This post was originally published in May 2022. It was updated with podcast links in April 2023 when this interview was also published as a podcast episode.
2013. It’s Cedric Dussud’s first week on the job as the new engineering director at WeWork.
WeWork hasn’t yet become the beast of a company that changed the corporate world. It’s only managing a half-dozen offices in New York at the time.
But it’s growing fast.
That first week, Cedric got called into CEO Adam Neumann’s office.
“So it’s him, and our head of our department, and me,” Cedric told me. “And Adam basically says, ‘Hey, I need you to build us a billing system. And I need it in three months.’”
So that’s what Cedric and his team did. And his main responsibility during the following five years was to scale WeWork’s billing system as the company grew from eight locations to 400.
In the process, Cedric learned a thing or two about managing data.
In fact, Cedric and Ahmed Elsamadisi, the senior data engineer at WeWork at the time, learned so much about what it took to scale a data operation that they devised a whole new system to make it easier.
Together, they founded Narrator, a data platform that makes it easier for companies to run business intelligence without a large data team.
Cedric sat down with me recently to explain the key lessons he learned about data at WeWork — and how those lessons became the foundation that powers Narrator today.
There were a lot of lessons that stood out to me, including:
- Why metrics often don’t add up.
- What happens when there’s not a single source of truth.
- Why you should be skeptical about internal data.
- When you need a data scientist to help you (and when you don’t).
You can watch the full interview or read a few of my main takeaways from our chat below.
Full Interview: Audio Only
Full Interview: Video
5 Lessons From Cedric’s Experience
I learned a lot listening to Cedric’s story. Here are five of my biggest takeaways from the interview:
1. Data Formulas Are Subtly Unique
Data science has a key difference from software engineering.
In software engineering, when you work on an existing system, you can build on what past coders have done before you.
But with data, any time you create something new, you can’t just copy the code — and because of that, the definitions often have subtle variations across reports.
Cedric used the example of a CEO asking for a quarterly sales report broken down by region.
“You generally have to start over and write new SQL for it,” Cedric explained.
“You kind of hope that if someone else made a quarterly report for sales, that you can then figure out how to reuse that to break it down by region, and maybe copy their logic. But if you don’t see their logic, you’ll make your own, and then your logic is going to be subtly different.”
When you’re using these reports to run your business, small differences add up quickly.
“You end up with just dozens of subtly different things across the company,” Cedric told us. “And it is actually a fairly common thing for the CEO to come in and say, ‘Why are these numbers not matching? What’s going on?’”
2. There Is No Single Source of Truth
Platforms with built-in analytics can tell you a lot, but they can’t tell you anything about data flowing through other systems in your business.
Cedric used the example of online ads.
A platform will give you reports on the number of ads run, the cost, etc. The platform might even try to estimate how many people saw an ad before making a sale. But if you’re running ads on multiple platforms, then there’s not a single platform with all the necessary information to give you a complete answer, because they don’t have access to each other.
“If you were to sum up all the orders that Facebook thought it was responsible for, plus all the ones Google thought it was responsible for, and so on, it would sum up to far above 100% of your total orders,” Cedric explained. “That’s a case where you can’t rely on those numbers. You have to realize the source of truth actually doesn’t exist in any of them. It actually exists with all them combined.”
3. Be Skeptical About Your Metrics
Cedric recommends that you not trust a platform’s analytics unless you know exactly where the numbers are coming from and how they’re calculated.
In other words, you need to dig deep enough that you fully understand and trust the numbers you’re working with as you make business decisions.
“It is always worse to overdo data than it is to underdo it,” he explained. “If you’re thinking, I’m not doing it right, you’re going to engage your brain a lot and really second guess what you’re up to; whereas if you’ve gotten some fancy system that answers every question for you, you’ll be like, Cool, I’m doing data.”
4. Bring in a Data Analyst When Your Data Is Spread Across Multiple Systems
You don’t need a huge team of analysts in the beginning. In fact, getting too deep into your data can be a distraction:
“When you’re a small company, and you’ve got your mind wrapped around how everything is going, you usually understand your metrics,” he explained. And in that situation, you don’t need an analyst — not yet, anyway.
But when your tech stack starts to grow, that’s when the numbers start to become untrustworthy for founders and CEOs.
“When you get to a point where data starts to get spread across multiple places, that’s when you start to lose your ability to understand what’s going on.”
That’s when it’s time to hire an analyst or build out a more in-depth data strategy — someone who can help you fully trust the numbers again.
