Episode 4: Botkeeper and the "What Sucks?" Framework
In this episode of AI at Work, host Rob May has a conversation with Enrico Palmerino, CEO of Botkeeper. Tune in to learn more about Botkeeper and the "What Sucks?" Framework that they use to approach automation.
Rob May, CEO, Talla
Enrico Palmerino, CEO, Botkeeper
Rob May: Hello, and welcome to the AI at Work podcast. I'm Rob May. I'm your host for today. Our guest today is Enrico Palmerino, he is the co-founder and CEO of botkeeper, which is the future of bookkeeping. Why don't you tell us a little bit about what you guys do?
Enrico Palmerino: Thanks again for having me, Rob. We have created a robot bookkeeper, essentially. We use a combination of machine learning and some AI to automate a lot of the bookkeeping and accounting tasks that a traditional, full-charge bookkeeper would do. We deliver our clients real-time data, what's happening 24/7 around the clock, and more accuracy, 99.97% accurate. You don't see human error, at about half the cost of hiring a person or using an outsource firm. Pretty much better, faster, cheaper bookkeeping and accounting.
RM: You're a successful repeat entrepreneur, and your last business and some problems you had there are what led you to this business model, correct?
EP: Yeah, I actually had, two companies ago, started a lighting business that grew pretty quickly. Our bookkeeping wasn't able to keep up, and that led us to run out of cash in a month because we didn't know what our cash on hand was, due to a 45-day close. From that, having that pain point, I said, “I'm never going to let bookkeeping tank a company”. You know, we were $8.5 million in recurring revenue, some awesome margins. It didn't tank the company. We got back on our feet with a line of capital, but I was never going to let bookkeeping jeopardize a business again.
RM: That's interesting. When you started this company, did you approach it from the real-time perspective first, or from the AI perspective first? What was your thought process? Did you always envision it being this sort of robot bookkeeper, or was that an evolution?
EP: I think out of the gate we thought of it as, and actually the way we started marketing it was, “We're going to use quant, decision trees, algorithms to automate bookkeeping processes. Then, whatever we can't automate, we've got a skilled team of accountants there to assist and go the extra mile”. What we realized when we first started marketing it was that it was super confusing to the people that were buying bookkeeping services. They didn't know what AI or machine learning was. They didn't know how it was going to do what they were asking. It meant we had to spend a lot of time educating them on what and how the software enabled our accountants.
We realized that we should really just be saying, “We've got a great team of skilled accountants for you that are going to do all of your work. Oh, and by the way, it'll be cheaper because on the back end we automate a lot with software and tech.” We had to water it down a little bit. We saw it as less of “we're getting a robot bookkeeper” and more that we're augmenting the need for more accountants on our team.
RM: Gotcha. One of the things that we've seen is that a lot of the people listening to this are tech executives somewhere, and the last wave of enterprise software, which was the SAAS and cloud wave, was very different than the waves that came before it. People were used to buying and implementing software, and it took time, and you needed expertise. Now it’s become that you need a web browser and you sign up. While some packages require implementation and whatnot, a lot of it is super easy.
But, with AI, everything I've seen across my angel investments, here at Talla, and other companies that I know, is that these AI companies typically take a little more on-boarding, right? Training the software is sometimes like training an employee. Sometimes, the worst the software ever performs is at the moment of deployment, and just gets better and better over time because it gets smarter. What have you guys seen are some of the challenges to on-board customers, and how do your customers think about that? Have you had to sell them on it? Have you had to educate them? What's that process been like?
EP: We figured, typical bookkeeping deployments take months. If you were going to start using an outsource firm or hire someone, you’ve gotta train them. It's like a month worth of training, and a month worth of getting access to data and different systems to be able to do the bookkeeping. My initial thought was, “Hey, anything less than two months would be awesome, and if I could boil it down to on-boarding in a week, or in a day, or a matter of a few hours, that would be worlds different than your other options”.
What we've found is because we've created bookkeeping to look so much like a pure software service, and like you said, with SAAS, everyone expects that is it to be a snap of your fingers and you're up and running. We now have people that sign up, and when they realize they have to put maybe 3.66 or 3.62 hours into training and on-boarding Botkeeper, and that they're just like, “This is way too much, I'm not doing this” or “I don't have time” or “I'll wait to do this later.” And I'm like, “It's less than four hours of your time, and you'll have a solution that you never have to train again, and it's just going to be running and getting better forever forward.” But, for whatever reason, we have a decent number of people that sign up, and even though we show that it's going to take about two hours or so to go through to do the setup, plus the on-boarding call, they still are shocked when they have to put any effort or any energy into it.
It's frustrating, because the other options take more time, and you're comparing us against a plug-and-play, a simple click of a button where you do nothing. How do you expect a bookkeeper to learn about your business, know the role, the process, with one click? You’ve got to at least link up your bank accounts and credit cards. That takes all of five minutes, and yet that's like too much work for some people. It's frustrating.
