021 AI-Generated Emails for Accountants Have Arrived

Oh, it's happening. The AI feature I most wanted. Somebody just shipped it. What's it mean? It's big. How does it impact the kind of bigger accounting tech space and even how you go about choosing tools now? I think there's some ripple effects. Come on in. Water's warm. Are those new trunks? Let's do it. Today, it's Jason Daly. Okay, if you haven't heard the news yet, the lovely folks over at Canopy just shipped generative email. So that is using our AI overlords to help you communicate things to clients because you're an accountant and you're bad at that. So let's run through, let's run through the press release what it is and what it isn't. I'll put a link to it in the. In the show notes. Using ChatGPT in your accounting firm. With the evolution of AI and OpenAI based architecture, almost every industry must decide how to incorporate AI into their work models, including ours. So they got a little. The most helpful thing here in the release is they got a little video that shows what it does. Basically inside of Canopy, you've got like a basic email authoring interface. It is not like super, I don't know, it's not like a full blown email client. I think, I don't know that this has really historically been Canopy's big area of emphasis is giving you a cool place to write emails. They've got an email integration where it sits on top of your Microsoft 365 or your Google workspace. Like, the biggest value there to me is it will pull in all of the emails so that you can see any communications that anyone in your team has had with the client. They do have a way to send emails out of there. Historically, I don't know that it's been like core part of their product, but it's moving that way now. So you've got this basic email writing kind of experience. But now you got a little pencil with some magical sparkles by it. You click that and it pulls up a dialogue. The AI Assistant beta, you put the subject in there, you put the recipient's name so that what it generates can refer to that person by name. And then you got a little box to put like prompts in there, what exactly you want it to say in the email. So in this little example that they give in the video which I have inspected, there's no cuts in the video. They didn't cherry pick this one. This looks to be like what it actually generated. The prompt she gives it is more Work was required. Out of scope, higher price. U okay, question mark. So that's literally the, like the, the extent of the context that they've given it. They have it generated email and I'll read what it generates. Let me get my bifocals. Hi Brad, I hope you're doing well. I just wanted to check in with you regarding the additional work we discussed for your project. As we went further into the project, we realized that more work was required than we initially anticipated, which means that the price is also slightly higher than originally quoted. I wanted to make sure that you're comfortable with the new price and that we're still within your budget for the project. Please take some time to review the new quote and let me know if you have any questions or concerns. We're always happy to work with you to find a solution that works best for you. Thank you for your understanding and we look forward to continuing to work with you. I mean, totally fine. Like the prompt they gave it was scant on details. Like maybe a better version of that is this is why we're over scope and here's how much I think the price is going to increase. I think that would have been a more interesting generative email. But you know what, it's pretty darn okay, which is how a lot of AI things are right now. And I think like my mental framework for this sort of thing right now is that I will rarely just send out the thing that it generates, but it is still better than starting from scratch. So that is how it works. If you are generating an email from scratch. A couple other odds and ends that you can do with the magical pencil button, you can select some text in the email and pick the AI option to rephrase the selected text, to revise it to be friendlier, revise it to be formal or correct grammar and spelling. All pretty basic stuff, but honestly all pretty handy stuff. Like that's all stuff that I use ChatGPT for right now. And I would absolutely prefer to just have a button inside of the app that I'm using to do it. So that is handy. That is not like rocket surgery by like technical, from a technical standpoint. But you know what, people are shipping stuff and so good on Canopy for making this happen and actually getting it out there because that's how you learn. That's how you're going to figure out what's going to land and what people are going to use long term and what they won't. It's not by sitting back and waiting. Now, interestingly, today Now Carter Gray posted on Twitter, if there was a bot that could do any task for you, what would it do? And naturally, because we're accountants, everybody replied with just these vastly different highly technical little minutiae things that frustrate them, most of which were different. But the like the single common thread where maybe, maybe a third to 40% of people replied at something to do with email handling email for you or saving time on just the amount of time that you spend on email. And this is why this has been like the AI application I'm most excited about. So just to kind of set some context why I think this is really valuable, here's something I wrote on Twitter last December. I said a truly transformative AI application for the accounting profession is a