5. Focus Efforts on Projects That Will Attract New Customers
Even though Cedric built the billing system that runs a global business like WeWork, he does not recommend that SaaS founders should do the same.
It took months of full-time work for Cedric to bring a first version of WeWork’s billing system into production, an effort that — in his opinion — is not worth the investment for small SaaS and software companies.
“Work on marketing; work on understanding the customer and on new features — things that will bring you new customers,” he explained. “Don’t work on building your own billing system.”
Note: In addition to out-of-the-box dashboards, FastSpring maintains robust API and webhooks documentation so our customers can dig into their own financial data. Want to know more about what all our platform can do? Set up a demo or try it out for yourself.
Full Transcript From the Interview
All right, Cedric, thanks for taking a couple of minutes today. Go ahead and introduce yourself.
Hi, Cedric Dussud and I am currently a cofounder at Narrator.ai. I used to be an engineering director at WeWork. And before that, I worked at Microsoft as a senior software engineer. And I built WeWork’s primary billing payments invoicing management system over a number of years. And at Narrator, you know, I do all kinds of stuff. We’re a data platform company that that thinks that data can be done in a fundamentally better way. So happy to talk about anything related to my experience. I’m really excited to be here.
Cool. Very cool. Well, thanks. Thanks for sharing your story I have I have so much I want to ask you, because I was reading up, you know, on the notes you sent me, let me start with this. So. So tell me more about, like, the billing platform that WeWork like, what was the need?
Yeah. Okay. So back in this is going to be what, nine years ago, back in 2013, I joined WeWork. And the first thing my first week at WeWork, I get pulled into Adam Newman’s office. So it’s him and our head of our department and me. And he basically goes, Hey, I need you to build build us a billing system. And I need it in three months. So that was the gist of that entire conversation. So like, why why did he ask that? At the time, the company was, it was much smaller, it was eight buildings total all in New York. So as you can imagine, a little bit different than they are now and at their height. And at the time, they didn’t have a system to do this. And I was unaware. And I think the rest of the company was unaware of any other systems that existed. I don’t know, if you guys existed, then I’m pretty sure you didn’t, you know, Stripe billing did not exist. And so we sort of realized, oh, shoot, we better build something at the time, the business was effectively run on spreadsheets. So you know, the community managers, that was the name of the folks that ran a building and responsible for it, they would have a spreadsheet that said, here’s the person, here’s the office, here’s how much they owe, here’s when they moved in. And here’s when they’re planning on moving out. If so, right, it’s blank, if not, and every month, they would hand write invoices, and right, like literally go through the spreadsheet and type in invoices by hand, and send them out. And it got to the point where, you know, they were spending the entire day before the end of the month, writing up these invoices and double checking them, right, because this is something you can’t you can’t mess up. Obviously, this doesn’t scale. But at the time, the company had grown so quickly that you know, they just hadn’t had time to put anything in place. So the first job there WeWork was basically was the thing I worked on, the whole time I was there turned out was to build a building system on top of like, replace this particular thing that I had going and give them something that could work in scale. I know I’m getting along here. So I can I can talk about what I did after that in a second if you’re interested.
Sure. That the it’s a whole back inside to write they send the invoices, and then they gotta like accept credit cards or checks or like the whole, the whole back into processing all that kind of stuff, too. Right? Yep.
Well, yeah, at the time. I’m not going to say what they did, because I’d hate to get anyone in trouble. But they figured out a way to get the money. I’ll put it that way. I wasn’t automated charging was done manually by hand.
Okay, interesting. So. So you, you come in there doing this manual process? How do you even tackle a project like that?
So this is where I think the story gets somewhat interesting. They wanted it done in three months. And Adam has a sort of a tendency to push very hard and have very, very ambitious goals, and you know, where things are gonna go. And so the case he made was in three months, the company is just going to not scale this is gonna be a problem. It’s gonna, no big deal. Big problem, right? So three months. So when I looked at this, with a software engineering background, right, and I have to build this in three months, I realized, to your point, right, like, how do you pay? How do you reconcile invoices? How do you make sure that they work? How do you connect back to the, you know, like, whatever ERP system is on the back end, I realized there was no chance we can do this in three months. But maybe if we didn’t have a UI, in other words, like maybe if this system had no interface at all, it could work purely as a back end system. And so what we concocted, which is horrifying, and also just really interesting, was we decided to actually use the spreadsheets as the front end as the UX. So what we collectively did is built a system that would read the spreadsheets, from there automatically generate invoices, you know, finalize them at the right time, send them to the ERP system and charge automatically, you know, connecting through at the time I think they were using JP Morgan or something, I forget exactly which bank it was, but we connected straight to the bank to shoot the charges. You know, the credit cards were collected through UI that was based The only UI we had was a sort of UI for credit card collection that was hosted by the bank. And so it, you know, it took three months to build, but it actually works. So the you know, as the company was, would would move, people in the spreadsheets would get updated by the community managers on a regular basis, you know, every hour, we’d sort of like, check the spreadsheets, show, these are all the invoices that are gonna be finalized at the end of the month, feel free to check them as you go. And then right at the end of the month, they all immediately get finalized. And then we just start charging. And I gotta say it actually it actually worked. So, you know, there was there was, there was a couple of hiccups, as, as any software project done in three months might have, but there wasn’t anything fundamentally problematic in that, you know, like, we saved a huge amount of time on the community managers side. And then from there, you know, like, we eventually obviously got rid of the spreadsheets built out of your system, but that that was the basis of the system that powered WeWork up until I left in 2018. It probably still does, I just haven’t checked, but it you know, it eventually easily went through billions and billions of dollars of invoices.
Fascinating. So for the technical people on on the call, or listening to this, you say spreadsheet, is it Microsoft Excel? Like?
Oh, yeah, so it’s the technicals nowadays, Google Sheets. So these were Google sheets with API’s, right? So we could we could hit the API’s as much as we needed. There was a lot of trickiness to it, to be honest, that you wouldn’t have expected. So sometimes you get rate limited, so we can’t hit the API too often, sometimes the spreadsheets can get too big, there’s too many columns. And so the response object gets too large. And Google doesn’t love that, you know, we had to deal with the fact that, hey, like someone could easily type in any garbage into the spreadsheet, right? Like, you know, we had to lock the columns and other spreadsheets, they don’t have a similar format. So you can change columns and move columns, because the columns are important. But if someone puts in like, some text instead of a number, then the whole thing is going to break. So there was a huge amount of work, which potentially might have been more work ultimately, than just building the UI from the get go. To make sure that as an interface that actually worked in people when they caused problems and accidents are there were problems just due to processing that it didn’t actually like, block the entire building from from getting their invoices done.
Interesting. How many, like, how many locations are we talking about? By the time this is? I mean, you’re talking about hundreds?
No, no, thankfully. So we were eight when I started the project. And by the time three months later, we finished we might have been like 10, or 11. I mean, I don’t even know if we were maybe if we had even gotten any bigger than eight. I mean, this was something to like, make sure as the company grew, it wasn’t going to be a problem. It took us probably another six months beyond that to get rid of the spreadsheets entirely. So by then, you know, probably 20 or 30 buildings, you know, I’d have to really look and remember better, but, yeah, it by the time it was truly scaling the system had kind of taken over in the spreadsheets where there were thankfully gone.
Okay, so you built the UI? Eventually? We did we deliver? Yeah. And so you didn’t have spreadsheets powering hundreds or 1000s of offices all over the world now. 2018? No,
no, no, like, I mean, it’s one of these things where a startup the scales fast enough is always so far ahead of its own, you know, grows ahead of itself. So like, the intent was always that we do this in a very robust way. Because it’s like a business critical thing to do, particularly since we have a serious accounting team, that that really needed to make sure the numbers worked. And so it was critically important to do it correctly. And so it was important to get off of them. As soon as, as soon as reasonably could. Once we met Adams, three month, like, you know, boom, like you guys have to do this in three months,
right? You may have said this already, but let me ask it again, like how many how many people was at work at the time WeWork? With? So how many people were in the company? Like it? Was it? 10 people 20 100 In the whole company?
So WeWork at the time there were when I joined there were a little bit bit about 50 people. By the time this was all happening, let’s probably didn’t grow much higher, you know, probably 6070. Yeah. So still, it was still you know, like post Series A, like not a huge company, big ambition, but not enormous yet.
What would you build the the UI on?
Cool. Cool. So So you, you, you run this, this thing that starts off at spreadsheets that goes into this whole app. And somehow you get from there to founding your own, like, co founding a company that does that does data. So how does that happen? Sure.
Well, you know, let me start by saying that my interaction with the data team at the time, you know, the timescale we’re talking about, there wasn’t a data team. So you know, the first thing I had to do was sometimes backfill that a tiny bit like, you know, occasionally the, the accounting team would need something that that wasn’t obviously available through the ERP system, in which case, I would go in and build a query and find it for them. But, you know, they started building a data data team, and the first person they brought in was Ahmed, who is the founder of narrator, CEO. And so he actually grew the entire data organization. And so I got to know him pretty closely, WeWorked very closely together. And as we were grew, him, this is effectively the origin story of narrator. So as we all grew, you know, to put it in context, when I started, as I said, it was 50 people, when I left, it was 5000. So it grew 100x In terms of people that grew 100x, in terms of members, you know, By every metric you could think of it basically grew 100x, which, in five years is pretty actually nuts. And so data was no exception. And by the time sort of, like, you know, around 2018 rolled around, you know, the data world that WeWork was, like 40, data engineers, and scientists, you know, hundreds and hundreds and hundreds of data tables, you know, the big old warehouse, like dozens and dozens of like, systems built by engineers, and my department, you know, all doing their thing, right. And it had gotten pretty messy. So, you know, the apologies for the non data, folks here, I’ll make an assertion about data. And I’ll try to keep it brief. But I will say that data can get annoying and very one very particular way, which is that numbers don’t match, you know, there’s not a single source of truth. So you think to yourself, Okay, you know, CEO, I want to know, the quarterly sales. And someone’s like, cool, here’s your quarterly sales, but like, is not the same as quarterly sales that the head of growth is getting when they say I want quarterly sales, sales by region, and then if you sum it all up, are those actually going to be the same thing. And it sounds like they should be, but they’re actually not. And at least not always. And the reason they’re not is that every time you make a new metric, like a new plot, like a new some kind of new number, you generally have to sort of like start over and write sequel for it and look at a work other people have done but unlike software engineering, you can’t really like use stuff that people have done in the past. So you kind of hope that if someone else made a quarterly report for sales, that you can then figure out how to reuse that for to break it down by region, and maybe copy their logic. But if you don’t see it, you’ll make your own, and then your logic is going to be subtly different. And so you end up with just dozens of subtly different things across the company. And it is actually a fairly common thing for the CEO to come in and say, Hey, like, why are these numbers not matching? What’s going on? So we saw this in spades that WeWork and realize that basically, these analysts whose job it is to give numbers and give guidance, we’re spending 80% of their time just reading the SQL queries, trying to understand what was the actual source of truth? Like, why do we have three versions of this concept called sale or occupancy or whatever for WeWork, right? And that just like, as the company grew, that became more and more problematic, because as you get more data just becomes more messy. So, you know, I could talk about whenever it gets there, but you know, the question I’m supposed to be answering is, is well, how do we get there? How did I actually move from, from software engineering over to narrator is, by just being close with Ahmed and the other two co founders, we realized that this is just insanity. And there was a really big opportunity. And so Ahmed had actually prototyped out a completely different way of doing data on top of WeWorks data that he was really excited about. And you know, we all took a look at it and realized like this, this is actually fairly transformative. Let’s go and see if we can build something out of this. And so Ahmed kind of started on his own and ran off and started building it and you know, quit first and I joined him a little bit later. Once it was clear that yes, this technically could get built. So you know, it’s it’s a little bit of a weird story of being a software engineer to go to a data company, but it’s mostly because I realized, wow, like, there’s this whole area where people are doing work that that can be made substantially substantially better.
Right. So there are a whole bunch of founders and CEOs who at some point will see this video and they’re, they’re cringing along with you, when they, when they hear that like sales total and sales total to by region don’t match because that’s a common problem like that. Maybe not that one specifically. But like, I think it’s interesting that you point out that like data data sets are. They seem like they should be objective, but they often don’t. Like they don’t tell a consistent story is that I mean, that’s, that’s what I’m hearing. Yeah.
it’s it’s like, it’s not even a philosophical thing, it’s interesting as purely a mechanical is maybe the wrong word. But like, when it comes down to just doing the work, it’s difficult to put different disparate datasets together, it can be a lot of trouble. And so you do it once, and then you need to do it in some other subtly different way, it becomes a real challenge to go, should I change it? Should I make a new one that’s similar? Like how close are these and then you get really lost in the details?
So it’s so hard, it’s so so I’m going to open the door to this, how does Narrator solve that, and then I want to ask you some questions about like, what, what people can learn from the store.