RM: Well, it's one of the things that I think a lot of companies are going to have to start to understand about deploying AIs. If they want the benefits, they really are going to have to go through this sort of larger on-boarding process. We actually had an issue with Talla that I mentioned on one of the other podcasts, where we were trying to explain, this. There was a company that was sort of saying, “This thing performs well, but not quite as well as we hoped.” And I said, “Oh, you have to spend a couple minutes a week training it, and here's how you do that. Here's how it'll pop up and talk to you”. They were like, “Look, my reps are busy. I don't have time to do this”. I said, “Well, how long does it take to train a new rep?” The guy kind of said, “Oh, well, it takes about two weeks…” Then it hit him, “Oh, I see. You're asking my 20 people that are here now to spend five minutes a week, so that I don't have to bring on a new person and bog them down with two weeks of training that new human, which is very, very slow”.
Your business model is predicated on the fact that you have real human accountants who can deal with the higher-level stuff that a company needs. Your business model is constantly recording everything that happens, how people categorize transactions, interacting with multiple systems, etc., and figuring out how to use AI to automate more of that all the time. Do you have a framework for how you approach things and think about what to automate first? Is it a framework that listeners could take to their own businesses when they're looking at, “Hey, I've got all these areas of my company. Where should I look? What are the properties of something that is right for AI-driven automation?”
EP: We call it the "What Sucks?" framework, and literally, we survey our team at least weekly to ask them what sucks about what they're doing today. We're looking for a commonality. Are enough people saying this one component of bookkeeping sucks and they hate doing it? Especially for accountants to not want to do something, that's usually the first thing that we'd want to automate.
We'll take that and then we quantify. We track all the time that our accountants are actually putting in to see how much time they are spending on this thing that they think really sucks. If it's 10 minutes a month cumulatively across our company and there are other things that we see them spending hours and hours or days on, we're going to look at those two to understand what the investment is going to be on our end to automate the thing that sucks versus the thing that's consuming the most time.
More often than not, we still automate just the thing that sucks, even if it's a little savings, because you find if our accounts are doing more of the things that they love, that's also allowing them to be happier, have happier contact with our clients, maybe even work more hours. They'll go the extra mile for us. So, we've kind of taken that approach of, let's just try to make our accounting team as happy as they possibly can be, doing the work that they need. That should translate into us automating more over time.
RM: Is there a hierarchy to some of these things where there are things that you look at it and you say, “Well, we couldn't automate that until we automate this step before it”? Do you plan for that long term in the way that you think? Or, is it just a little more ad hoc in how you’re making these automation decisions?
EP: I think we're crossing the chasm on the long-term planning. We're seeing that when we first set out and we had fewer people on our dev team and fewer moving parts, we had a more ad hoc approach. We'd see something, we knew it was inefficient, we could quickly realize that if we built a little bit of software, we could learn or train something and then have it repeat and replicate whatever that process was, so literally combining machine learning with RPA and then kind of getting to this, call it quasi-AI, because I know AI takes, usually, one step farther if you really want to talk semantics. We are now realizing, to your point, that sometimes we're setting out to automate things, and there are much bigger projects.
On the dev roadmap we'll say, “All right, we want to automate this”. We start going about it and realizing that thing, whatever it is, is a much larger project and it's going to mean we either have to sidetrack the five other things we'd planned on automating this quarter and just focus on this one, or do a recalibration of, okay, how important is this? What is the cost going to be? What's the ROI on the build-out, and how many other clients do we foresee using this?
Then there is also the other side of the spectrum, which is, if we did have it, even if we're not using it a ton today, could we teach our sales team to sell more clients like this that we could use to automate? We do a bunch of dental practices, ironically enough, and there's things that we would automate, but we haven't focused on it so much just given that I don't think our sales team is in a position to do a whole lot of selling to dental practices. So, there are things that we see that we just turn away because the business side of the equation and how we're going to sell or market it isn't as sexy or doesn't make as much sense.
RM: You use a bunch of human in the loop processes where you have people that are looking at transactions, classifying them, doing certain things, and you're capturing what they do so you can automate that. Do you have a framework or measurement or objectives around how much you automate with that, how fast it improves, and then do you see it anecdotally? Are you automating things faster, now that you've been around for a couple of years, or has it started to slow down?
EP: I think we're automating things faster now that we've been around for a couple of years. It comes down to the fact that we've started to build a really good dev team that works well together, and so, whereas we had a lot of inefficiency around people either not communicating enough to each other, and were building things that aren't connecting the way they should, or leaving out certain variables that we should have been considering, database schema changes. We tried to build a hard-coded database schema on PostgreSQL and then realized we really need to focus more on Mongo because Mongo allows us a lot more flexibility, and our business is changing and shifting so much. Let's just assume there's going to be constant changes in the future, so there's this build now or plan for the future approach.
