How TaskRabbit Solves for Endless Human Variables

0:00When someone comes to my door, I know, because I'm using SimpliSafe. I get an alert straight to my phone and I can see, is this someone I should unlock the door for because I know them? Or is it someone I need to tell to get lost? But here's the thing I love most about SimpliSafe. It works when I am not looking at my phone too, because when I'm not keeping an eye out, SimpliSafe is. SimpliSafe is proactive with a double layer of defense that stops crimes before they start. First, AI-powered cameras spot potential threats outside. And then live agents step in, talking to the person through the camera and letting them know they're on video and police will be dispatched if they don't leave.

0:45They can even trigger a loud siren or spotlight. That's how you help stop a crime before it starts. And SimpliSafe makes it easy to use. There are no long-term contracts, no hidden fees. You can cancel at any time. It was named Best Home Security Systems by U.S. News and World Report for five years running. It offers a 60-day money-back guarantee, so you can try it and see the difference for yourself. This month only, take 50 % off any new system. This is one of the best prices you will ever see for SimpliSafe, so don't miss it. Just go to simplisafe.com slash problem solvers. Again, that is SimpliSafe, S-I-M-P-L-I-S-A-F-E, simplisafe.com slash Problem Solvers and lock in your discount.

1:33There's no safe like SimpliSafe. From Entrepreneur Media, this is Problem Solvers, a show that solves entrepreneurs' toughest problems. I'm Jason Pfeiffer, Editor-in-Chief of Entrepreneur Magazine, and each week I take a problem that you're probably dealing with and get tactical solutions from people who have been there and solved that. So let's get solving. I was recently talking to the CEO of TaskRabbit and I asked her, what is the hardest thing about running a marketplace business? You know, marketplace, which is to say that TaskRabbit, like many marketplace businesses, really just sits in between two sets of people.

2:15There are people who are providing a service and there are people who are requesting a service. So Uber is a marketplace. You've got people who want rides and people who provide rides. eBay is a marketplace. You've got people who are selling things and people who are buying things. TaskRabbit is a marketplace. You have people who need tasks done, and you have people who do those tasks. So what is the hardest part of running a marketplace? And the CEO told me something that was so simple that it almost sounds comical, but also quite profound. The fact that humans are involved makes it really interesting, but also super challenging.

2:57Humans. Humans. Because if you are running a marketplace business, then you are connecting humans to humans. Humans who can be unpredictable. Humans who don't always exactly deliver or articulate in the way that they should. Humans that are complicated. Boy, it would be really great to run a marketplace business connecting robots to robots. Super easy. But that's not what we do. We connect humans to humans who are complicated. So let me introduce you to this one human, the CEO of TaskRabbit. Her name is Anya. Hi, my name is Anya Smith, and I work on helping taskers find work and on helping clients make lives easier for them.

3:40And if you want to understand marketplaces, there is truly nobody better to ask than Anya. So actually, I've been working in marketplaces now for, I guess, 15 years, over 15 years. It's horrifying when you have to put a number to it, isn't it? Exactly. And honestly, my first sort of, quote unquote, real job after I got out of management consulting was at Expedia, which is sort of one of these OG marketplaces, really. It didn't involve humans or like the gig economy part of things, but it's where I sort of started to learn the beginnings of how a marketplace works. And then, you know, I then went and did a stint at Walmart, but really started digging deeper into marketplaces and what they are when I started Airbnb when I was still a fairly young company.

4:32And then after that, I spent some time at Uber and then became the CEO of TaskRabbit. And Anya told me that people often underestimate the complexity of a marketplace business and just how sophisticated you have to be to truly grow one. And so I said to her, well, what's the primary thing that people aren't really thinking about? And she said the answer is variables, endless variables. So there's all sorts of things that come up, you know, it's like someone shows up and they need help with putting up a fan, but they didn't mention that they have 20 foot ceilings. And so now, yes, this tasker has a ladder, but that ladder is not tall enough.

