Perspectives from the Profession: AI Adoption Starts With How Your Firm Works

IPA - Perspectives From the Profession

 

 

 

By Joanie MannJoanie Mann

AI is moving into firms faster than most have decided what they actually want it to accomplish. Teams are testing copilots, generative assistants, workflow agents and specialized tools, often comparing features before agreeing on the problem they are trying to solve.

The more important question is: What kind of firm are we designing?

Every major technology transition tempts us to make the new system imitate the old one. We put desktop software where paper ledgers used to be. We move familiar applications onto remote servers. We adopt web applications but keep downloading data into spreadsheets so established processes can continue.

Those steps create value, but they also carry old constraints forward. The risk now is that firms will bolt AI onto workflows built for human searching, copying, checking, routing and remembering, then call the result transformation.

We Have Seen This Pattern Before

Over the years, I have watched business information progress from documents to records, reports, dashboards, intelligence and ultimately action. Each stage changed what firms could do with information. But the greatest value appeared when the operating model changed with the technology.

Early accounting software reduced arithmetic and retrieval effort, yet many systems were still electronic filing cabinets. Hosted applications separated software from a particular machine, but firms still had to rethink access, support, responsibility and security. Cloud applications made new capabilities easy to acquire, but that convenience also produced app sprawl, shadow IT, forgotten integrations and data scattered across systems.

AI can amplify every one of these patterns.

A firm can use AI to draft the same memo, populate the same workpaper and move the same information through the same approvals. That may save time, but it doesn’t necessarily improve the workflow. Worse, departments may independently add chatbots, transcription services, research assistants or automation tools. Each looks useful and inexpensive on its own. Together, they can create duplicated knowledge stores, unclear data flows and unmanaged access.

The Lessons We Are Overlooking

Start with the problem, not the product

Technology conversations too often begin with a demonstration. Start instead with the work. Where is information entered twice? Which reports arrive too late? Which approvals stall? What depends on email, spreadsheets or one employee’s memory? Which decisions are made without complete context?

A useful AI evaluation asks more than whether a model can perform a task. It examines whether the firm can improve the end-to-end workflow, including inputs, handoffs, exceptions, review and accountability.

Data readiness is operational readiness

AI does not create clarity on its own. When clients, jobs, revenue, costs or status are represented differently across systems, AI can repeat those inconsistencies at scale while making the results sound persuasive.

Can the firm trace its key numbers? Are important processes documented and consistently followed? Do permissions reflect current responsibilities? Is there a reliable version of the truth? If not, the most valuable AI project may be fixing the foundation.

Trust must be designed

People did not adopt previous technologies simply because the software worked in a demonstration. They adopted them when responsibilities were clear, exceptions were handled, support was available and someone remained accountable when the system was wrong.

AI needs an even stronger trust structure: visible sources, defined review thresholds, audit trails, escalation paths and clear limits on autonomous action. Keeping a human in the loop is not enough if the reviewer lacks the time, evidence or expertise to challenge the output.

Protecting Judgement and Accountability

Another risk deserves particular attention: automating away the apprenticeship through which professional judgment develops. If junior staff no longer perform foundational analysis, review exceptions or explain their reasoning, firms may gain capacity today while weakening their future expertise. That experience is essential to developing sound judgment, and firms need to consider how they’ll preserve it.

Firms may also keep AI and manual processes running in parallel indefinitely “for safety.” Comparison is appropriate during a pilot, and some independent controls will remain necessary. But permanent duplication of entire workflows increases costs, obscures ownership and creates uncertainty about which process is authoritative.

A More Practical Way Forward

Before selecting an AI tool, define the desired business outcome and map the existing workflow, including workarounds and decisions that depend on human judgment. Establish which information is authoritative and who owns it. Decide what the system may draft, recommend, retrieve or execute, and what still requires human review.

Pilot with real work, including incomplete records, unusual clients and exceptions. Measure the entire workflow: cycle time, quality, correction effort, risk and downstream effects. Then decide whether to scale, redesign or stop the initiative, retiring redundant steps where appropriate.

From Intelligence to Action

AI changes the economics of producing language, analyzing patterns and coordinating routine work. It does not eliminate the need for professional judgment. It increases the value of people who can frame the right question, recognize missing context, challenge a convincing answer, explain uncertainty and accept responsibility for a decision.

Firms have an opportunity to reconsider how knowledge is assembled, how exceptions are surfaced, how decisions are supported and how accountability is maintained. AI makes new approaches possible, but realizing their value requires examining the work itself.

Before investing in another tool, firm leaders should ask: If we were designing this work from scratch today, would we build the same process?

About the Author

Joanie Mann is CTO of Mendelson Consulting and director of cloud services at Noobeh. With more than three decades of technology consulting experience, she has worked extensively with accounting professionals to adopt new technologies, modernize client services and turn financial and operational information into better business decisions. Today, she helps accounting firms and small and midsize businesses connect cloud platforms, data, automation and AI to practical business outcomes.

Categories

Recent Posts

Subscribe

Sign up for the IPA INSIDER: a bi-weekly news round up sent directly to your inbox.

Related Stories