“Brand marketing at Uber taught me how to rewire how I thought about measurement”

An interview with Sundar Swaminathan, Uber's former head of data science for brand.

This conversation was probably the most impactful I’ve ever had on the “brand” debate.

I had the recent pleasure of sitting down with Sundar Swaminathan who amongst other roles was the data science head for brand at Uber.

Uber to me is OG performance and data.

Cynics have long argued Uber’s approach to growth was an epitome of ‘growth at all costs.’ But as someone who studied it as much as I could from the outside, it always made sense to me as a performance marketer by trade.

In Andrew Chen’s Cold Start Problem, he tells a story about the tipping point at Uber. If the wait time for a cab goes over seven or eight minutes, then riders leave the app. If riders leave the app, drivers leave the app. The liquidity falls apart. And so driver incentives become key. Surge pricing becomes key. They are the growth levers to keep that engine going. This isn’t splurging cash, it’s science and maths.

And so my default view has always been Uber valued data over all else.

Danny (previous guest of the blog) introduced me to Sundar a few months ago and it was a real delight to get to chat.

Sundar started off his career at the Department of the Treasury, but unsurprisingly it was his five years in Uber’s data science teams that took up the majority of focus.

I hope you enjoy reading this as much as I enjoyed doing it.

The measurement story for Uber’s $1bn of brand spend

What’s the 3-minute career overview?

After graduating in finance, I became a software developer for two years, before going back into finance at the US Treasury – a very interesting experience. However, I decided being a developer or working in finance wasn’t for me, and jumped into marketing analytics, which has since made up the majority of my career. 

Most of that time was spent at Uber, where I built the brand data science team. I did a lot of performance marketing, specifically around acquisition, activation, lifecycle marketing, and email channels across the entire funnel. 

After that, I worked with a couple of startups based out of Amsterdam, one in travel tech, the other in beauty and wellness. There wasn’t a lot of room for analytics at that early pre-seed/seed stage, though, so I was let go before long. 

At the beginning of 2023, I started consulting. Then, at the end of the year, I took some time to figure out what kind of consulting I really wanted to do. And I think I’ve narrowed it down. I’m now a marketing analytics consultant, either setting up or fixing marketing analytics teams for consumer marketing leaders. 

What was so interesting about the Treasury?

At the time, the US Treasury was managing about $20 trillion of debt. It was going through a transformation where it wanted to bring in more data science and analytics to explore how the US Government could manage all its data. It brought in ex-traders, ex-quants, data scientists – there was a micro-pocket of innovation, almost entrepreneurial. We launched a new fixed-income product, something that happens once every twenty years or so. I was lucky to experience that innovation while I was there. 

But that wasn’t all.

Every year, the US debt ceiling becomes a political showdown: Republicans and Democrats argue that if they can’t push certain things through they won’t allow the debt ceiling to be extended, meaning the US could default on its debt. This would have terrible economic implications; the entire world is essentially priced on US debt. 

At the age of about 25, I got to build models that showed when the US would run out of money if it didn't raise the debt ceiling. These would go to the Secretary of the Treasury, who would then take them to the President. There are very few times you'll get that much responsibility at such a young age. I loved it.

So why didn’t you stick around?

It was one of those times when you look around and realise this isn’t what you want to do. It just wasn’t appealing. 

Around that time, a friend asked if I knew anyone who did analytics. I told them I was doing analytics for the US Treasury – if they taught me the business side of things, I could pick it up. 

Timing was everything. It was around 2014/15; digital marketing analytics was a buzzword, everyone wanted to be data-driven.  It's very rare to be able to jump into another industry, so I took the opportunity. And I’m glad I did. 

What did you like about marketing analytics when you got into it?

It offers a unique perspective. If you’re good at your job, you can understand both the quantitative and qualitative patterns of where people are in the funnel and why they make decisions. If done right, it’s hugely empowering to be able to produce insights that a marketer or product team can use to change decisions, make an impact and ideally improve the customer experience. 

But, while it’s a cool field to work in, it’s very challenging. You’re often thrown into a role where the expectation is that you’re insanely good at prioritising communications, putting together decks, and number crunching. Nobody teaches you how to prioritise, though. Nobody teaches you how to do anything. 

The industry continues to evolve, of course; lots of new things keep coming out. So it’s a very valuable skill in my opinion. 