high context way to autocomplete emails. General AI autocomplete is just around the corner, especially as Microsoft is tied up with OpenAI. Won't be long until that's just part of the email experience. That's copilot. They announced that a couple of months ago. Now I said but its intelligence, that being Microsoft's solutions intelligence will be limited because it won't be able to see into the rest of your workflows. So the long term advantage lies in its ability to see into the rest of your business. We went deep on this in the here's your practice management system on AI episode and I actually said here, for example an app like Canopy Tax that handles file management. When you get an email requesting a copy of your 2020 tax return, your email client inside Canopy suggests a block of text and says hey, here's the file that I'm pretty sure they're asking for. Do you want to attach it to the email? Now in my like prediction AI prediction thread that I made at the beginning of the year on January 1, I said generative email replies was the thing I most wanted. And on like a guesstimate of timeline I'd said how quickly will our industry move on this? Not fast enough. We probably won't see a mature version of this in 2023, but in a few years time this functionality will put the file final nail on the caution of traditional desktop based PMs. So I said we wouldn't get a mature version of this in 23. What Canopy's done, it isn't a mature version of this. It's like a first step. It's an excellent first step. It's an important first step that is necessary to get to the even cooler implementations of it. But does this revise my timeline or I don't know, expectation of when we get a really rad, mature version of this that can see into the rest of all that you do. Honestly, I'm feeling optimistic today. Maybe this is because this is still new news, but it's kind of got me thinking, like maybe somebody will move on this and we'll actually have a super rad version of it by the end of the year. Now to just a recap, like, what does a more mature version of this look like? What we're doing right now is basically what you could do in another tab in ChatGPT. It is excellent that it's pulled into Canopy. The reality is 99.9% of people have not used ChatGPT. They're not gonna hop over into that other app day to day and stuff like this will be the first time they interact with this technology now that it's built into the app they already use. But the fancier versions of this is that generative reply that takes into account everything else in your business. The, you know, the next step is probably just taking into account an email thread. So right now my understanding is this is just for drafting an email. Whether it's a new email, whether it's a reply, it's just going to draft it based on the prompt that you give it. But the next version of this is seeing that most recent reply from the client, maybe seeing that whole email thread. It's a little harder to build around because that could be implied impossibly long, but it enriches the context with which you reply. And that's ultimately what's going to make this really helpful, is the more context it can take into account when you send that reply. You think about all the things that are happening in your business and the amount of information you have at your fingertips. When you reply to an email right now, you are truly shooting from the hip. It is all just what can you remember off the top of your head about that client. And in the grand scheme of things, that's really not much because you're not looking at their tax return unless you open it up. You're not looking at their accounting. You're not looking at all these different things that may actually be helpful things to have in mind while you are replying to that email. And so the more mature version of this is the one that pulls in a bunch of that stuff. Another wrinkle, wrinkle here, positive wrinkle. Another good wrinkle here is ultimately the quality of your experience with generative email will not just be how Good is what it suggests, but it will be how true to me does what it suggests sound. So if I send that to a client, I, I don't want there to be a day where they realize, oh, kind of beyond this day, the stuff they started sending me was like machine generated. Because if you've seen a lot of it, you can kind of tell the people who are posting GPT generated comments on my LinkedIn posts, I know who you are. It's not that hard to tell. And people will be get like more aware of what this stuff looks like. So much of the quality of that output or the usefulness to me will, will be how true it sounds to me both in general and my voice. But also within the context of that conversation, if we're having a real happy slappy talk back and forth and then somebody comes in with a total robot response, that's weird. Also, if that's just not how you generally write, that'll also come off wrong way back when I did like a ChatGPT video at the end of last year on the main channel, one of the things we ran through was dropping a, you know, a blog post or something that I wrote into ChatGPT and saying, Give me a one paragraph explainer of my tone that an AI can understand. And so you can basically copy paste that paragraph that describes your writing style, paste that into ChatGPT or any other AI writing experience and say, use this one paragraph descriptive tone if you have something like that that becomes part of kind of your identity within your practice management system. I mean, it can already see all of your old emails, right? So like building a profile for how that