So Narrator basically, you know, kind of just turns things on his head a little bit. So whenever it does is it says, Let’s not think of data in the same way that we normally have words, these are a bunch of tables. And these have to all be joined together to make something instead of things in terms of activities. So everything is customer centric. The idea is any data that your company is interested in doing anything with, at all, like, at some point, there’s some kind of customer taking some kind of action at some time. And this is the thing we always let people challenging us on, because we can sort of always frame it that way. So let me give you the I’ll give you the WeWork example. Because we were talking about WeWork I’ll assume that maybe look folks listening, no, you know, WeWork leases out office space, like that’s maybe the simplest way to think about it, right? That’s probably good enough for our purposes. So a customer might be a person, you know, that decides to get an office. So you know, day to day care about, somebody signed a contract. Like that’s them saying that they’re going to you know, pick up a subscription for an office for a few months. Like that’s an that’s an action, that that Narrator then then sort of works with customer moved out, a customer reviewed a webpage customer submitted a support ticket customer posted on the internal social network, you know, all these activities is whenever it calls them, or activities. And apologies, if I go long, I’m trying to explain this quickly. Because I know we have a lot of interest, we’re also gonna keep going keep. But the fundamental idea is Okay, so once we define these activities, and this is all based on the data that’s sort of already in someone’s data warehouse, then what we do is we put them in a special format that we called the activity schema, and effectively, it’s in that format. So we make a big table, that table literally says, here’s the customer, here’s the action they took, here’s when they did it over and over and over and over. That’s like just one big table, like millions of entries, all saying this person did this at this time. And then from there, we realize, hey, look, we can like answer just about any data question by just reading these activities together. And what makes this different, is that it’s actually very easy now to combine anything together. So previously, I’ll give you a real example. Because I know this is a little abstract. Previously, you know, the data team that WeWork was asked, Hey, like, we really care about, you know, leads coming in people, you know, deciding to take a tour with with the, with the community manager and then deciding to take an office. And but we want to know, hey, do phone calls have any effect on that, like, do they call first and then take a tour. And the poor data, people are like, Oh, my God, like, I have no idea how to connect a person in the phone system. With a person that took the tour, there’s like, the data doesn’t have that connection in there at all, like, and so it’s a huge amount of work to try to relate these things together. It’s this fundamental problem that people just call user attribution. And so Narrator What we’ve done is we just say, like, look, every time you get a phone call, you just say who you think that user is, every time you get a move in form, you say who the user is. And then we do the work of sort of stitching that all together with a takeaway being that if you ever want to say, oh, what’s the conversion rate from somebody visiting a website to somebody buying my product, like visited website, buy products can always be matched together Narrator immediately, and you can do that with any number of activities like no matter what they are, so the phone call to move, inform example, same thing, like soon as you want to know how they relate, you can just do it immediately. So effectively, once you’ve built out this core, little bit of your data system, the narrator can then go across all of them to get any data you like. So I’ll say one more sentence, which hopefully ties it all together. And then I’ll be done with this super long answer, which is basically, in the old data world that I spent all this time describing Whenever someone asks a new question, like a new flavor of a question that wasn’t asked before. So you know, I mentioned give me quarterly sales by region. What if it’s quarterly sales by city and yeah, I know I said region before but now I want city I really want to dive down, right? Like any other facet of that, if you didn’t think ahead, some poor data person then has to go and rethink how to make it work. And so every time you ask a new question, the old way it could be weeks of work before you get your answer. A Narrator wants it set up. This is what we do fundamentally at Anytime you won’t have a new little facet on a question, it just happens immediately, there’s no extra work to be done. So you can ask a question, then a follow up, then a follow up, then a follow up, which is we feel like the right way to do data.
The difference, I think, tell me if I’m wrong in, in most orcs, there’s datasets in different places, right? Like, they’re all over the place. And as a as you’re trying to make sense of it, you have to sort of figure out how to tie them together every time somebody asks a new question.
That’s right. That’s right. And in narrator, you do that, I guess ahead of time is one way to put it. You don’t actually ever have to relate them directly. You just say, okay, you know, I’ve got this data set coming from Zendesk, which Zendesk, you know, does all my support tickets. So I just have to say, user submitted support ticket user closed support ticket, and that’s it, I’m done. As long as I sort of expressed that, then I don’t have to worry about how Zendesk relates to any of my other systems, as far as the data goes. So you just individually sort of put each one into the right format, and then you’re good to go.
What, what, so say, you’re talking to like a SAS founder, or SAS, SAS, CEO, small company, $10 million, something like that, in that range? What are the like, what are the pieces of data that they need? letter that are most commonly frustrating for them that like they should be looking at? Because I feel like, there’s a lot of fear of missing out when he used to come to data to the sort of entrepreneur and growth world, right? It’s like, you look around, do you think all these people are doing cool things with data? I can’t even seem to figure out what my quarterly sales are. I’ve heard that some flavor of that version for from a lot of different people in a lot of different corners. So talk to that person for a minute, like, how should they be think?
Cool? Yes. So there’s a lot of different ways to think about this, the first thing I will say, is, it is always worst to overdo data than it is to under do it. You know, if you’re, if you’re thinking I’m not doing correctly, I’m not doing it right, you’re going to engage your brain a lot. And really second, guess what you’re up to, and really try to make sure you’re doing it correctly. Whereas if you’ve gotten like, some fancy system that answers every question for you, you’ll be like, cool, I’m doing data. And so you know, it’s one of these things you don’t want to overextend yourself on. So you know, to answer a little more specifically, you know, I think the very first step is to make sure that the metrics you care about, you know them, and they’re available to you, and you trust them. Now, any CEO, I think is going to know the metrics they care about. So at minimum, there’s like, you know, what’s my revenue? Someone’s going to ask you, How much money are you making? Like, you’re gonna want to know how, like, how much time do I have literally, the state of the business is something you’re already going to want to know. And if you don’t know, you’re going to move heaven and earth to get it figured out as soon as possible. Like, when we were tiny, we could figure it out by literally adding it up in our heads, because we had like two customers, and you knew how much they were paying you. But you always know that number. And so that’s the first step is to is to have a reliable source that you trust that and I mean, trust me, like, you know, understand where it’s coming from. So, you know, for example, if someone is using a billing platform, like let’s say, fastspring, I assume you guys have things like that, like you have got revenue, I’m sure you have MRR I’m sure you’ve got. And if that’s the only platform you’re using for all your sales, then you can be pretty confident that those numbers are going to be close enough for your purposes to sort of, you know, and if you ever want to check, you can always manually double check. I’m emphasizing this because it’s actually less common than you realize to have data in a place that makes sense that that you trust. So that’s step one is like the metrics you care about. And we can talk about what metrics should you care about, but metrics you care about, make sure you’re getting them from somewhere that that you believe, and that you’re getting them consistently, this only becomes a problem when when you start to get bigger, and things come from multiple places. So you know, like real example, we had Narrator are using a billing platform for certain subscriptions. But for others, people are paying us in different ways. And so like, I can’t now ask our billing system, what’s my MRR? He doesn’t actually know, because I’m getting, you know, effectively checks in the mail. Right? So like, they don’t know about that. So the trick is to understand is this is someone looking at the source of truth. It doesn’t know the totality of the world. If I can give you another example, that’s maybe a little less intuitive, but still pretty interesting that illustrates this. This applies more to maybe marketing, folks. But when people run online ads, you know, they’ll run Facebook ads, or they’ll run Google ads, or they’ll run you know, whatever, Pinterest. And frequently these these ad platforms will say, you know, obviously, here’s the number of ads we ran, here’s how much it cost, whatever. But then they’ll also take a stab at guessing, hey, all the sales you got, I’m going to try to say whether or not someone saw one of our ads before like you made that sale. And the problem is these platforms don’t know about each other. So if you were to sum up all the orders that Facebook thought it was responsible for plus all the ones Google thought it was responsible for, right and so on, it would sum up to far above 100%. And so that’s a case where you can’t really too much rely on those numbers and you have to real Lies, the source of truth actually doesn’t exist in any of them. It actually exists with all them combined. Which is actually a segue to Well, why do you need a data team? But I’ll let you sort of lead this discussion and get there.