Was Uber the first tech startup you were in as an analyst which had a growth team? What did that team structure look like when you were there?

Uber was the first one I was at, although I understand Facebook was one of the first to pilot the idea of data-driven growth, investing heavily in data analysts and data scientists to squeak out every little conversion rate improvement. They already had crazy network effects and product-market fit going for them, but this helped them accelerate faster than anyone else. A lot of understanding came from this of what data science and data analytics could do for a growth team. 

The way Uber was structured meant you couldn’t push an initiative through unless it was backed by data. When I joined in 2016, there were what they called “driver ops” who’d analyse the supply side, and “rider ops” who’d analyse the demand side.

It was a very local model: you’d report to the general manager of that region, who reported to the head of US, who then reported to Travis Kalanick. And they’d shred any narrative that didn’t have any relative data backing. Being data-driven became part of the culture. It was by far the most data-literate company I've ever seen. 

Over time, this structure moved away from a localised model to a more centralised function, although it was still heavily matrixed - I reported to the head of marketing data analytics, who reported to the head of data science and analytics.

“I’d sit exclusively with either a brand marketer, performance marketer or lifecycle marketer in a “pod”, where I was considered a partner. We were judged by how successful the marketers were, rather than just how many analyses we did or how many tickets we closed.”

It’s an approach I take now, in my consulting. 

Uber did a lot that I’ve never seen any other companies replicate or succeed at. And it was at Uber that I learned how much growth teams value data. 

What was the process in those pods? Did the marketers come to you with ideas they wanted you to look into? Or did you – as an analyst – come up with insights after digging into something?

It was a healthy mix of both. In some cases, we were expected to look at data proactively and surface the insights. Marketers were very good at trying things out. Roadmaps were planned quarterly, but not set in stone. They were very open to being flexible. 

In other cases, the inspiration came from other people. For example, our CACs would fluctuate by crazy percentages week on week. But nothing had changed. We were spending millions each year on performance marketing in the US, but spend levels were consistent. So I took a step back to dig into the data. 

The only reason for those wild swings in CAC was that while our numerator spend stayed the same, our denominator fluctuated with seasonality - one week it would peak, the next it would drop.  It turned out we were non-incremental because everyone in the US knew us – there was actually no need for us to advertise.

“That saved us $25m a year on Meta”

To give you some additional context, I was also responsible for owning Uber’s saturation analysis for the entire US – the total addressable market versus its penetration. From the third-party data we'd purchased, it was clear we had extreme signup penetration in most major and all secondary cities. Using that knowledge, we looked at our data on spend and realised that spend wasn’t related to the number of signups. 

We then ran a three-month incrementality test. It came back to say the entire span was not incremental – we just had such deep penetration. We didn’t have to target at that level all the time for the generic use case of “use Uber”, because that’s all people were doing at that point. Either they knew about Uber, and they were using it for that use case, or they hadn’t found the use case. It didn’t make sense to be blanket top of mind all the time. 

There was some nervousness, but it was ultimately an easy sign-off from everybody on the next steps. We turned off Meta. And we saw no change. This saved $25 million a year on Uber Rides.

Turning off that level of spend would be catastrophic for most companies. Where does the journey go from there?

It’s a classic case of taking a step back and asking how you can increase customer lifetime value. At this point, you recognise how many people you’ve signed up and have just never activated. 

It starts to create new strategies – rather than filling the top of the funnel, you need to continue to nurture and build a better funnel. You get a lot of ideas around using your own channels for ways to re-engage, reactivate, and bring customers back. 

You become a lot more customer-centric and force yourself to build better retention foundations. It forces you to get better at creating systems to monitor spend and wastage, such as more frequent incrementality tests. It’s a wake-up call to leadership to invest in better measurement tools. 

It was a soul-searching moment. But that's when you get a lot of innovation. Performance marketing teams will talk to CRM teams about partnering on re-engagement campaigns, for example. Ultimately, though, it’s an opportunity to try out new approaches and realise when it's time to move on.   

Uber would have been quite large at this point. What was the breadth of your work? How was it all divvied up?

There were probably five or six hundred marketers at the time, plus around 75 analysts and data scientists. There was a lot of breadth across geographies, as well as across products, with Rides being a completely different product to Eat. 

We saw an emerging desire to cross-pollinate and cross-sell. People who were exclusively Rides or Eat began trying a mix of both, using diverse channels like performance marketing, lifecycle, and push. 