person emails and how they sound in their tone of voice, that's gonna make that perceived experience much better. Whether that's actually like ultimately a better reply or not. Maybe it doesn't matter, but the value there is that it will sound true to you. Because none of us are looking to now switch to sound like sounding like total robots. So that's another way, like, interesting way that we can build on this. But exciting times, super exciting times for most of us, but maybe not for some of us. Let me explain why I've talked about this a bit, but you know what? I think this really does sticks a nail in the old coffin of every system that doesn't have email built into it. Hear me out. It's a hot take. And I'm saying it like I'm there, I'm probably 75% there on this hot take. I shared this on Twitter the other day. I Take that's inconvenient in many ways and steamroll some products I love. Email is now a mandatory aspect of a practice management system. We've talked about this before, the idea that AI is only as good as the context that you give it. So ChatGPT is pretty good, right? But it has zero context. It's everything on the web. The version of that that's way more helpful for me is the one that can see the way that I work and, you know, my information, all that stuff. And we spend a whole heck of a lot of time emailing every day. And for AI to be helpful with my email and it needs to see into the other things that I do, right? Otherwise you've got this like orphan set of data that is not then informing how you reply to emails. In the past, email didn't do anything meaningful with all that other information. It's not suggesting stuff based on projects that are outstanding or that meeting transcript from when you talked to the client a couple of days ago. AI enables this in a really cool way, pulling in all this other context to impact how it suggests that generative reply. But if all that data sits in a different system, that email is just going to be dumb. It's not going to have the ability to do any of that stuff. We're going to circle back to this. Let's look at Canopy's kind of approach to their initial pass at AI and let's look at what a couple of other tools use that tools that a bunch of us use, ClickUp and Notion and think through what this ultimately looks like. So right now I'll link this in the Show Notes ClickUp AI overview. A lot of people were excited about this, myself included. Here's what it does today. You can do a little forward slash in any of your ClickUp docs and have it generate something with AI, very similar to what Canopy has in its email client now. You can have it summarize for you. You can have it continue writing. So from a certain point, continue based on what's above it. You can have it explain things to you. You can have it make something better, make it shorter, make it longer, fix spelling and grammar, use simpler language. Honestly, a lot of stuff that people use ChatGPT for right now, but it is nice to have it built in. But that's it today. It's not doing anything meaningful with all of the context that you have in that system already. It's not going to write it according to everything else that it sees in there. It's not gonna Write it in your own voice. It's not gonna make meaningful decisions based on the other stuff that's in there. Now that's today. Doesn't mean it won't do that stuff down the road. I think it definitely will do more of that stuff, but that's where it's at today. Now Notion. Let's take a look at Notion. Cause it's pretty similar. Notion actually moved really early on the AI stuff. They've had their AI product out for a few months and it's pretty much the same as what ClickUp does with maybe just a little bit more functionality. So like here they've got an example where you can summarize something and ask it to make a table and it makes like an inline little like spreadsheet style table like Notion has. But at the end of the day it's largely doing the same sort of things that ClickUp is doing. And that's cool. And in certain use cases like company documentation, stuff like that, that's really helpful. But you know what? It's not, it's not making my email any better. And it's also building organizational documentation in a place that is totally segregated from my email. And that's the bigger long term problem. I don't see a future timeline where Notion or ClickUp let me send emails to my clients. Like I think ClickUp has like this real basic kind of built in emailing thing, but it's fundamentally not an email client. Right. And so if the ultimate value of AI to me is gonna be limited by how much context that communication platform can see into, then should I be building organizational information and knowledge and all that stuff on a platform that my email is never gonna have eyeballs into? Like I said, it's a hot take. I'm maybe 70% down the road to this hot take. But if I'm making technology decisions right now, I'm thinking long and hard about that. If my team collectively spends 20% of their time on email, a third of their time on email, ultimately my tech stack needs to enable email. Like that email can't sit out on an island. We can't be doing that stuff in a totally manual way like we do now. Good job, Canopy. Bravo. I'm not saying I called my shot, but if, you know, you know, other companies in our space doing AI stuff I think can be takes the prize right now for like the most meaningful thing that's been shipped in the accounting space. Client hub, doing a cool little thing around task generation. So if you have a project, it will generate like boilerplate tasks in there based