30 years to get up to speed with AI tools. Successful firms have the most accurate data to begin with and that sets them up to make huge efficiency improvements.