I read every article I can find on AI productivity gains, looking for measurable results. I suspect many of you are frustrated by “knowing” that AI is improving your personal productivity, yet unsure how to measure its effectiveness. While I continue to write this column without AI assistance, for client work, I frequently start with my legacy knowledge on a topic and ask AI to prepare summaries of key information. The quality of the output is sufficient to address client questions, and I frequently learn something new from the analysis. I have chosen to disclose my AI usage to clients so they can have confidence in my direct expertise and can see the AI-generated responses.
We continue to have concerns about “AI poisoning” of results, where marketers have gamed the systems for AEO (Answer Engine Optimization = the practice of structuring and optimizing content so AI systems like ChatGPT, Google AI Overview, Gemini, Claude, and Perplexity can extract, trust, and cite it as authoritative answers). I appreciated Claude Anthropic’s recent announcement that it will incorporate invisible watermarks to comply with the EU AI Act, which will help identify AI-generated answers. AI engines will be able to spot what other AI engines are producing. We expect utilities to be released that can surface these invisible watermarks, which would be helpful to management and schools. Recent statistics suggest that 60%+ of TikTok videos are AI-generated, a far higher share than in the recent past. AI-generated email is already at production scale, with over 50% of email now being AI-generated by agents.
As a warning and reminder, the days of one-off prompts (chats) are coming to a close. Agents using MCP (Model Context Protocol), controlled by Orchestration platforms (n8n, Agent 365, and now Anthropic, OpenAI, Amazon, Salesforce, and more), are taking AI productivity even further by organizing and delivering agents.
With recent releases from all major AI providers, capabilities, result accuracy, and the surfacing of viable solutions improve with every passing month. We are consuming our token budgets on large-scale projects that were not practical in the past. If we can conceive it, we can build it. Developing practical new ideas is becoming more difficult. Additionally, not every piece of information is accurate, even when carefully sourced. Current information, as well as information that should be kept private, is prime currency for frontier models like Claude, ChatGPT, Gemini, and more.
As AI technology is released in products, my co-host, Brian Tankersley, and I discuss the breakthroughs on the Accounting Technology Lab podcast, recently discussing Microsoft’s Agent 365. More and more, AI is being thoughtfully built into products, and the build-versus-buy discussion is warranted. For unique needs, building is viable. If you are less experienced at using AI, buying your applications is wise. You can review the licensing costs and compare them with projects and tasks to see whether the work completed was faster and easier with AI. I’m fortunate enough to have easy time tracking for costing, and I can see how long various tasks are taking (pretty universally less), so I have real measures of 20, 40, 60%, and more, in productivity gains. One AI Assurance platform I reviewed over the past week has reported a 60% increase in productivity among firms using it.
So, What Has Changed with Your Strategy Using AI?
My guess is that each of you has one or more unique skills to help clients and businesses. Historically, you have used your skills to solve compliance problems, including tax returns, audits/comps/reviews, and Client Accounting Services (CAS). These services were transactional but delivered in the context of a client relationship. Additionally, many of you have the superpower to quickly read financial statements and to get to the root of problems. I drive my own team crazy when I can spot an operational problem in the financial statements before it is discussed. Perhaps that comes from experience or intuition. I recently had the pleasure of working with an accounting professional who could routinely spot ways to radically improve profitability in business operations with a few hours of observation. Some people can just “see” the problems and the root causes.
AI agents can be set up to capture some of your best knowledge. They won’t get tired or miss obvious issues. Currently, I do not see AI possessing a skill that I think many of you have. Creativity! If I can get you to think back, I know all of you started out creative as children. While many accountants are embarrassed to admit they have the creativity skill, probably because creative accounting is so frequently associated with fraud, creativity is your way to address and solve client issues. When a situation arises that is difficult, sticky, or unusual, I see the creative juices flow in the best accountants. Further, the relief for the client and the joy of solving the issue are the reasons many people got into accounting. Solving problems for money reactively is called consulting. Proactively identifying and resolving issues for clients is the basis of advisory work. To do advisory work well demands creativity and a client-centric, whole-of-client approach. Most of this work is best billed by the value, typically at a flat rate, as Mark Koziel, CEO of the AICPA, commonly observes.
Can AI be creative or will it get there? I suspect so. You can game AI systems to ask them for creative solutions, but we don’t want to push them so hard that they make up answers or hallucinate. Frequently, AI systems can list multiple solutions, allowing you to exercise your professional judgment to suggest the best approach to the client. Thomson Reuters Ready to Advise uses this method for tax advisory.
For example, when I find a solution that is close to what I want, or a client asks about a particular solution, I have an agent that evolved from an AI Chat that lists the top 10 competitors and includes attributes like the website, pricing, key strengths, critical weaknesses, and overall advantage against competitors. I do note that AI is getting less good at answering these types of questions because of AEO manipulating the answers, as cited above, to favor a particular platform, so exercise caution (yes, professional judgment = human in the loop) if you use this technique. Occasionally, you can process these results with one AI engine and have a second AI engine verify them. Using one or more AI platforms is easier to do with AI Agents and AI orchestration platforms.
So, What Should You Do Now?
If you have not been using AI to solve real client problems, you need to begin now. There are some critical steps:
- Set your AI policy.
- Implement your governance. Microsoft has Purview tagging available in both the Business Premium version with Microsoft Defender and Purview Suites for Microsoft 365 Business Premium for an extra $15 per month and in the advanced E5 ($60) and E7 ($99) versions. The free or $20 Microsoft products don’t have the same protections. Likewise, the $20/month paid versions of ChatGPT, Claude, or Gemini don’t protect the data strongly enough for my taste.
- Don’t use free AI tools. Your prompts and data are the product, and this is far more critical than the historical use of data by Google or Facebook. As my long-term colleague and friend, Dr. Bob Spencer always said, “if you don’t pay for the service, you are the product.”
- Be cautious putting any client data, Personally Identifiable Information (PII), healthcare information, bank or financial information into most platforms, even the paid version. While the $30-per-month Microsoft Copilot doesn’t perform as well as Anthropic Claude, its protection of client data is stronger.
- Build a library of common client questions that you can collect together to begin building your own prompts/chats that will result in a catalog of agents.
- Consider off-the-shelf AI products that solve specific problems for you. Watch the ROI of each subscription. It is amazingly easy to subscribe to so many services you can’t use them all. You may not have the problem at home, but think about paid TV streaming services. How many do you have? How many do you need? How easy is it to cancel the ones you don’t want? AI services for business are similar.
Real value comes from compliance augmented by advisory services. AI tools are providing productivity leverage in Tax, Assurance, and CAS. If you have time saved by AI Automation, how are you going to use it? If you invest some of this time in yourself and some in your clients with Advisory services, you’ll find that AI can provide the benefits promised. The result will be more work, done better, in less time, and more time for leisure and the things you like to do. That is where AI will provide real productivity gains.
Sign in to get access to this free resource, and all of our whitepapers and reports.
Download this content today!
Register Now Already registered? Click here to Log In
