Subscribe
Share
Share
Embed
Business success is dependent on a solid financial foundation & success looks different to everyone & there is a lack of equity of access to resources and information for small business owners and independent contractors & there is a societal narrative making us believe “balance” is our ultimate goal & … There are so many “&”s that impact being your own boss. Let’s have some frank discussions on the basics of business with a holistic focus on everything that helps business owners define and find success.
Molly Beyer: [00:00:08] Welcome to The Ambiguous &: Business Basics and Beyond, the podcast, where we have frank discussions on the basics of business, with a holistic focus on everything that helps business owners define and find success. Each episode is a reminder that success isn't one thing, it's a whole lot of ambiguous ands. Like, subscribe, or follow, and let's explore these ambiguous ands.
[00:00:33] Hello and welcome to the Ambiguous &: Business Basics and Beyond. I'm your host, Molly Beyer, and I'm here to lead you through frank and holistic conversations on the basics of business. Let's just put this out there right away to kick off the conversation, AI is going to change bookkeeping. The question is whether we use that change to eliminate human work, or to eliminate unnecessary work, and make the human work more valuable. The research right now, in the fall of 2026 gives us considerably more substance to this conversation than that usual "AI will replace Accountants" headline. The provocative question here isn't whether AI can categorize a transaction. It demonstrably can. Quickbooks is using AI to do things like suggest transaction matches and categorize with high confidence indicators. It is surfacing anomalies in financial statements. It's maintaining the books continuously and providing AI assisted forecasting. Zero describes its AI as handling reconciliation, document capture, invoicing, and cash flow insights, and Sage is developing predictive cash flow, AP automation, and continuous close capabilities. There's even some academic work that specifically is talking about transaction categorization in QuickBooks using some relational deep learning.
Molly Beyer: [00:01:49] The researchers are reporting that their model outperformed the production model they compared it against, while also addressing the cold start problem of making useful predictions with limited historical data. So we won't spend too much time arguing that AI can't do bookkeeping. Instead, let's ask if AI can do more bookkeeping, what exactly are we paying a bookkeeper for? Now, bookkeeping involves many different kinds of work, and I think we should explicitly separate them into a few categories. Those are data processing, pattern recognition, professional judgment, and then there's the human relationship and context. So data processing includes things like downloading transactions, extracting receipts, categorizing transactions, matching transactions, reconciling, entering bills, sending invoices, and repetitive reporting. Pattern recognition is things like identifying unusual transactions, spotting anomalies, noticing cash flow trends, predicting future cash needs, identifying changes in margins, and comparing current behavior against historical patterns. Professional judgment encompasses answering questions like: is this really a business expense? Is this transaction being classified appropriately? Does the financial statement actually represent what happened? Is the business owner making a decision based on bad information? Does a strange number represent an error, or something that actually happened? And then human relationship and context goes deeper into determining: why did the owner make that decision? What is happening in the business that the numbers don't explain? What does the owner actually need to know? What should they be worried about? What are they afraid to look at? What does this information mean for this particular business? AI is moving rapidly from the first category into the second, and the evidence is beginning to show that this can change the third and fourth categories rather than eliminate them.
Molly Beyer: [00:03:50] A 2026 Journal of Accounting Research field study is particularly useful here. Researchers studied 277 professional accountants and more than 200,000 transaction level records from 79 SMEs. They found gen AI was associated with productivity gains, more granular ledgers, faster month end closes, and a shift away from routine data entry toward communication and quality assurance. But they also found an important warning. Accountants were more likely to intervene when AI confidence was low, and relying on AI recommendations that weren't supported by consensus could increase risk error risk, which all gives us a much more nuanced premise. The evidence isn't showing us that humans disappear. It's showing us that the location of human effort changes. So this is a great place to talk about a real world example and Bench Accounting software is a good one to highlight. Bench was supposed to be a major industry disruptor by using significant AI bookkeeping with human review, and the disruption got more intense when they suddenly shut down and left business owners unable to access their data at the end of the year. But Bench did not ultimately disappear, it just paused operations from December 2024, when it was acquired by employer.com, and it resumed operations in January 2025.
Molly Beyer: [00:05:19] Bench says it continues to use an in-house bookkeeping team and describes multiple bookkeeper reviews and year end quality assurance, and bench represents an important model: its technology enables bookkeeping, plus human bookkeeping rather than AI replaces bookkeeping. So with examples like bench and the movement into more AI tools in the other major accounting softwares, the bigger question really becomes what happens when technology gets good enough that the ratio between technology and human labor changes dramatically. And that's an economic question I really want us to explore. So the early tech enabled model was the software collects the data, technology organizes it, and then humans do much of the bookkeeping review or much of the bookkeeping, and then humans review it. So the emerging model is more increasingly that the software is collecting the data, AI is interpreting and processing it, humans are investigating the exceptions, and then humans provide context to judgment and accountability. And the next stage really could be AI maintains the financial system continuously while humans become responsible for the integrity, interpretation, and application of the information. And that is a very different profession. One of my regular messages is as a business grows, the business owner's bandwidth shrinks, and AI potentially changes that equation, not because the business owner should suddenly become their own bookkeeper, though, it's quite the opposite.
