Skip to main content

Accounting

The Accounting Technology Lab Podcast: Zoho Analyst Day 2024

Hosts Randy Johnston and Brian Tankersley, CPA, offer a report on the 2024 Zoho Analyst Day conference. The event, held in McAllen, Texas, gave attendees an inside look at tech developments the vendor is making, as well as the opportunity to speak one-on-one with developers.

Hosts Randy Johnston and Brian Tankersley, CPA, offer a report on the 2024 Zoho Analyst Day conference. The event, held in McAllen, Texas, gave attendees an inside look at tech developments the vendor is making, as well as the opportunity to speak one-on-one with developers.

Use the video player below to watch, or the podcast player below to listen to the podcast.

Thanks for reading CPA Practice Advisor!

Subscribe for free to get personalized daily content, newsletters, continuing education, podcasts, whitepapers and more...

Need more information? Read the FAQ's


Audio Player:

Transcript (Note: There may be typos due to automated transcription errors.)

SPEAKERS

Brian F. Tankersley, CPA.CITP, CGMA, and Randy Johnston.

Brian F. Tankersley, CPA.CITP, CGMA  00:00

Welcome to the accounting Technology Lab sponsored by CPA practice advisor. With your hosts, Randy Johnston, and Brian Tankersley.

Randy Johnston  00:10

Today, welcome to the accounting Technology Lab. I’m Randy Johnston with my co host, Brian Tankersley. And we’re going to cover the Zoho analyst Day held in McAllen, Texas. In early February, we think it’s important that you know a lot about Zoho O’s event, because of the significant technology changes that they announced during Zoho day. Now, we’ve had the good fortune have been around the host since the inception, have known much of their management team throughout the years. But Zoho day was intended to be a press event for analysts. And they flew hundreds of analysts from around the world into McAllen, Texas. So Brian, you had the good fortune of being at the event as well, and many of the other analysts for us press were there as well. So what some of the things that you’d want us to know about the Zoho analyst day 24, and Zoho in general? Well,

Brian F. Tankersley, CPA.CITP, CGMA  01:13

you know, the analysts they bring in are like people from IDC And Gartner and Forrester and others like that. And they, they go through and they talk about their strategies, and they talk about the products they’re going to work on. Now. They told us a lot of things that we can’t talk about yet. Okay, so it’s gonna be an interesting year from, from Zoho, for sure. But what we can talk about is some of the AI and how they’re incorporating AI because you’ve as Raj, you that’s that Zoho is one of our favorite, both of our favorite people to talk to about technology, just because he’s so insanely smart. And so articulate, and, and just, he’s just a once in a generation kind of talent. And so he, he actually did some stuff on small bit on on large language, small language and medium language models, and some other things about how AI is actually incorporated into the applications. You know, the the event generally was a, you know, a lot of a lot of the people from big cities complained about going to McAllen because he had to change planes, you know, but that’s, that’s life in the big city. I mean, it was a I thought it was a great trip. It was great to go somewhere really warm for a few days. And, you know, there were probably as many there were probably 100 analysts there, I’m guessing maybe 150. But then there were probably another 100 or 150 Zoho people there. And many of most of them probably were on probably came came from overseas. So it was a, it was a wonderful trip. And it was a great time to catch up with old friends, you and I got a chance to catch up with Val steed from Zoho, as well as Doug Sleater. And Ted Needleman and Bob Scott. And a lot of a lot of people that we’ve worked around for many years. So it was a was a wonderful event for many, many reasons.