I also got to work on brand, which was hugely interesting because there’s a complexity of channels just for brand delivery, many of which I’d tried for performance marketing. I was also working at different stages for different products. In 2020/21, for instance, I was working on campaigns for Uber Eats to not only build out the category in France but also build brand awareness into that category - it was a very different adoption curve to the US where Uber Eats had really taken off. I worked on so many different strategic projects, too – one year it was about growth, the next it was profitability, and the next it was profitable growth. 

Brand marketers often don’t view the world with the same data lens as performance marketers. You did data science for brand at Uber. I’m fascinated to hear what you learned. 

Of all my experiences at Uber, that’s the one that gave me the most unique perspective and, arguably, an advantage compared to others working in marketing analytics. 

There are brand marketers who are data-centric in that they know their brand numbers. But they’ve never been asked to prove the ROI of brand. That’s when they embrace it and start to play around with learning these new ways of approaching measurement. 

At Uber, once brand marketers understood I was trying to help them prove the ROI of their work so that they’d get more budget and keep doing more interesting things, it became seen as a useful flywheel. 

Most people are trained to think they need to see immediate ROI. But brand doesn’t have an instant effect like performance. You’ll be asked how to know if brand is working.

The answer is that, instead of proving ROI, you have to prove the metric you’re trying to move is moving. In an awareness campaign, there are plenty of marketing methodologies to prove that awareness has moved. 

I’ve always said that it’s only there if you can measure it. Brand’s the opposite. It’s there, it’s just hard to measure.

Uber taught me to rewire how I thought about measurement. 

Did you find there was a link between the increase in awareness and the increase in consideration? Did you find a way to measure that gap?

Yes. There are some assumptions and connections you have to make. For example, the Superbowl is a big moment. It’s around $5 million+ for a 30-second ad. And we ran lots of ads for Uber. When I was there, it was the first time Uber had run a Superbowl ad. The immediate question was "What's the ROI?" The only way to link to the immediate ROI of a Superbowl ad is by the increase in search volumes, app downloads, and orders relative to last week’s or last year’s baseline. 

You have to outline how you’re going to measure and make those assumptions beforehand, so it doesn't look like you randomly pulled connections after the fact to make yourself look good. 

Uber has a very large addressable market. But if you don't know how big your total addressable market is, how would you – as a consultant in a business – approach trying to discover something like that?

I'd want to do it quickly, between four and six weeks. I'd start by trying to do as much customer research as possible on your current population. It's the classic Pareto rule – where is 80% of your business coming from? What does the other 20% look like? Many companies don't have that clarity. 

Then there’s some hand-waving to say, most of your 20% are going to be early adopters. So I'd match that to an adoption curve. The other 80% that haven't converted to power users are probably late adopters, so there's still room to grow there. Then there are the adjacent markets you can expand to within six months to a year. Again, it's a case of making a few estimates to give you a weighted saturation: "I've probably already got X% of my power users. Therefore, if I still have Y% left, I know they'll bring in Z amount."

That's how I'd approach it by hand waving. I'm not convinced you should spend more than a couple of weeks on a penetration exercise because the additional accuracy you get isn't worth the time it takes. It's okay to make assumptions. You can find out if you are off by doing marketing or advertising tests.

Few companies could match Uber's brand marketing budget. From what you understand, is it worth doing early? Or is there a threshold you reach at which you should probably start because you need to be able to commit X amount every month for the next six months?

Brands invest in performance marketing because they can see the immediate effect it has. Smaller brands need to prove product market-fit because it allows them to do more product, message, and audience testing. 

More mature brands, however, invest in brand marketing for a reason. They know performance marketing has a cap or isn’t as cost-effective. So instead of adding more channels, they start to layer in brand. 

Once you’ve reached a point where your creative and your flow are good, it can be hard to continue to optimise in-platform. However, investing in brand opens up more channels and makes them more efficient because of the stickiness of the touchpoint. 

If a company hasn’t found product-market fit, it shouldn’t think about brand or tone. It should find the audience first, then find ways to attract more of that audience by doing more top-of-funnel brand marketing. 

Don’t get me wrong, I’m a huge believer in brand; I’m an advocate for it. But I think it’s the least understood marketing discipline. Most don’t understand brand marketing. And that’s fine. Because if you do understand, it means there’s still a lot of room for it to be your competitive moat.