5:17And so these things happen all the time. We have done millions of tasks. In fact, I just saw that we have completed over 30 million tasks on a platform, which is a lot of tasks. And so we've learned a lot. And there's still a lot of learning to be had. And on this episode of Problem Solvers, we are going to extract some of that learning and really focus in on how does a company build with all those different variables, endless variables. That little thing that Anya just said a minute ago about someone who hires a tasker to install something on a ceiling and that ceiling is too tall for the ladder that the tasker showed up with.

6:01It is so simple and yet it is so complex. How do you solve for something like that at scale? It is the reason that we will, throughout the course of this episode, continue to return over and over to that very specific problem. This is a strategic challenge for any kind of business to be thinking about how do you scale while also taking into account all the infinitely expanding variables that come with growth, especially if you're serving humans and even more so if you're serving two different sides of humans. So how do you solve the problem of human variety in business, especially a marketplace business?

6:48That's what's coming up with Anya Smith of TaskRabbit right after the break. All right, we're back. I am talking with Anya Smith, the CEO of TaskRabbit, about the complexities of building and growing a marketplace business. And just before the break, as you'll remember, Anya had to use this example of the ladder, the task that requires reaching up to the ceiling but showing up with too short a ladder because, you know, the height of the ceiling wasn't communicated in the listing that the tasker showed up to perform the job on. How do you solve for that? And how do you solve for that at scale? How do you start to tackle the endless variety, the endless variables?

7:30Well, that's what I started by asking her. So much of it has to do with data. And today with AI, we have a lot more help, but it is about gathering all the information. So in this example, you know, the system should know that when someone asks for a ceiling fan, one of the first questions that the system needs to ask is, well, how high is your ceiling? And then we also need to know that taskers have ladders. First of all, do they have a ladder? Yes or no? If yes, how tall? And then the match becomes much easier. So the sort of secret sauce is being able to produce the best match for the job, regardless of the number of variables.

8:15You don't want to ask infinite numbers of questions because, you know, then that process is not easy, but you need to know which ones are the ones that are most likely to cause a job to go wrong and make sure that you have that information. And all of it needs to be automated and smart. And so with the use of AI, we can't get that much smarter in really producing the best match that we can between a client and a tasker. Tell me about the difference between pre and post AI then, because TaskRabbit is a company that predates AI. So what kind of problems was a pre AI TaskRabbit running into that are now solved in different ways.

8:55So it's honestly really hard to think about sort of pre and post because it wasn't for us like one day this happened. We have been using language models forever, for a long time, really synthesizing data and trying to optimize the match. Right. I guess I should just clarify that I'm thinking the popular boom of generative AI, not necessarily the availability of the AI technology. So as those large language models got much better, we can leverage all sorts of ways so we can use chatbots to really gather information in a much faster way than manually gathering information so that we can be smarter and ask the right questions versus just the same standard questions for every task.

9:40and then we can really ensure that the data that we're getting is being used in a smarter way to find the right tasker for the job. So that means we maybe don't need as many taskers in a given market. We need more taskers that are available on a Friday night to do sort of last minute tasks like your toilet being plugged in your apartment in Brooklyn and all of a sudden you need someone to help you and you need that person quite immediately. It's not a task that you can wait four days for. With AI, we can better understand what the gaps are and who are the taskers that we need to acquire or make sure that we have on a platform to solve those types of problems.

10:24What does this look like internally for you? And what I mean by that is I'm thinking as you're talking, you know, not everyone who's listening to this is running a large marketplace like TaskRabbit, but everyone in some way or another is trying to ingest every variable of something that is going wrong in their business and then trying to learn from it and build it into the systems to optimize. And so I'm curious what this looks like internally. How do you structure your teams to be ingesting and responding to this information? How are you gathering the information in the first place? What does this look like?

11:03So it's a great question. And we certainly don't claim to have it right. But I think at the core of it for us is data. And again, having clean data, having data that is usable, having data that you can easily manipulate, Having data that everyone has access to so that you don't have to sort of put together a ticket and then, you know, get some chart back two days later that you can quickly be able to figure out how to answer the question that you have. So it's more self-serve. So actually having agents on top that can do that for you, but all of that relies back on the quality of the data. So we're doing a lot of investments there so that we can make sure that when the data comes out, it's actually right and you can make clear decisions based on what it says.

11:54What does that look like then? And maybe just to keep it human as we're talking abstractly about data, just go back to your example of the ceiling is too high and the tasker doesn't have the ladder that's tall enough. So first of all, how would something like that get reported into the system? How would TaskRabbit know that that specific problem happened? And then how are you structuring it in a way in which that turns into usable data? So in this instance, usually either the tasker will try to solve it, right? So they may have a friend who has a ladder or something like this, and we may never learn.

12:33But usually what ends up happening is either the client or the tasker contacts us and says, hey, we can't do this because this is the problem. And they're contacting you on some form inside of... No, they'll contact our customer service team and say, hey, this isn't working. and the task has to be canceled. Or they may just cancel the task and then not give us a reason. So then we don't know. We just see that the task has been canceled. So do we follow up and understand that? And then how do we log that data? It's better if they actually contact us in a way that gives us a reason because then we can actually store that data and understand, well, how big is that problem?

13:15Is this one-off problem that happens, you know, once a year in this city? then it's probably not something we're going to dive into because there's, you know, thousands of other problems that we have to solve. Or is this a recurring thing? And then is it similar to other problems? Can we build a product that actually corresponds to the issue so that no one has to contact us so that we can get ahead of the problem prior to this? Right. So in this case, we would be asking the right questions ahead of time. And we would have that data. So we would have never sent a tasker who indicated that they only have an eight foot ladder to a house where the ceiling was 20 feet high.

13:57OK, so if you're if this is coming in through customer service, the customer service has some kind of system by which they're flagging this kind of very particular sort of problem. Ideally, you don't have to even call the customer service. You can say I'm canceling and this is the reason. And the reason can be specific enough that we can actually do a search and just pick up keywords and start understanding what the major issue was. And then we can add it to all of the other issues and start grouping and understanding, is this a small problem or a big problem? How can we solve it? I'm enjoying thinking about this because you said the solution casually when you first introduced the example, which is that if someone is going to hire a tasker to hang a ceiling fan, then you need to ask how tall is the ceiling.

14:53But in actuality, I would imagine that getting from that problem to that solution is not that simple, right? first of all, enough of it has to have happened. It has to be recognized as a pattern inside of your system. And then there has to be some solution built inside of a hyper-specific flow, which is that somebody's now asking for this specific kind of task. You see where I'm going. I'm curious how on earth the million varieties of this gets managed. But this is where AI can be really helpful. So you can understand that the chatbot can ask if you say a ceiling fan, the biggest question is how high?

15:37If you say a TV, well, the biggest question is how big is the TV? Because if it's too big, I need two taskers. And so this isn't going to work because one tasker can't lift it. Or what type of a wall? If it's a brick wall, that's a very different wall than a drywall. And so you can sort of go down these paths and the systems are quite smart now to ask those types of questions. We just have to know how to then harness that data to then match it with the other side. Because at the end of the day, it is all about optimizing the match and ensuring that you have the right level and number of taskers with the right skills at the right price at the right time in the right location for the problem that a client may be having.

16:22So how are you solving for, because that's a bunch of those things that you're describing are things that are outside of the system that we're talking about. Now you have to create some kind of incentives or structures to have the right people available at the right time. I don't even know how to begin to guess what it is that you're doing. So what are you doing? One of the keys to a marketplace is to make sure to look at a market or at a specific area. We operate in sort of liquidity state where we need clients and taskers to sort of mesh in this one area. Oftentimes, people kind of don't get this right and they look at totals and averages.

17:05And for a marketplace, that really doesn't work because in total, you may say in the U.S., we have X amount of taskers and X amount of clients. That seems about right. We're good. But it really is just not the right way to look at it, because if all of your taskers are based in New York City, and in fact, you have not enough taskers in Seattle, now you're spending marketing money in order to acquire more clients in Seattle because you're seeing that that seems that we don't have enough clients. but then they actually come to your app and they're not seeing any taskers who can fulfill those jobs.