Randy Johnston  03:13

Yeah, I appreciate that. Well, now, so hold themselves. I have I have quite a history, of course, they’re privately held and publicly responsible. They’ve been around for about 28 years, currently running 15,000 plus employees. Their stack is a collection of many applications. 50 Plus, the way I count them, it’s actually 60. Plus, they currently are deployed in over 150 countries, they have 100 million plus users on their accounts, and 750,000 plus paid companies on the accounts. Now the announcements that were made throughout, like Brian has said, some are non disclosure. So we would love to talk about some of the other things that are out there. But they do definitely have products grouped together. Now, of course, it’s been six years, I believe, and no seven years since they first announced selling all of the products as a bundle known as Zoho one. But you can recognize that they have sales and marketing. They’re well known for their CRM product. And they just recently have a tiny CRM inserted into the Microsoft 365 platform called begin. But they also have IT tools for customers service, desk and assist and lens and manage engine. They also run a complete productivity suite with email and equivalent word processing, spreadsheets and so forth. Frankly, the business intelligence and analytics and others are good. I was surprised and forgot, and I was actually stunned that my memory didn’t help me recall. All their HR products started in 2008. I perceived it was just a few years old, as it turns out is longer. But then the finance Zoho books and inventory we’ve done another podcast on invoice, which we think is an ideal product for those of you to put small clients on, and so on. So Brian, I’ve kind of ranted and there’s so many more products that I could have spoken about. But I know you have a few other different favorite ones in this particular stack.

Brian F. Tankersley, CPA.CITP, CGMA  05:31

Well, I’m a big fan of Zoho social, which is a tool that integrates all your social media management, it also integrates with their CRM tool. And one of the things they did announce is that the integrations between the applications are going to become more seamless in the future, which was pretty exciting. I’m also a big fan of creator, which is one of my favorite tools to, to integrate and create dynamic workflows. It is a low code, no code solution. And so you can write the code in a language called deluge, actually successfully wrote some daily Cash Code with chat GPT. So it’s a you know, it’s not brutally difficult to do it. And, you know, again, it’s a, it’s a pretty robust system. You know, I think their BI tools, they actually talked about some of the AI and their BI tools and some of the things coming in a bi session, I was very excited about that. And I think it’s, I think it is, with the new user interface that they announced, I think that it makes that tool, a much more worthy competitor for the tableaus and clicks and Power BI ‘s of the world.

Randy Johnston  06:44

Now, I know you have some others on the list that you also didn’t call out with the Data Prep tool, and also the low code, no code tools. And again, I I go back to your digital plumbing word, which I still admire. But are there other things you’d like to say about the Data Prep or the low code, no code world?

Brian F. Tankersley, CPA.CITP, CGMA  07:02

Well, the Data Prep tool really just does is think of it it’s very similar to Power Query in Excel and the central power query that you use in Power BI. It’s really just an extra tool for extracting transforming and loading data. So you get data out of systems in CSV or in database format, or in a, you know, XML or any other format like that. And then you create a routine to clean that up and reformat it into the format it needs to be, and then you go ahead and load it. And they gave us the the number of lines of code of data that they’ve imported with that tool, but it’s in the billions or trillions, as I recall, it was a stupidly huge amount of data, you know, they get the custom solutions there, the way of thinking about those as creator is kind of the kind of a low code, no code tool catalyst is a pro code tool. Flow is a tool for kind of this digital plumbing that we talked about. And they are investing in flow now more than they have in the past, simply because there is a need to do integration. So flow, for example, and the BI tool, already have integrations with QuickBooks Online, and a lot of other applications that you might use. So these tools may be accessible to you without having to do any of the hard work of interfacing or dealing with tools like Data Prep. But Data Prep, again, is menu driven, very easy to use. And I think you’ll find it to be to be very, very helpful in the conversion process.

Randy Johnston  08:37

Well, thank you, Brandy. So you know, we also heard about the corporate values of Zoho. They are one of the, to my knowledge, actually the only company that have been profitable every quarter since inception. So over 28 years, they’re privately held. So they do not publish their numbers. But since I know the founders, well, we’ve had those discussions in the past. So they’ve been very, very prudent about their financial position. They, by module have an extreme customer focus, trying to deliver features and items that the customers want. And then they also have done a variety of social efforts, including their Zoho University, the employment of, you know, lower income people, the ability to form operations in smaller communities to drive local economies, they’re doing that globally. So the corporate values from the top which Sridhar and his team come all the way down through the business. So any other key items on corporate values, Brian, you

Brian F. Tankersley, CPA.CITP, CGMA  09:51

know, Zoho actually goes in and does does a lot of work, you know, with homeless folks. They actually have a I can Recall the word that Sridhar and Rose you use to describe it, but there’s actually a term that is that that means that if somebody is homeless and needs food or needs to sleep, one is obliged in Indian culture to provide that to them. And so Zoho is actually set up in some of these smaller communities where they are, you know, little, little sleeping pods. And if somebody’s hungry, and they’re near a Zoho office, they can just walk in and eat with everybody else, and walk right back out. And so they, they really are doing some interesting things from a social conscious perspective.

Randy Johnston  10:40

Yeah. And one other thing that I know is near and dear to your heart is working with the homeless people in the Knoxville market. So you’ve, you’ve done all sorts of things there through the years, and we encourage our listeners to help in their own individual markets act locally, think globally type of thing. Now, this technology company is strong, it is clear that their data center built out are becoming compliant, I had the good fortune of speaking at account techs in Toronto in November, and the day after they introduced Soho practice, which we have recorded in a separate podcast for you information on Zoho practice, which we recommend broadly. And for those of you in public practice, we’re going to talk more about so hos public practice strategies in a future episode. But the data centers in Toronto and Montreal allow the Canadian regulatory environment to be followed. They have data centers in every significant jurisdiction, including the EU, and they’re standing up data centers in Mexico and the GCC this year. So that’s a big deal, because they own all of their data centers and all of their equipment, they are not built on anybody else as your Google. AWS, as Oracle said, they’re not built on anybody else, they own their own infrastructure, which gives them greater control over privacy, and speed. And so that’s kind of a big deal. Observations on the data centers brand.

Brian F. Tankersley, CPA.CITP, CGMA  12:15

Well, they’re they’re getting, they’re adding four in this next year, you mentioned the one in Mexico and the two in GCC, they’re also adding one in Singapore. But, you know, this really gives you the opportunity to deal with these emerging privacy statutes that are out there that require you to have data stored locally as opposed to in other countries so that it’s not the so it’s within the range of the the reach of the long arm of the subpoena. So that’s, I think that’s a very good deal here. But you know, I think the, I think the other thing about this is that, you know, they’re really a platform, and not just a not just a piece of software. And I think that’s really important, because they really control more of their own destiny than some of the other platforms, you know, with books, then some of the other platforms like sage intact, and others do. Also, since they’re doing their, their, their AI in house, they don’t have to go out to outside API providers who may have different privacy, privacy requirements, then, then maybe you would like, but this is a company that is very, very aggressive on privacy, and is not interested in in what they refer to as surveillance capitalism.

Randy Johnston  13:33

You’re absolutely right on that point, Brian. And the other thing that I do not have the exact date, but I think it was 2012, when they determined they were going to build their own acceleration hardware. And so they built their own risk and Sisk technologies. And we’re just now seeing that type of strategy announced by Microsoft and Amazon and others in probably in the last 90 days, they’ve been doing it for over a decade, which is interesting. They also accelerate a lot of their technology with Edge, acceleration. And then they also deploy local pops for acceleration. So they know that they’re assess product, they can see where their traffic’s come in that coming from and they’re doing everything they can, technically to make the product run quicker and more reliably. So I think they’ve got that done fairly well.

