What's Next For Accounting? - Issue #135

📅 OpenAI Launched A New AI Model!

Their spring update was a fun watch, about 30 mins. Here are the 3 most relevant updates for accountants:

  1. Better, faster, cheaper

    The new GPT-4o model is around 2x as fast, at ½ the cost. Usage limits will be higher, and they’re even making it available for free to non-paid users. They say it’s equally capable to the last model, but in human evals it outperformed every other model out there. Paid users should have access to the new model already.

  2. Really good voice support

    This one is rolling out over the coming weeks, but the ability to talk with ChatGPT is getting a massive upgrade. It’s much faster, you can interrupt it, and it’s super expressive. You can tell it how expressive you want it to be, to talk faster, talk like a robot, sing the next sentence, etc. It sounds trivial, but if you ever used Pi, you know how novel a really good voice assistant is. But all of that is made better by 👇

  3. A ChatGPT Desktop App

    A desktop app that can see what’s on your desktop. Meaning you can have a conversation with it over voice as it sees what you’re doing on-screen. This eliminates the need to navigate to the ChatGPT web app, and means it can now be a helpful companion where ever you work.

The desktop app serves as the first step toward a helpful assistant that can complete tasks for you on your desktop. These desktop agents are the most significant near-term application of AI that will significantly change how we work.

💎 This Week's Sponsor

📅 This Week

  • I’m currently in Austin at RightNow - would love to see you if you’re here!

  • The hottest book in accounting right now is Down To 40 Hours by Geraldine Carter - I chatted with her about it on the pod

  • I roasted a firm that pivoted to vCFO work

  • Lots of AI this week:

    • OpenAI dominated the headlines with their announcements, teasing other cool capabilities like the model’s ability to create sounds

    • The next day Google announced a bunch of AI stuff at their annual dev conference, big updates summarized here - Nothing got me particularly excited as their best models still lag behind those of OpenAI for most applications

    • I suspect we’ll soon reach peak chat assistant saturation, where crammed into every app we use will be an overly cheerful assistant waiting to do your bidding

    • But I don’t see these chat assistant being useful long-term. I see the best AI applications being one of two things:

      • A feature integrated directly into an app (not an assistant) - For example a summary of an email, a draft generated by AI etc

      • Or universal AI agents, an assistant that works across apps. The assistant built into Excel will be less helpful than the assistant that can see both the excel file, and the PDF that’s open alongside it

  • But none of them can top this app that creates child-like crayon drawings. Here’s an accounting conference expo hall

  • I shared how I’ve built my online network, and how doing the same can lead to a more profitable firm

  • This app is an example of how AI stands to improve workpapers - similar to Google’s NotebookLM, workpapers live in the middle, AI chat & info extraction lives on the right hand panel. It’s a great example of something that’s possible with today’s tech, that we need smart people to bring to life

🔮 Q&A

What is the right format for an SOP? Text video etc Link
Video + text. A short video is best for teaching a process the first time, but a pain to scrub through after that, and a challenge to keep updated. A text version with screenshots of each step (using a tool like Scribe or Whale) allows for easy reference and updating later on. The text SOP should live directly under the video.

How do you ensure SOPs are kept up-to-date by the team? What systems or incentives are needed? Link
Updating them must be tied to the team's performance evaluations and incentives. Randomly pulling SOPs, and having a different member of the team try to complete a month end close is a great way to keep folks on their toes. It must be part of how their performance is measured.

How do we build systems that accommodate exceptions without being wildly inefficient? Link
Example: You have a 3 day response time expectation, but a client is selling a biz and needs more frequent support. This isn’t a reason to make the global standard fast enough to handle this exception. Design a system around what’s normal, then create a different system for handling exceptions. Tina, if Anne calls you can let her straight through for the next few days. This is also why you shouldn’t have your email open all day just in case something urgent comes through. Make urgency monitoring someone else’s job, so you don’t torpedo your focus every day, whether it’s necessary or not.

Putting AI chatbots on accounting firm websites for lead generation and client support is becoming more common. How will this be perceived by clients? Link
Right now whether it’s useful is almost secondary to whether it’s acceptable. The jury’s out on what will become normalized over the next couple years. It may be a good hack now, particularly if it can be useful, but just how far we can take this is still TBD. Could you charge for access? Is there a version that could be productized for non-clients? The answer is yes, but today it still feels a little cringe.

😍 From The Community

My private peer networking community Realize for firm leaders, and it’s now reached its cap of 500 members 🎉

▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮▮ 500/500 members