跳到正文
原文
Accounting Today(美国会计行业媒体)· Ahmad Tamimi·· 4 天前站内评分42

Ahmad Tamimi:会计师事务所应将AI作为“员工”管理

AI isn't a tool. It's a hire.

导读

Ahmad Tamimi主张会计师事务所将AI按“员工”而非工具管理,并以客户文件接收实验说明其执行效率与监督边界。在一家税务事务所的实验中,管理员和申报人员处理同一批真实客户文件用时刚超过84分钟,AI在32秒内完成分类归档,并在1分47秒内完成清单核对、客户邮件草拟和数字摘要。停止计时时工作量约减少98%,但管理员仍需复核分类,申报人员仍决定缺失资料并审批邮件,决策权保留在员工手中。

正文

Most accounting partners I talk to have tried a chatbot or two. Some have found useful applications. Others have concluded that it is one more thing to learn and gone back to work. 

Processing Content

I think part of the problem is how we have been asked to think about AI. 

AI has largely been presented as a tool: something a person opens, prompts and operates. That can be useful for a specific task. But when someone has to initiate every step, supply the context and move the result into the next system, much of the work still belongs to them. 

For recurring work, a more useful frame is AI as a hire. 

When you hire someone, you define their responsibilities, train them on your processes and establish what they can handle independently. You also explain when they need to ask for help. Over time, you judge their performance by the quality of the work and the supervision it requires. 

Accounting firms should bring that same discipline to AI. 

A test case: client intake

Consider client document intake. In many practices it is the seasonal bottleneck, and the usual answer is to hire temporary staff to receive documents, chase what is missing and get files ready for a preparer. That cost repeats every year. 

Identifying a document is one task. Managing intake involves more: organizing what arrives, recognizing what is missing, following up with the client and knowing when an ambiguity requires someone at the firm to make a decision. The useful question is how much of that process AI can reliably carry, and how much checking it creates for staff.

To find out, I ran a head-to-head experiment with a tax firm that handles both individual and corporate filings. We used real client documents, as messy as they arrive in practice: multiyear engagements, mixed corporate and individual returns, files with inconsistent names sent through different channels. We timed the firm's staff doing intake the way they actually do it, then gave the same files to AI set up as a team member rather than a tool: given the job, not prompted step by step. 

On the files we tested, the firm's staff took just over 84 minutes of combined administrator and preparer time to complete intake. Downloading and renaming one client's documents alone averaged 10 minutes and 37 seconds. The firm's most complex multiyear engagements can run to four and a half hours. 

The AI classified and filed the same documents in 32 seconds. By 1 minute and 47 seconds, it had also compared them with the checklist, drafted a client email and prepared a summary of the figures it found. That was the point at which we stopped the timer: before review, roughly 98% less time than the staff took. The administrator then checked the classifications, and the preparer decided which missing items to ask the client about and approved the email. 

The people did not disappear. They moved. The administrator's job shifted from renaming files to checking them, and over time to guiding and training the AI on the firm's own patterns. Decisions stayed with staff. Execution did not. 

Supervise it like a hire

The analogy holds here too. A new employee does not sign the return; the firm does. So the firm decides what the AI handles on its own, what it escalates and what a person approves before it leaves the building. In our case that meant every client email was approved by the preparer and every missing-item call was made by a person. Those rules were set before the first file went through, the same way you would onboard anyone new. 

Treating AI as a hire does not mean trusting it blindly. It means giving it a defined job, clear boundaries and a manager who reviews the work, and judging it on results. 

For a busy practice, the measure of success is whether the work gets done accurately, with less effort from the people responsible for it. That means measuring the whole process, including review, corrections and exceptions. The hiring analogy is useful because it changes what we evaluate: whether AI can carry a defined responsibility under appropriate supervision.

Ahmad Tamimi

Co-Founder & CEO

Ahmad Tamimi is co-founder and CEO of Pennix, an AI document intake platform for accounting and tax firms. He has spent two decades in software... Read full bio

来源:Accounting Today(美国会计行业媒体) · accountingtoday.com

审计热点 是聚合摘要和阅读索引,原文版权归各来源所有,引用请以官方原文为准。如果你是来源方,希望更正、下架或调整展示方式,可以通过反馈页联系我们。

来自微信公众号的内容,本站仅呈现摘要与原文链接,不转载文章全文;文章权利归属原作者及原公众号

关于使用条款隐私信源