So when do you need? So I guess I zoom back? And I’ll just ask that question in a second. But like, you talking about it, WeWork if I go back to the WeWork example, you built a billing platform, on your own, like, hacked together in three months, and then over several years, developed it out to something that was functional. That’s yeah, that’s the right timeline, right. So it’s not easy to build a building? What’s like, okay, yeah, let’s, what? What, what were some of the things that you ran into, as you’re building it that there were like, this is way harder than
Yeah, um, there’s a couple things. One that was pretty interesting is that I mean, it’s, I’m gonna say two. So if I forget, one, remind me, when that was interesting was when we first did it, at the very same time that we were kind of working on launching this billing platform, the company decided to switch their ERP system from QuickBooks to SAP. And so we have to sync all our invoices straight to that system. So right, so we had to write a sync to go to QuickBooks. In two months, we, we can only build monthly. So I had a test month where I tried to do it on one building to make sure the system worked. So that was month two, and then month three was when we launched it with everyone. So month two, we did it with QuickBooks, and everyone was happy. Yeah, look, it worked with this one building. So then month three, we go live, right, we couldn’t test in between least not in a substantial way. And so we went to SAP. And then I, you know, a few days in the month, I get, you know, effectively the CFO running into my office saying, We can’t collect on like so much revenue, just super pissed. Like, the accounting team is stuck. There’s like, however much revenue like just outstanding we’re not getting. And what happened was, we obviously charged you know what we could automatically but as you’re more than well aware, that doesn’t always work. And some people you gotta go and like get money framerates called dining, and they didn’t have the ability to see who that was anymore. So QuickBooks had it. SAP will, however, was configured did not have that, for some reason. I don’t know the details. And obviously, there’s some little bit of there, that might not be 100%. Correct. But I mean, the takeaway was, they were like, we’re not actually sure who owes us money, and who to follow up on to get that money. And I was like, Oh, I wasn’t aware, like probably failure on my part not to be aware that that was a thing they were going to lose by switching over. And so we had to like emergency, you know, as fast as possible. Bill, this is the first day why we basically built into the product was a UI for the accounting department to get a basically a list of all the folks that the payments didn’t work out, and have them have a way to then go in and ask, you know, to charge again, which clearly is a fundamental part of any billing platform. But I had to go build that sort of an emergency because we didn’t anticipate them not being able to do that.
Right. Right. And so then, that’s yeah, I mean, that’s, that’s fascinating how many people how many people,
it was immune, one person building that, and then very quickly grew to be a team and then eventually became like, teams of teams, you know, like, by the time I left, we were probably 3040 People just in that, on that system. I imagined, you know, that makes sense. The other part that was hard that I find to be really interesting, and this was somewhat unpredictable, is that it’s the rules, and the business rules are not obvious for the business. And they change and they can get weird. So you know, when I decide to say, Okay, this is an invoice and here’s how much is on the invoice, you know, the system is making that that call, there’s actually a lot of really interesting business logic involved in there, it’s like, so, you know, the company will sell an office for a certain amount of money, but then one day, they’ll take that office, and they’ll break it up into four desks. And they’ll sell those four desks individually. And so all of a sudden, you got to like move someone out of this one thing that’s like being paid in a unit and then break it out into these different units. Or you get, you know, businesses that say, say, I want this invoice to go to a completely separate department, but the business is all the same. So you then you got to find, you know, come up with some idea to represent businesses that have these multiple hierarchies. You know, you have to start to say, oh, you know, now we’re going to start selling virtual things that have no office at all, like, can you represent that and these are, how the rules of them are going to work. And, and so in some cases, for certain products, the, the business logic involved to sort of actually build it the way that businesses wanted to, was pretty nasty. And one of them in particular was, you know, they said, Hey, let’s do this new thing where we’re gonna give people desks on a sort of subscription basis, but you can come in whenever you like, and just use the desk. So you pay a certain amount but then you get to use like three days, you know, a month or something like that, right? Like a sort of a virtual desk thing and billing for that became just a nightmare. It was Huge amount of work. So it’s interesting because the reason I say that particular case is that it’s a blessing and a curse. The the curse is, you know, this company is on the hook for building this thing. And frankly, they don’t want to invest in it. And like this is clearly all your customers I think are gonna agree with this. Nobody wants to invest in building a building system, like it is not core to whatever your business is doing. Except for you guys. Right? And so you know, what is we were doing, like having this thing and, you know, costing them a ton of money. And the thing is, is that with these complicated things, the business was growing so fast, trying out so many different things that we could just do whatever we wanted, as far as, as long as we could write it in the software, we can make it work. So we’re having this really, really tight coordination with with us doing the billing actually enabled a lot of stuff at the business, a lot of crazy things happen that that I never would have thought they were ever gonna try. So we periodically just to cap off that that point, we periodically we look around and say, Okay, is there like a system out there that someone has built that we can just migrate to? And every time we looked at the answer was always no, up until when I left?