Thing that we don't really track is, how is that feedback loop either being reduced, on a per client basis or in general, versus, our costs for that feedback loop and how much revenue we're generating, and are those costs going up in proportionate to revenue? We know for a fact that they're not, so the costs on that feedback loop are staying roughly the same, and yet revenue is climbing pretty significantly. Possibly it's even coming down a little bit, we're seeing in some of the more recent months because we just found we didn't need as many people, we repurposed some people to other roles. I think we look at it on a kind of revenue per dollar spent basis. I'm sure we could look at it on a more time basis if we wanted to, but there's a lot of variables. One client could totally throw off a model if they're a very large client, and we have a wide range of client size.
RM: When you look at big companies and how they're thinking about AI, a lot of big companies, what they do with any new technology is they take a “wait and see” approach. Let's let other people adopt it, let's let them work out the bugs and the kinks, and then we'll adopt it when it's ready.
Do you feel like that's a valid strategy with AI or do you feel like it's different enough in the tools, the flywheel from data learning, the workflow behavior change? Companies that don't adopt it sooner rather than later, are they smarter and going to be better off and have better products at the end, or are they going to be behind the eight ball and not up to speed and get beaten by their peers?
EP: I think there's a window for when you should start adopting a tool. I don't think a company can build enough of a product in a year that you know this thing to either be around or actually have benefit. It's one thing to say, “hey, we're building an automated bookkeeper and we hope by next year this thing…” If you came on early on and were one of the beta clients, knowing you're a beta client, great. If you're that early adopter, get in, lower pricing, etc., other benefits, but, knowing that it's going to take time and it may or may not work.
I think by the time an AI company has established itself for two years or so, has some client base that's been there for maybe 6 plus months, that gives you a really good sense of whether or not this thing is working. The cool thing with most AI companies is that where the software falls short, we all have people to pick it up or take care of it.
It is not a SAS tool that works or doesn't work, it's like you said, a flywheel of continuous improvement, and there's always a helping hand there to pick up where it breaks. I think you have lower risk, really.
RM: What's your general advice for, say I'm a VP of marketing or IT or accounting or whatever. I've worked at an insurance company, we haven't adopted any AI yet, I want to get smarter about it. I want to start thinking about whether or not it's the time. Do you have any good ideas for, here's how you should educate yourself, here's what you should look at first, here's some mental frameworks and models that you might have that you might think about? What's your advice if you're having a beer with somebody in that situation for how they should approach it?
EP: I think there's a lot of good material out there. You can certainly crawl YouTube. I'm sure there's some TED talks, there's a bunch of great events or groups. Like yourself, actually holds an AI meetup here in Boston, which I think is really great. So those conversations, you get to quickly learn and understand what other people are doing in the space, how they're approaching it, what kind of software and tools, what efficiency they're getting.
I would say that you're better off trying to adapt or bring in products or tools that other companies have built than going about starting your own internal AI division or department and setting after it. You're going to spend a lot of money that way trying to figure out where and how you can optimize or streamline. It's much easier for you to look at another product, let them do an ROI analysis on how much time you're spending on something and what they're going to reduce the cost to.
You're betting that they are not lying to you, but I think most businesses wouldn't be in business or have any longevity on clients if they were. You can search AI across any department and you can find a whole lot of results.
RM: Now, going back to the AI at Botkeeper and the automation that you guys are doing, you've recently changed your pricing because you've been able to become so efficient at this that you've been able to offer customers even a better deal. Tell us what it was before and how you guys have restructured that.
EP: We used to have pricing that started at $235 a month and then you could do a 110%, 120% increase on transactions, and it would basically double. Then you could do another 100% increase on transactions and it would increase only 50%.
The idea is that basically as you grow and get bigger, the cost per transaction comes down, which plays into automating more and more, greater efficiency. Our goal was to eventually get to the point where we could offer free bookkeeping. The idea was, let's be the thing that we say we are. If we truly are the leader at automating bookkeeping, we should eventually be able to build a product that is purely automated, has no cost or minimal cost to set up.
At that basis, we turned what was a professional service industry for the last 50, 60, 80 years, and created a freemium product. Starting July 4th, we're rolling out our free version of Botkeeper. If you're a business under 20 transactions a month, so certainly a small business. Think about it, if you can pay five or 10 bills a month and invoice 15 clients and be in that threshold, your bookkeeping will be free. No duration limit, nothing. As long as you're under that threshold of transactions, free bookkeeping for you.
If you need to increase up to 50 transactions a month, the cost only goes up to $99 a month. Our goal was to really say, “Hey, look, a lot of people claim to automate a lot, the proof is in the pudding if we can deliver a free solution, we get to put our stake in the sand that says we're certainly automating the most”.