Molly Beyer: [00:06:51] If AI can take care of more mechanical accounting work, the value of a good financial professional may shift toward helping the owner better understand and use the information. And that fits beautifully with my constant push for this data driven and heart centered work, because we really have the data plus the human side, and that gives us an interesting distinction. The old value proposition for bookkeeping really was "I keep your books". The emerging value proposition then becomes "I make sure your financial information is accurate, meaningful, and useful". And those aren't the same thing. And if we take this even further, the future bookkeeper may be less of a bookkeeper, not because bookkeeping disappears, but because the definition of bookkeeping expands. So predictive AI is where this gets especially interesting, because we're moving from what happened to what is happening, to what is likely to happen, to eventually, what should we consider doing about it? So, for example, QuickBooks Enterprise Suite currently offers AI assisted cash flow forecasts using historical financial data to generate a 13 week forecast. Well, Xero and Sage are developing similar predictive and analytical capabilities. That means bookkeeping could move from a more fundamentally historical discipline toward a forward looking discipline. And that is an important caveat, because prediction isn't understanding. An algorithm might recognize that businesses with this pattern of receivables historically experience a cash cash shortage in approximately six weeks. But the human may know that's because this owner intentionally slows collections during hunting season, because 60% of their customers are in the same industry.
Molly Beyer: [00:08:36] So the model then is seeing a pattern, but it's the professional that understands the story behind the pattern. And that's something that we've regularly focused on in our firm, that bookkeeping is an inherently historical. And then what we do is we use those numbers to help create that future profitability or profitability. And this is where many bookkeeping professionals are really starting to move, that the industry is moving into a more advising and consulting approach to truly help business owners understand their numbers and how to use them to realize the success that they've defined. And it also gives us a really good AI limitation discussion. So we've got that the AICPA CMA's current guidance is really remarkably aligned, sort of with this idea as well. Their 2026 ethics discussion specifically addresses predictive forecasting and AI assistants, while also emphasizing data quality, security, verification, transparency, and professional competence. So their basic message is also that AI should support rather than supplant professional responsibility. And the current professional literature is increasingly talking about this human AI collaboration rather than simple automation. So the AICPA CMA's September 2026 discussion specifically frames the opportunity around redesigning work so that humans and AI contribute where each is strongest. And my own ideas on AI are very similar, and they're really that it isn't good and it isn't bad.
Molly Beyer: [00:10:15] It's just a tool that exists. So for me, my general guidelines are: don't use AI simply because AI exists. Really use it where it makes sense. Don't ask it to replace expertise that I actually need. Don't hand it sensitive information without understanding what happens to that information. Don't confuse a confident answer with a correct answer. And don't outsource responsibility just because I outsourced a task. So I really use AI as a tool when it's appropriate to use. It kind of also, my way that I use AI really kind of leads into the environmental conversation surrounding AI because it's an extremely important trade off that comes with this tool. And again, it's one of the biggest reasons that I do want to limit my usage. So looking into the energy impacts, the International Energy Agency estimates that data centers consumed about 415 TWh of electricity globally in 2024, which was around about 1.5% of global electricity consumption. Its 2025 base case projected data center consumption to more than double to about 945 TWh by 2030. And the 2026 update adds an important wrinkle to this, that energy use per individual AI task is falling extremely rapidly, but people are simultaneously adopting much more energy intensive AI uses, including more of these reasoning and agentic workloads and data center electricity consumption rose 17% in 2025, while AI focused data centers grew even faster.
Molly Beyer: [00:12:00] So again, efficiency is improving and demand is exploding. And both of these can be true, and that is an important nuance. Then when we add the water piece to the conversation, it really becomes even more nuanced. And much like with the electric electricity impact, this is another place where the conversation also becomes too simplistic. A 2025 peer reviewed study from Lawrence Berkeley National Laboratory found that workload level water consumption can vary more than 10,000 fold depending on server efficiency, electricity source utilization, cooling technology, climate, and other factors. The authors explicitly conclude that there is no single recipe for minimizing water use. So a question like how much water does AI use isn't exactly a very useful question by itself. Better questions are where is the data center? What cooling system does it use? What is the local water situation? How electricity is powering it. How efficient is the hardware? How heavily utilized is it? What happens to the heat? How much water is actually consumed versus withdrawn? And what alternatives are available. And that really becomes a much more intellectually honest conversation. And then we move to how these technologies are being regulated, and what is the result of this. Because good regulation can create the conditions for innovation by forcing problems into the open and giving innovators something concrete to solve.