Very interesting. Is it? What What? What about a small company? What about like a two person company? Like how, like, what do you when it comes to Narrator
We we don’t write it ourselves? No, no, thanks. Right. Like, if you’re a subscription business, like that’s well understood, you know, if you’re like an E commerce shop, selling widgets, like that’s well understood, I think that the model built out by a lot of guys like you guys, Stripe building chargebee, you know, I not gonna express any preference. And in any of these things, I don’t advocate for anyone. But I think generally like, this has become a well understood problem over the years. And I imagine that having been in business, as long as you guys have, you’ve seen just about most models people want to do you know, someone’s like, oh, I want to charge weekly buy the unit, but only a unit is greater than 20. I’m sure you can have a way to figure that out. And when you get to, like cute with it, it’s also sort of a sign that your business is doing like something a little weird. So my take on it is if you’re a small business, like you definitely don’t need a roll yourself like just don’t, it’s absolutely not worth it work on your work on your business. Know exactly like having built one I’m not I don’t want to do it unless it’s strictly required.
Yeah, you say,
Yeah, work on what actually is gonna get you customers? Yes. Like, you know, understand your customers, like work on marketing, do something, but don’t don’t work on a billing system.
Okay, cool. What? Give me Give me like one or two, one or two things if if somebody listens to this interview, what,
What do you want? Well, as far as data goes, I think that might be interesting to talk about. The takeaway, I would say is you can go a lot farther than you think, without hiring real data. But there comes a point at which you absolutely have to do it. And, and it feels like a big chasm to jump in some ways it is, but it’s actually not that crazy to do. So maybe I’ll quickly illustrate that for just a moment if you’re interested. So you know, when you’re a small company, and you’ve got your mind wrapped around how everything is going, you understand your metrics, and you know where you’re going, you know, like, data is important. And it’s important, particularly so you don’t accidentally believe something that’s not true. But that’s the main reason it’s not going to predict the future for you. And it’s not going to tell you how to run your business. So you know, if you’re wondering why people are churning, and you’ve got like, 50 customers go ask them, right? Like, this is stuff that all startup founders understand. There isn’t some magic data that’s going to tell you something super innovative or super interesting. But when you get to a point where I illustrated this earlier, where data starts to get spread across multiple places, that’s when you start to lose your ability to understand what’s going on. So you know if like, like with, for example, when I said, you know, like, what’s my MRR? Well, I’m getting it from my billing system, and I’m also getting it from people just sending us money. That means that no, I no longer African, you know, one place to go. And when you do that, then is that is when you should hire somebody for data. Because it seems conceptually straightforward. But it’s incredibly hard to properly bring data together in a way that is doing what you expect. Because like, for example, you guys, like I said, I’m certain you have tons of great metrics, MRR, you know all this stuff. But it’s not your job to go get it from some other system. So the moment I want to understand MRR, I need to go like write that myself. Because now there is no system that will do it for me because like there isn’t like you guys aren’t gonna grab it from some other thing and sort of compute it for me based on someone else’s stuff. And so what I would say is be wary of, of tools that sort of try to do this, but not really, you know, like a tool that will tell you, hey, we’ll give you all this great data, just merge this and merge this and we’re highly generally skeptical of those things. Because once you start doing that you actually do need someone who understands I think data to sort of help you do it properly. You know, the way that you would proceed there is is bring someone on board that’s like generalists, you know, analyst background did engineering background willing just to work all across the stack, like, you know, don’t go ham and like try to hire a company that does in depth user attribution. That’s all they do, like, don’t worry about it, all you need to do is have your data in one place so that you can see it right start from that. And so that person will probably set you up with a data warehouse, they’ll find a way to send the data from, you know, n different systems into that data warehouse. And then from there, that all that is relatively straightforward. There’s great tools that do this already for you. From there, they can then sort of start to answer your data questions. And, and, ideally, you know, that the thinking that we have is, people know, their business, like, you know, marketing executives live on data of CEOs live on data, they understand this, right? So if you if you show them, hey, like, this is your MRR, this is your churn, this is your net revenue retention, they’re gonna have hypotheses for how to improve it. And what I would say is like, at that point, you can you don’t have to get to like in depth about it, you just form a hypothesis for you know, why is Why is trying to problem like, maybe we should look at fixing this thing. So go ahead and fix it. And then now that your data that you trust that you can track it, see if it changed. That’ll take you incredibly far, like if you guys, if you end up with a data person that is able to sort of get your stuff in place to give you the fundamental metrics for the business, you’re basically ahead of most companies, and I say that actually, without being facetious. It’s actually true. So I’d say don’t hold on – you don’t need to do in-depth analysis, that early on; focus on the basics.
Right, so you’re talking about just getting accurate data that you trust first?
Particularly when it starts to get spread across different places.