RM: Even though you can offer that with for these really small companies, one of the things that I've noticed over the last 18 months is you guys, while still having a good SMB customer base, it's started to move up market to some pretty big companies where you're really seeing some big gains in the work that you're automating away. Is that something you see continuing for Botkeeper?
EP: I think we'll continue to move up market in terms of the client that we capture, but I don't see us changing our focus. We're going to keep focusing down market. The idea is if we can be your first bookkeeper, when you're starting a company, that's when you need to get a sense of your bookkeeping the most, and you don't know what you're doing.
If we give good financials to companies earlier on, that's going to help with the business model, with potential of raises. There's also a lot more of them, so they'll learn about Botkeeper, our product. They'll be on the platform, they'll grow with us. They shouldn't outgrow us ever. That’s what we've been able to prove with our current client size.
The reason we won't focus on selling to an enterprise client is that there's a human variable there, emotion, that we can't plan for. We can tell you better, faster, cheaper, half the cost, more accurate, quicker reporting, faster delivery, but we can't plan for that you have a brother, sister, or cousin on that bookkeeping team, or that you feel bad about letting that bookkeeping team go, or not having a place to repurpose them. We found that some companies have come to us where they either lost bookkeepers, they're large companies, or they're really looking to improve accuracy, get more reporting, cut costs.
If you come to us, we've got a really high win ratio, and it's a short sale cycle. If we go to you, it's a lot of time convincing you that it's going to work and then a much longer time with you mulling over letting people go. That's the other negative connotation that we have in the AI space is we're putting everyone out of jobs.
RM: Yeah, it's interesting. I was having a discussion with somebody over drinks the other night about that very topic. I think one of the things that's going to happen, when you look at other things like clothes and furniture and other areas that used to be done by humans and then went to manufacturing, it's like now there's an allure again to having something that's handcrafted and handmade.
I don't know if there will be a day where it's a status symbol to have a human accountant still, but, for a lot of the service industry jobs that are going to put people out of work, I think there will be a status symbol with having a human do your tasks. Being able to afford that because AI will probably make everything so cheap.
EP: I think in our space what we're trying to really position as enabling accountants. We let them do a lot more in less time. I think, yes, there are bookkeepers and companies that won't get jobs because Botkeeper is there, or, some larger companies might get rid of a couple people and replace them with Botkeeper. The most that we've done is up to eight people that Botkeeper's replaced.
I think more often accountants and accounting firms think of Botkeeper as an enabler. We do the books for other bookkeeping firms, which is the most ironic statement. They bring us in and we then do all the blocking and tackling, the data entry, the stuff that they don't want to do. Then they can spend more time doing advisory-level services and work and actually tell you what the numbers mean.
They can spend 80% of their time looking at the data and telling you how that's going to change or impact your business, versus 95% of the time getting the data in and then 5% looking at it. That's the numbers that we hear in the industry. I think if anything, we're going to replace some people. I think we're augmenting a need for a market that is limited supply and has a lot more demand for bookkeepers. Just look at the cost of a bookkeeper, $50-$75 an hour for someone to do data entry is asinine.
RM: Good. And then one of the questions that I like to end on with all of our guests, and we talked about this a little bit before the show, but Elon Musk and Mark Zuckerberg have this big debate about where AI is going. You have Musk on the AI is going to kill us side, we need to do something today. You have Zuck saying, this is either really far out or we never need to worry about it because even if we do create a general artificial intelligence, there's no reason to think it's out for us.
Working in the industry and seeing that maybe this stuff just doesn't move as fast as we thought with creating generally AI and everything else. A lot of people don't realize how hard it is. Where are you on that Musk versus Zuckerberg spectrum?
EP: I think I'm more aligned with Zuckerberg, just because I look at what we're building and how much effort and time it takes. I think with AI are still getting there. Focusing in on super nichey things, you can get the models to start working well. They're not perfect.
It’s hard to imagine that someone is going to be able to compile enough models that work in hand with each other. Even if you went on an acquisition spree of every AI company and try to mishmash it into one general AI thing, I don't think the code would talk or the models would assimilate or integrate well.
There's so many things. You can train the model and it starts working really well, and then more data comes on or too much data and the model starts regressing and retraining itself in a way that is reducing efficiency or reducing accuracy. You've got to have human labor and time there, constantly teaching it. Maybe one day it will get to a point where it can keep teaching itself better and knows what those thresholds are, but I don't see that anytime soon.
RM: Good. If you're interested in learning more about Botkeeper, you can go the website, botkeeper.com. Enrico Palmerino, thanks for being on the show.
EP: Thank you, Rob.
Subscribe and share AI at Work with your friends! If you have feedback or questions, we'd love to hear from you at email@example.com or tweet at us @talllainc.