Sacra Logo

How does in-house development of recommendation systems affect's strategy?

Andrew Yates

Co-founder & CEO at

Absolutely. The only competitor we care about is “build in-house.” Of course, we would care quite a bit about Google or Facebook making it a very direct competition, but we'd actually be pretty excited about it, because it would help some of our go-to-market. And given what I just said, well, I don't really think that's going to really be an important part of their strategy.

The funny thing about the way you described that is it takes for granted that you have a top engineering team that is going to build it, and it works. That's how people describe it. It's like, “Oh, well, won't everyone just build it themselves?” Let's look into that a little bit. This goes back a little bit to some of our origin story in hiring engineers and building these teams at Facebook, Google and Pinterest. And everyone on the team, by the way, has experience with building this at other similar companies like Grab, as an example.

First, you have to hire these people. How much does that cost? Well, have you tried to hire senior staff engineers in machine learning and real-time infrastructure lately? It's going to be at minimum $500k or millions of dollars per year per engineer. And that's just to get them with a company laptop, and everyone's remote -- so not even in the door. They don't even show up at your office. You're like, “Okay, now you’re on payroll,” and you send them a laptop. Now they have to build the thing. Okay, so how many of these do you need? Well, look at the industry. Is it one? Is it five? Is it twelve? Is it 50? Is it 100? Is it 1000? Is it 10,000? The number of engineers keeps growing, and there are companies employing this number of people on every point on this curve, and it keeps going. When your company is really serious, the number of engineers keeps growing, at least super linearly.

Then you think about what are these engineers working on? Well, let's say you're just getting started. Okay, we're going to hire a top team. We're going to hire a VP out of Facebook. They're going to put out a job listing. We're going to say, “Look, we just raised at a $2 billion valuation. We're super hot. We'll pay you a $200k salary, and our stock options are going to be super awesome, and you can make up to a million dollars a year.” First of all, you're going to be competing against Google and Facebook who will pay you in cash. Where's the growth opportunity for you worth that much money? Second of all, after hiring all of the people and getting them to work together, how much time do they have to go build a really fantastic system? Have they worked with each other before? No. Now you have one quarter with a bunch of new people who don't know each other, and you’ve got to slap it together.

We've seen it over and over again. You cut all these corners and you get something out the door. And voila! You're showing a banner ad at the top of your app, and you're working with some big brands. Look, congratulations! But let's look at Coupang. Coupang has several hundred engineers on ads -- just ads -- and several hundred engineers on just search and discovery, and they have all the supporting infrastructure, data, client logging, and then, of course, all the connections back to fulfillment house. It gets super complicated. It's like, “Well, is it really a reasonable prototype that you showed a banner at the top of search, or you pinned the top result and you have a simple crud app about now you can control who's at top?” I mean, I guess technically that's on the same path -- in the sense that if I jump up, that's closer to the moon, depending on where the moon is in the sky.

That's what we've seen with “build in-house,” and we've seen it over and over again. If you've already gone to SoftBank and raised a billion dollars to go build this, and you're dead set on being a technology leader or the next Uber, that’s awesome. That's probably not a customer that is going to be great for us today, although we have a lot of specialty tools that maybe you should look at and that maybe you haven't already built. But for companies like JOKR, how much time do you have, and how many businesses can you be the best in in that amount of time? “You're going to be the world leading media technology business and the real-time delivery business, and you need to do it in six months.” That doesn't sound like a successful strategy.

But I don't think people have really thought through this carefully, because, as you described it, “We'll just build in-house. How hard can it be?” It's deceptive, because it's not that hard to hire a few engineers. It's not that hard to convince yourself that, “Hey, look, someone worked at Facebook before. We're going to rebuild Facebook to get started.” But then two or three years in, your marketplace isn't growing in any sustainable way, you have this huge mess, and you just keep doubling the size of the engineering team. It's not really progressing in any sort of way.

I think people will start realizing that they should focus on being an expert in one business versus two, in terms of where you're going to invest your top talent and resources. Is it going to be logistics? Is it going to be fulfillment? Or, is it going to be the commercial media optimization in your app? Which of these do you want to specialize in? If it's all of them, why wouldn't you just be outcompeted by someone who focuses on one and outsources the other pieces?

Find this answer in Andrew Yates, CEO of, on driving marketplace ARPU with personalization
lightningbolt_icon Unlocked Report