Brian F. Tankersley, CPA.CITP, CGMA  14:27

Now, let’s back up here for a second, okay, now, why don’t you look at all this stuff in the infrastructure and you go, Why have I never heard of these people? Okay. And the answer is they spend marketing to the extent that it makes financial sense to do so. As opposed to you know, and marketing is not the, you know, the engineering, it’s the thing that drives this. Also, remember that this is a, you know, they they talked about some new customers they had that had, you know, millions of users. And so the the thing about this company is this is an enterprise company, and so they’ve got to do deliver. And It’s browser based everything. So, you know, with that browser based world, they have to stand in to deliver every day. And so that’s one of the reasons why you wouldn’t have heard about a lot of these innovations is that honestly, they are, you know, they’ve just kind of flown under the radar. They enterprises know who they are, they white label a lot of things, including their BI tool as embedded bi. But the thing to know about them is that they’ve got, you know, they, there’s gonna be a lot more going on. And you might be might think, for the company for the profile that you might have seen for them, you know, Superbowl ads from these people, you don’t have to worry about, you know, they’re, they’re not going to be a double digit percentage of NFL revenue from advertising. They just make stuff that works. And that’s a different strategy than some others.

Randy Johnston  15:50

It certainly is. But all 15,000 of their employees are on the proud platform, just like the you know, I guess the way we used to say it is, we like companies that eat their own dog food nowadays, it’s the drink their own champagne. But, you know, as it turns out, Oracle is working that way. So who’s working that way, and so forth. And those are important points, because everybody’s on the platform. But the other major learning, Brian, for me, was the concept that Raj you advanced, which was that one AI model doesn’t fit all. Now, I kind of knew that. But it was not anywhere nearly as clear in my mind before Raj, you spoke as it was after and I’ve had time to ruminate on this a bit. And it is really clear that one model doesn’t fit all. So knowing that that is the case, the need for multiple models, which they defined as narrow, small language models, medium language models in large language models, work, interesting differential. Now there’s been an AI frenzy on large language models with chat GPT and copilot 365, and the new Google Gemini. And I value those large language models. But I also understand the places that other models make more sense. And, you know, that became real clear with some demonstrations. So you know, Brian, I know you know this stuff pretty well, as well. But you know, when we look at the narrow models, there’s narrow models are used for purpose driven, one task at a time tuck type of functionality, things like grammar, which you see today in Microsoft 365, or Grammarly, for example, which Zoho can do in their own word processing editors throughout their systems, the natural language processing which they’ve had in their Zoa, Zoho intelligent assistant for quite some time, and I wrote about this in my February column for CPA practice advisor as well, because it was so interesting to me. But narrow models also can do predictions and events and anomaly detections, particularly for time series. It also enables very domain specific capabilities in finance, and Legal and Security. But the key here is that multiple models are deployed across Zoho. So Brian, I think they had a really nifty example, on the narrow models and small models, but do you want to pick up and explain the small models to our listeners today.

Brian F. Tankersley, CPA.CITP, CGMA  18:40

So there are as we’re thinking about the small language models, just a reminder here that, just like there is a need for different scales of databases and different scales of solutions. That’s the same thing here. And so the beautiful part about the small language models is that these have three to 7 billion parameters, they are easier to fine tune. And they are less emergent that the medium language models and large language models, the inference and the analysis is done on the on the normal CPU as opposed to the GPU. So it’s much more efficient to create these on the local GPU. So they use this for translation, noise cancellation and transcript generation. And so as we think about, you know, a, we think about the models here, if we ran a standard narrow model against the recognition of this receipt, it would pick up the $100 because that’s the big number in bold print on this receipt. However, if we put the small language model that has that has some context and intelligence and other things in here, it can pull out of here that the amount due the amount paid for this was $9 and not $100 and not the $91 and change that happened here. So that that’s really that’s that’s the thing that we’re trying to talk about here. And we’ll talk about that a little more here in a second.