17:46So you have to be looking at it. Yeah, then you're chasing these pockets of supply and demand imbalance all over the place and it's madness. All the time. And so you have to look at it market by market, category by category, and constantly understanding how much higher does one side versus the other need to go so that there aren't these gaps because they produce very poor experiences for both the taskers and the clients. As I mentioned, you go on, you saw this ad that said we can do all these things and you go on and there's no one to do those things. You're not likely going to come back, yet we've spent a lot of money in acquiring a client.

18:25What we don't want to do is continue to onboard taskers. It's really expensive to acquire taskers and to, you know, have them go through a background check and so on. And we want to make sure that we're bringing the taskers with the right skills, right? So maybe we have enough taskers in Seattle who can assemble furniture, but not enough taskers who can help you with your yard. And so you have to understand the nuances and categories as well. And so it is very data heavy type of a system. And we call this marketplace dynamics, constantly trying to balance across markets and categories. So does that mean that you are constantly adjusting your recruiting techniques?

19:06And what does that look like? That gets expressed in advertising and other modes that you're using to acquire taskers, and you're just being hyper-specific about the kinds of things that you're looking for? That's right. So we're being specific and understanding what types of skills they bring onto the platform in that specific market, given what kind of gaps we have. Fortunately, for the last several years, we have had way more taskers in some sense than likely that we need. And many markets have wait lists because, again, we don't want to bring on taskers who then are going to be unable to get any jobs.

19:43It's not a great experience for them. We want to make sure that if you're on our platform, you have the work that you're expecting to have. And so we go market by market, but we don't always have to acquire new taskers or onboard new taskers. We can also incentivize current taskers to become more available, right, or to jump into new categories to learn a new skill. And that also works. So there's a fine balance of trying to see, you know, how optimized is a current supply? How much new supply do we need? again, this comes back to, do they have the right skills? We haven't even talked about pricing, right?

20:23They may have the right skills, but they price so high that they're still not getting hired. And so all of these factors go into balancing of the marketplace. You know, this is really fun to hear you talk about this because I had just never thought about how this kind of marketplace is a constant micro-balancing act across almost infinite cross-sections. The first time I'll tell you that I had wrapped my head around what the challenges of a marketplace like this are, and this goes back to what you had said just a minute ago about how if you measure the wrong thing in the marketplace, then you might not actually have a healthy marketplace was when I was talking to Andrew Chen, who's a partner at A16Z and investor in Uber.

21:12He had written this book called The Cold Start Problem, which is about network effects, which TaskRabbit is a network effect business, which is to say that it becomes more valuable the more people use it, which isn't true for yogurt. It doesn't matter how many people are eating yogurt. It's still just as delicious. But TaskRabbit is not all that useful if there aren't enough people using it. So he pointed out how Tinder, when it started, did not try to just onboard the entire country onto Tinder because you could say we have people in all 50 states, but that's not all that useful. It's totally that useful, right?

21:50You think they started at a university in Texas, right? That's exactly right. They started in California, but yeah. And this is how TaskRabbit started as well. Leah started in Boston. And you have to work at a small market and then try to understand what do customers need? What are their biggest problems? And can you find solutions for them with the right supply or the right taskers? It is a really challenging problem. We're now expanding to new markets. And, you know, we've done this many times. So we have a playbook. But, you know, we don't go into a market and say, hey, look, everyone, TaskRabbit is here, because it will really actually turn off customers.

22:32We first sort of soft launch and really focus on acquiring taskers and trying to figure out what do we think the right balance will be for the market and how many do we need before we really start focusing on acquiring clients. And then, you know, and then we're off to the races, really trying to build both sides. But we definitely start on the supply side. Oh, that is so interesting. What's a new market that you just recently entered? I will say this because I've been sort of, it's near and dear. It's my hometown, Sioux Falls, South Dakota. So it's a much smaller. And, you know, I think there are probably a population of about 200 ,000 people now.