on the project title, the sort of things you would expect to need to do for that project. Again, not as like here is the ultimate answer for the things you need to do for this type of product for this type of project. Because it doesn't really know enough information to say exactly what the right thing to do is. But it's another, another one of those things where it is probably better than starting from scratch. You can use some of those tasks, you can modify em, you can wipe out the ones you don't want. And like that Canopy email example, it's better than just starting from nothing. Financial sense also they are doing something like shipping an AI feature every week for the next eight weeks or something like that. The first couple have been similar to what client Hub's doing and they're in task generation. So when you create a task template you can have AI generate a set of tasks for that task template and then within a project you can also have AI generate kind of a set of subtasks for that project. And it's all good stuff. And like in order to get to the earth shattering applications of AI that impact everything that we do, like you gotta start somewhere and we're going to find out what's the most valuable stuff and what isn't the most valuable stuff. And again the answer is not to sit back and wait and watch everybody else do it. The answer is to like start investing in that now and see what's going to resonate and what isn't really exciting now like a counter argument to the whole discussion of one system to rule them all, which I fundamentally don't like. Like what I loved about like kind of the cloud paradigm and being able to integrate all your cloud apps was just being able to pick and choose best and breed of all these different things. But it sure feels like we're going in a different direction now. Like coming more back to all in one, which is generally bad for innovation, bad for like small companies creating cool novel things. Those cool novel things will still be out there, but it's getting harder because for example ClickUp, I've got all my organizational history in there, but what about my files? Like my email needs to not only see into all the things I'm doing in ClickUp, but it needs to see into like the contents of all of my work paper files and stuff like that. When a client asked me for a banker letter and I've made 200 of those banker letters before, I want AI to see what those 200 banker letters look like so that it can generate one for me, that's 95% of the way there. Where do like the little startup companies plug in to that ecosystem? It's harder. It's not as simple as like integrating cloud apps as we have in the past. This isn't as simple as pushing a bill from Bill.com to QuickBooks. It is like all of the context from the contents of the files to projects and tasks and all that stuff, which in a software and like an AI software system is, I mean it's called embeddings. It's embedded in a certain like, format that they use to then make it searchable and discoverable when you're say, writing an email. So the notion of like integrating something to be able to see into all of that stuff, it's very different. That's much harder than like a vanilla cloud integration that will push, you know, a task on a proposal acceptance or something like that. So it is getting harder to integrate that stuff in a meaningful way that the AI can see into. Will that be necessary for absolutely everything we do? Maybe not. But it is taking us in a direction that, that more so than in the past, would lead me to look toward a all in one option that is leaning into AI. Those seems, those, those platforms seem like over the next three to five years they're gonna be the big winners and some of the incumbent folks will get it right. And five years down the road, some of those folks that are getting it right, they're not out there yet. I think this is a big enough paradigm shift to where you will see companies spring up simply because of their AI expertise. In the same way that we saw like SaaS developers kind of blow the socks off what desktop developers were doing for the cloud. This is a big paradigm shift. And so I wouldn't be surprised if we see new providers come in with like an AI first solution that are doing some cool, exciting stuff. I tell you what, like, even if I am, even if I was running my, my cast practice on ClickUp, so like, I know the pains of the, the notion of, oh my gosh, is this not the right platform? I sure hope it is. Even with all that, even if I'm running it on a system like that, that maybe there's some concerns longer term. Man, I'm still pumped because the destination here is such a better place than where we are today. It is such a more intelligent version of the tools that we use right now. Whether it's the tool that you're on today or not, it's just fundamentally better than how we do our work today will make us look smarter for clients when we can see into all the other context in the firm and maybe a conversation that a different client had with them in a zoom meeting. It's just gonna make the way we operate so much more intelligent. So, I don't normally, like, go deep on a thing that a specific company did, like we did today, but Canopy deserves props, like, for actually putting something out there. There's actually a lot of risk to being the first one to move on that, but somebody's got to do it. Like, who's going to take the lead and actually take us to where we want to get to. So good on Canopy for doing that. That's all I got today. Tomorrow, we're going to have some more fun. So come on back. I'll be here by.