The Chatbot That Couldn't Count

The Ledger and the Machine — Part One of a Four-Part Summer Series

For the past few weeks I have been writing about soccer and the lessons learned from the World Cup. Sports have always provided me with ways to connect with people and to share life lessons. They offer us these micro moments under stressful conditions to help us bring awareness on how we can improve leadership and teamwork. We saw moments of brilliance with leaders who maximized their team’s potential and moments of disappointment from others who abandoned their strengths in the biggest moments. Most importantly, the World Cup was a great reminder to us of the importance and joy of connecting with one another and relishing each other’s uniqueness.

Thankfully, we still have many memories to create over these next weeks of summer. I thought it provides a good time to “chat” with you about where we stand with AI in accounting, how we got here and where do we go from here.

So, as we take in those rays over the next four weeks, let’s add in a little added AI color.

This is Part One. It starts with a chatbot that could not count.

―

In November of 2022, a research lab in San Francisco released a website that allowed you to type a question into a box, and something on the other end typed back an answer. It could write a poem about your dog. It could draft a wedding toast. It could explain the infield fly rule, more or less. It was called ChatGPT, and within two months, a hundred million people were using it. Usage exploded! It was the fastest adoption of any technology ever!

I remember the first time I used it and was blown away by it. It left me with the feeling that I was connecting with someone “human” on the other end. At first, I used it more as a google enhancement and a letter generator.

I remember the first work question I asked ChatGPT, “what was the most beneficial way to value an asset for estate planning”. It came back to me quickly. Confidently, in a well-organized paragraph. I copied the answer and sent it to one of our tax experts. Unfortunately, it was wrong. Not wildly wrong. But, that’s the issue. The fact that the answer wasn’t wildly wrong, potentially deludes you from thinking you have what you need because it was wrong in a way that sounded exactly right. Thankfully, we have experts who could spot the error immediately. However, the people who may not know the nuances involved to determine the accuracy of the response may wrongly act on it. For our profession whose entire product is earning trust from delivering accuracy, this was disqualifying. You cannot bill someone for a confident guess.

But, November 2022, for many of us, was a glimpse into the future of a profession that will be disrupted. It’s like the moment Steve Jobs visited Xerox’s Palo Alto research lab in 1979 when he was shown a computer with windows you can open and a pointer you could slide across the screen. Granted the technology wasn’t where it needed to be to execute well, but it was a glimpse into the future. Jobs saw the future. It was obvious to him that this was the way every computer would someday work. ChatGPT was accounting’s Xerox PARC moment.

AI in November 2022 was a fluent talker that although it may not reliably do accounting “yet”, it could read, reason, and respond. It could take in a messy paragraph of human language, understand the intent behind it, and produce something responsive. It was the beginning of how the work of a knowledge profession is changing. Ours included.

ChatGPT as a standalone was never going to cut it in a profession like ours and that was precisely the opportunity. A general-purpose chatbot with no access to the client’s books, no memory of the client’s history, no audit trail, and no way to verify its own claims is not an end-all product for accountants, but it provided the framework for what is to come in what it lacked: the verification, the workflow integration, the domain depth, the connection to real financial data, the trail a reviewer could follow. All of this needed to be built with purpose. In the distance, you could see an entire ecosystem was needed that did not yet exist. Founders and those who looked around the corner began moving, and so did the money.

Venture capital, the engine behind the innovation in America saw the opportunity. The money that funded the PC, the internet, and the smartphone turned some of its attention to, of all things, accounting. The VC’s saw that the new technology can finally handle these laborious, often inefficient tasks. Coupled with the trends in the market of a profession hemorrhaging talent, with hundreds of thousands of accountants having left during COVID and the pipeline behind them collapsing, the opportunity for automation and tools was exploding.

Andreessen Horowitz aka “A16Z” is one of the most influential venture firms in the world. They have been ahead of most others and published a piece called “Death, Taxes, and AI” that mapped the opportunity. They have been a major catalyst in creating the new ecosystem. They have been funding a new generation of companies to build the things the chatbot could not do on its own.

A company called Rillet raised money to do something almost nobody had dared attempt in thirty years, rebuild the accounting ledger itself, from scratch, with AI in its bones. A company called Basis built AI agents designed for accounting firms, on the principle that every action a machine takes must leave a trail a human can verify. Companies began going after receivables, tax preparation, audit, close automation. Within two years of the chatbot demo, there were dozens of serious, well-funded companies building the ecosystem the demo had implied, and many of the world’s leading tech investors were competing to fund them.

The investors were also clear-eyed about the limits, which is exactly what made their conviction credible and personally provided me with sources to listen to and learn from. In January 2025, two of them at Andreessen Horowitz, Seema Amble and Marc Andrusko (now with Oak) described the state of things in one sentence:

“While AI pilot programs are prevalent industrywide and firms are enthusiastic to try them, fully deployed vendor relationships are still relatively scarce. This is probably because AI models are not yet as proficient with numbers as they are with text, and because accountants (unsurprisingly) are fairly risk averse.”

That statement “not as proficient with numbers as with text” was key. The machine could talk, but it could not yet count reliably enough for people whose signatures carry legal weight. But notice the word the investors used: “yet”.

The biggest software company on earth was reading the same map. Microsoft, which had invested billions in the lab that built ChatGPT, took the technology and bolted it into the tools the whole world already used, calling it Copilot. Emails summarized in Outlook. Slides drafted in PowerPoint. And, in theory, help inside Excel, please hold on to that phrase, ‘in theory,’ because what Copilot could and could not do inside a spreadsheet turns out to be one of the hinges this entire story swings on, and it is where Part Three begins. I’ll be getting to this, soon. 😊

We have been very fortunate at Wiss to have colleagues that saw the potential of the opportunity and helped create WissLabs. We knew that we were so close to the work and how the hours of an accountant’s week were consumed by reading, extracting, summarizing, and re-keying. We knew that our world both qualitatively and quantitatively would increase tremendously if a machine could replace even half of these tasks. We needed to build our own innovation arm to invest and partner with the companies building the missing ecosystem. It had been the missing piece we have long desired in accounting, to rebuild a more human way in accounting.

The chatbot was not the final product, but an inspiration for what can be and proof that machines could now read, reason, and respond. The real products that impact accounting will be built down in the plumbing of how businesses run, where the operating history lies, some of that remain in a filing cabinet, most in a system of record. The ledger will become intelligent.

That is where the story begins. Next week, Part Two: the thirty-year reign of the most boring software in the world, the day Salesforce cut off its own head, and the builders who decided the unglamorous plumbing of business was the most valuable real estate in technology.

The Ledger and the Machine continues next week with. Part Two: The Filing Cabinet and the Brain.