Randy Johnston  20:06

Well, before you leave that concept, which is important, you know, the word that Brian used was that small language models are less emergent. And if you recall from our prior Technology Lab podcasts on artificial intelligence, one of the things that we’re concerned about with artificial intelligence is the errors made hallucinations, if you will, and that many of the large language models and it turns out medium language models, do things that they weren’t programmed to do. So it’s much harder to control them. So notice less emergent on small language is good. And it actually worked well in the document or condition on the receipt. And we see other competitive vendors, Wolters Kluwer being a good example that have done a similar technique with their teammate document linker product where they used to just do OCR only, just like Zoho used to do OCR only. And then they applied these language models to improve recognition on their OCR product. And I noticed Zoho has improved recognition in Zoho expense because Brian and I both use that. And I didn’t understand why. And it’s like, oh, da, now I see why. So Brian, sorry to interrupt you, but let you pick up then with the medium language model

Brian F. Tankersley, CPA.CITP, CGMA  21:29

ideas. That’s great. It’s great. So the medium language models do take more, more time to, to train and program than the small language models. These were going to have 20 to 50 billion different parameters, you can fine tune them. They are more emergent and getting can hallucinate more than small models. But this is things like natural language query queries, where you go in and ask questions about documents or you mark the anomalous things, the things that the documents that look like exceptions, you know, $1,000 $1,000, soda purchase, or a, you know, versus a $5 hotel purchase, those would both be unusual. We can also generate intelligence over transcripts. And so we’re they’re starting to deploy those. But again, getting the right fit for this just like accounting software is a function of fit. People ask people often ask us, you know, what’s better QuickBooks or SAP? Well, it depends on whether your ford motor or you know, Bob’s flower shop. Because it’s, it’s pretty clear that one is probably better than the other in that in that example. So we have to think about the same way with these models, where the big ones and small ones are in mid or medium ones are used for different things.

Randy Johnston  22:50

So likewise, large language models have a lot of extensive training, they’re often 50 billion parameters, or more certainly chat, GPT and Bard, sorry, Gemini, the new name, I’m still using the old one there. And so hos models, and so forth, copilot. These all are large language models that have a lot of emergent behavior. Now, that’s handy. And I was using a well as recently as yesterday, because frankly, I use it almost every day, large language models to do diagnostics in very advanced ways. And these models are not easy to find tune. And there are a number of public large language models out there, the chat GP TS from open a eyes and the Microsoft copilots and the Google, Geminis, and many, many more hundreds, frankly. But Zoho has made the decision to invest in their own large language models to do content generation. And really, some of the capabilities that they describe for their intent of use, for example, was show me all the documents related to a project for my team in the past week. Well, that’s pretty common thing that we need. In fact, even before recording today’s podcasts for you, Brian and I were looking for some documentation. And, you know, Brian was able to find it with his Google tools. I was not able to find it with my Microsoft tools. It was fascinating, and very, very frustrating to me. Well, Zoho is working on making that stuff work. Well. We wanted to spend this little bit of time from the Zoho event to explain Nero language, small language, medium language and large language models. And the concept that Raju taught both of us and the entire group of one model not fitting at all.

Brian F. Tankersley, CPA.CITP, CGMA  24:48

And one of the things that that they also said is that you’re probably using AI in ways that you don’t imagine. And so the small language medium language models are not you know, they’re not going to be as visible, your use of them won’t be as visible as the large language models, because there’s just things that you expect to work like optical character recognition. So as an example, as an example here, but they actually use the different models in context. So to get context into, again, solve different problems with respect to data. So if we look at a legal workflow like this, where we need to sign a document, we email a document for signature, we view and translate, summarize and highlight we find this anomalies, then we make suggestions and trigger and other workflow. Well, as we’re thinking about this, now, this context, this context can come from Ai, for anomalies and suggestions for what kinds of things might be normal, just like my son turned in one of his first English essays in college today. And one of the things that, yes, and we were working on it yesterday. And one of the things that that I showed him in that was that you need to look at the pattern, and then figure out how you how you answer this question with this pattern. And so just like that this context, is really helpful. So you can see that, that we have a number of AI applications in here, including where we email the document for the signature, we got to detect for phishing. And for translation, we have a small language model. summarization is a medium language model, as is anomaly detection, and then the recommendations are going to come out of the large language model. So I want you to catch here that for some of these things, it we don’t need the large language model. And not only that, we don’t want the hallucination and the other problems that come with these models. So sometimes it’s better, sometimes less is more in this world. Yeah,