Molly Beyer: [00:13:40] The EU has already established reporting requirements for data centers covering energy performance and water footprint, and its sustainability rating work is intended to improve transparency and encourage efficiency, renewable energy, grid efficiency, and waste heat reuse. And the IEA points to a range of possibilities. So we have things like renewables, storage, nuclear, geothermal, grid optimization and more efficient hardware models. It also estimates that existing AI applications could potentially produce energy savings in other sectors that exceed the emissions associated with the data centers, but explicitly warns that AI isn't a climate silver bullet. So for me, this really comes down to a few things. We shouldn't pretend the environmental cost doesn't exist, but we also shouldn't pretend that the only responsible response is to stop innovating. We should demand better innovation, and that's where regulation can matter. If you say something like build whatever data center you want, you have less incentive to solve water use. If you say, if you're going to build this facility, demonstrate energy efficiency, disclose water consumption, account for local water stress, and meet defined performance standards, then you've just created a market for people who can solve those problems. The EU is already moving in that direction, including work toward minimum data center performance standards. And there's another important environmental distinction here because the question simply shouldn't be can AI do this? It should be: is this a good use of AI? If an AI system uses substantial computing resources to generate a cute picture of a bookkeeping spreadsheet holding a cup of coffee, that's a very different societal calculation than using AI to detect financial fraud or identify accounting errors, or forecast a cash crisis early, or optimize electrical grids, or improve building efficiency, or discover new materials, or assist with scientific research, or reduce waste, or automate repetitive administrative work for a tiny business.
Molly Beyer: [00:15:55] So the environmental conversation therefore really belongs inside of this responsible use conversation rather than being a separate anti AI argument. So like I shared earlier, I already have developed a philosophy around AI that I think should become sort of the backbone of this conversation. AI really should expand human capacity, not eliminate human responsibility. And this is a good framework for asking whether a use of AI is appropriate. So some questions to ask yourself in relation to using AI is does it actually solve a problem or are we using AI because it's fashionable? Another question is what happens if it's wrong? A wrong social media caption isn't the same as a wrong financial classification. What about what information does it require? This is especially important for financial payroll and personally identifiable information. You need to be very careful what you are entering into with these systems. Also, where does human judgment belong? Because not everything that can be automated should be automated. And finally, is the resource cost justified by the value created? This is really where that environmental impact belongs.
Molly Beyer: [00:17:10] I think that the future of bookkeeping is very interesting with the introduction of AI. I'm not one who believes that AI will take over the industry and that there won't be bookkeepers anymore. I think there are maybe four models that bookkeeping is sort of going to emerge into. We have the traditional model where humans are doing most of the processing and review with AI automating some of those pieces of the process, which is overall what's happening in the industry right now. Then we have more of an AI assisted model here that's emerging where humans are reviewing, correcting, and interpreting where well, AI does most of the routine processing. There's also this AI native model where the humans are owning the judgment controls context and advisory, while AI is maintaining much of the financial workflow. And then there's this fourth possibility that I think is going to be sort of this more likely eventual model. And that's the AI native plus human financial partner. And this is a system where the AI can keep the books continuously, but the professional doesn't disappear. Instead, that person becomes the person who is asking the questions like, do these numbers make sense? What changed? Why? What does this mean? What should the owner know? What decision is sitting underneath this number, and what isn't the system seeing? And that's a very different kind of bookkeeping practice. So I think the blanket statement of AI is changing.
Molly Beyer: [00:18:39] Everything is true, but it's also not catastrophic in regard to the bookkeeping industry in my opinion. Bookkeeping is one of the professions where AI can take over an enormous amount of the mechanical work to make the work more efficient, hopefully more accurate and likely more cost effective. And that makes us confront a deeper question than if the machine can produce the numbers, what is the human actually responsible for? And again, this brings us back into these conversations on trust, judgment, context, security, environmental cost, business relationships, and ultimately what we think the purpose of bookkeeping actually is. And that is a much more interesting conversation then AI will replace bookkeepers. So yes, the question is nuanced and the future isn't clear. But AI does not look to be going anywhere. So it becomes all about responsible use and maintaining the proper controls to keep the system on track. Thanks so much for hanging out with me today as I talk about AI. I'd love to hear your feedback on today's episode, as well as any requests for future content. Drop a comment or suggestion and join us next time for more frank and holistic conversations on the basics of business. Please also like, subscribe, or follow so you never miss an episode. And until next time, I am Molly Beyer and this has been The Ambiguous &: Business Basics and Beyond. Have a wonderful day.