What's Next For Accounting? - Issue #75

Author

Jason Staats
March 24, 2023

This week's big news was Microsoft's announcement of M365 Copilot. They're adding GPT to each of your MS apps, which largely do what you'd expect: analysis of what's in the doc, generative functions within the apps etc. But importantly they also enable shared context across apps, and via a new interface, Business Chat. A roundup of examples:

  • In biz chat, summarize the chats, emails & docs about [client] from the past 30 days

  • In Outlook, summarize information from [Excel doc]

  • In PowerPoint, generate a 5 slide deck based on [doc/meeting/excel]

  • In Teams, I just joined 10 minutes late, summarize what I missed and ᵈᶦᵈ ᵗʰᵉʸ ᵗᵃˡᵏ ᵃᵇᵒᵘᵗ ᵐᵉ

  • In Outlook, draft a summary of my meeting with [client name] yesterday

  • In biz chat, summarize the emails I got over the weekend, and create to-do's for the most time-sensitive tasks

More impactful than razzle-dazzle features will be AI's ability to pull together relevant context across apps. This only works in an environment where a system has access to a wide swath of information. It's the reason why in our industry, practice management systems have the most to gain, and the most to lose by not moving quickly on AI.

The AI goldrush we're in was kicked off by rapid improvements in the price & accessibility of embeddings & vector search, and the quality of GPT. But you're already noticing, everyone's sort of doing the same thing, right? Chat with everything (GPT), and making context more helpful (embeddings & vector search).

Long-term differentiation boils down to who has access to the most valuable context. The email client that can see your meeting transcripts will be better than the one that can't. The accounting transaction classifier that can see receipts & invoices will perform better than the one that can't. The project management system that can see into chat & email will squash the one that can't. The request system that sees you already have that information will be better than the one that can't.

Forward-looking folks are engaging in a land grab for the most valuable context. Transcripts built into everything is one such example. Ultimately services that have access to context, but don't anything of meaning with it, will be displaced by those who do.

  • I made a video about an ERC firm who lost its tax staff in January 🪦

  • If you missed the accounting transaction classifier automation from last week, not to worry, you'll likely see it built into a number of apps in our space soon 😍

  • I re-ran the Steve test - I previously ran this to see if I could use ChatGPT as an intern - now updated for GPT-4 (spoiler: 📈)

  • As everyone and their mother builds chat into their app, most are not using GPT-4 because it's bloody expensive. Ranging from 1 cent, to upwards of 25 cents(!) per prompt. This will come down over time & quality of chat will improve

  • Microsoft opened access to their Notion clone

  • I wrote an article summarizing the state of the tax pro software ecosystem for developers

  • Hypertype is a generative AI assistant that can see your email history - but is only seeing email history enough, or will the tools that can see everything else outperform

  • I did an AI Q&A for the questions you were too afraid to ask

  • Google rolled out a waitlist for their GPT equivalent Bard, but did anyone care? Does Google have too much to lose in their core business of search to make something of meaning with Bard?

  • The fastest way to a tax professional's heart is...

  • Github announced their chat assistant for developers, and it's a different UX paradigm that puts a chatbot alongside the app itself - I suspect we'll see more of this

  • Midjourney launches v5 of their generative image model, and a bunch of people are about to be dooped by photorealistic images of public figures

  • Alleo is an AI app that creates a "Lifeboard" - everything you do all in one place, by gobbling up all your context

  • Wingspan raises a round - payroll software for contractors bc worker classification is so 20th century

💎This Week's Sponsor

🤖 Automate My Thing

AI With Zapier & Make

Developers are wowing us with neat new AI stuff, but what about us normies? Let's check in on Zapier & Make's OpenAI integrations to see what's possible today:

ZapierCheck moderation - Send some text to OpenAI's free moderation API endpoint, to check for abuse.Generate image - Generate an image with DALL-E by providing a text prompt, the number of variations you want (1-10), and the image size, up to 1024x1024Send prompt - Generate a GPT completion based on a prompt - now supports GPT-4, this is probably the easiest way to build the transaction classifier I shared last week into your toolsSearch embeddings - Provide a search query + a list of documents to search across. Embedding are generally most helpful across a big volume of data, so I'm unsure what the use case would be here

Make currently supports the same 4 connectors that Zapier does, in addition to an API call connector, to simplify the process of building your own custom API call.

What are you building? Reply to this message, I'd love to hear about any automations you've been working on lately, AI or otherwise.

😍From The Community

My community Realize is a private space for people to swap ideas & collaborate. At $180/mo it’s also the way I finance everything else I do, pay a full-time creative director, and a part-time developer in an effort to make the accounting industry a slightly better place 💗 The community is currently about 280 members strong.