23:14And so we haven't really been loud about it because we're in a process of acquiring taskers before we announce it and do some work to acquire clients. Yeah, I know Sioux Falls because I grew up in Florida and I'm a Miami Heat fan and the Miami Heat's G League team, which is their kind of affiliate team, is in Sioux Falls, the Sioux Falls Sky Force. So that's how I know them. Oh, Sky Force is a great team. I mean, I don't follow it anymore, but yeah, they've been around for quite some time now. They've been great. And some Sky Force players have become great Miami Heat players. But OK, so this is really interesting.

23:47So when you're entering, let's say Sioux Falls, right, you don't want to announce to consumers that TaskRabbit is there. because then they're going to go on to TaskRabbit and they're not going to find enough taskers. And then what is the point? And so instead you have to, I'm going to guess what you're doing is you're looking at the demographics of Sioux Falls and you're kind of mapping that on top of other markets. And you're saying, okay, based on Sioux Falls, we can anticipate that this is the kind of demand that will be there. And these are therefore the taskers that we want. And so you'll try to onboard those, then you'll go to the market.

24:17And then after that, you just start in one more market, this crazy, endless game of data gathering and rebalancing. Correct. We then really start focusing on acquiring both sides, but not leaving a strong gap in between. But, you know, one other thing that is important to mention is that we can't sort of for months on end acquire taskers because they know that we're not yet fully launched, but they only have so much patience as well. They also have lives. They have things that they need to do. And so, you know, we have to do this quite quickly in order to sort of start gaining the momentum and get more and more folks on so that when the clients come on, we're, you know, operating this sort of mini marketplace to start with, but are really starting to focus on marketplace dynamics and growing both sides and seeing which categories are most important.

25:14A lot of it has to do with weather. So, of course, you know, in Sioux Falls right now, it's really hot. So we know, for example, that there's going to be a lot of AC installations. And that's a good thing. We know we've learned a lot. And so we understand sort of which categories we should really lean in on based on, like you said, the type of a city, the location of a city, the weather patterns, and so on. Right. Okay. So final question, and very interested in what you say, I don't know what you're going to say, which is, imagine someone listening to this right now, and they have been thinking about launching some kind of marketplace.

25:49And they're now hearing about all of this balancing and data gathering that needs to take place. How would you advise that an early marketplace company structures its staff? What does it need to be able to be ingesting the data, reacting to the data, constantly shifting the balance of how it is creating supply and demand on both sides of this? TaskRabbit is at this point a well-oiled machine and it's big, but how would you start the basics of this kind of staff build? I think there's few things, but one of the key things is really early on investing in clean data. So you need strong data scientists, data analysts, machine learning engineers to really help you think about the models that are really running in the background that are helping to really provide the best experience for both your clients and, you know, the other side of the marketplace, whoever that may be.

26:46And it's really important that someone is, especially having some marketplace experience, trying to hire people who have had some of this experience. There are now many marketplaces and this sort of gig economy space, if it happened to be in this space, it's been around for 15 years now. So there's been a lot of lessons and a lot of learnings. And we're much further along in understanding what marketplace dynamics mean. And obviously 2008, 10, 11, you know, we do things where we understand now. For example, if we do something nicer for clients and allow them to cancel jobs, you know, five hours before the job, they like that.

27:34But that really makes it challenging for our taskers. And so you have to be able to think of the second and third order impact across both sides of the marketplace. And that gets challenging. So having that data and clean data and understanding that I think is really important. Well, now I can really appreciate, given the complexities of this business, why someone like you, who gets to understand marketplaces really well, then stays in marketplaces, becomes in demand. Fascinating. It's really fun. And, you know, I spend a lot of time with clients and taskers and I still learn so much. They share stories or things that they like and don't like.

28:15And even after this many years, I'm always like, wow, I've never heard that before. I didn't realize that's actually the pattern I should be looking for. It's really important to spend time, obviously, with your customers as much as possible. Well, Anja Smith, this was so fun. Thanks for chatting with me. Thank you so much. It was fun, Jason.