Randy Johnston  26:58

as a matter of fact, its performance then at that point, too, because one of the key differences between Chad GTP, 3.5, and forro. And we suspect four, five, and so on his speed, you run on chat, GPT, three, five, man, it’s fast, but it’s not as accurate for many situations. And this idea of accuracy on small and medium language or neural models is pretty stunning.

Brian F. Tankersley, CPA.CITP, CGMA  27:21

I will say that the you know, the seminal event on hallucination for me, I think I’ve mentioned this a couple of times, but was when you and I were in a in a thought leader meeting in New York City and Chet GPD told us that our friend Doug Sleater, was dead. So you know, it’s very much like the Paul is dead thing off the Beatles albums, I suppose. Now, our next thing to cover is going to be finance. And so we’re going to talk next about expenses. And so in here, we have a process, that’s basically we add a photo of an expense, we submit an expense report, and we get it approved. But when we add the photo, it’s going to go in, and we’re going to extract the text from the photo with a note with a narrower OCR model, we’re going to extract the information from it like merchants and vendors and currency with a small language model. And then we’re going to generate the inference with another small language model. Furthermore, we then go into submitting the expense report. And it’s going to use the context of the anomalies in the policy deviations with with a medium language model and a large language model, respectively, to analyze these things, and to identify the unusual things in here, and then it will go through and approve it. So what I want you to catch here is that in both of these cases, we just demonstrated and, by the way, I’d like to thank rescue for, for making these slides available to us because I thought this was the the aranea you and I both saw when we saw it that it was one of the better explanations of these models, and when you use which ones and so forth, that we’d seen. But again, in this case, the goal of this is for things to just work and users to not even know that they’re using AI. Furthermore, the private Furthermore, when we’re using things like small language models, because there’s so much more efficient to create, we can create a small language model on your organization for context. And we don’t have to create this big large language model that also has all these problems and potentially leak to others. So this way we have, you know, we can have a small language model on a single tenant as opposed to having it on a on a bunch of tenants. And that way we it makes the data privacy and the data leakage problems, much less acute.

Randy Johnston  29:40

Yeah. So Brian, you and I both learned a significant amount. It is a whole day this year, and not only in the products and oh my wish we could talk about the non disclosure things because I’m just itching to create other podcasts on those topics. Any parting thoughts on Zoho and our experience with Zoho product? And it’s Ohio days.

Brian F. Tankersley, CPA.CITP, CGMA  30:01

Well, and we’ll talk about this in a future accounting, accounting Technology Lab. But I think with the with the demise, impending demise of QuickBooks Desktop, I think that Zoho accounting or Zoho practice and Zoho books are both important things for you to consider. You’ve got in that finance suite, you’ve got an outstanding expense reporting tool, you’ve got everything you need to run a business, and with their focus on making things even more integrated than they already are. I think this that, you know, this integrated platform is really a is really a concept whose time has come. And so I think it’s a, I think it’s something that you should be you should be looking at, you should be thinking about. Now, you may need to get a consultant implemented or other things like this, but I think you should at least be kicking tires on some of the tools, Zoho books, so practice, Zoho expense, possibly Zoho Creator or Keszthely. They’re their tools for creating workflows and their BI tools. You know, again, you don’t have to dive in and take on every tool. But there are a lot of good ones, and you’re really missing out if you’re not trying any of them.

Randy Johnston  31:20

super well. Friends, we appreciate you being with us on this accounting Technology Lab. We’ll have more content for you in future podcasts.

31:33

Thank you for sharing your time with us. We’ll be back next Saturday with a new episode of the technology lab, from CPA practice advisor. Have a great week.

+ END +