Retrospective on 3 Years at Uber

Up until mid-2014, I didn’t know that Uber existed.

This was because even though at the time it was valued as a multi-billion company, it was still very new in Canada.

At the time, there had been less than a few hundred thousand transactions in Toronto, and it was not even a name in Vancouver yet.

The only reason I realized Uber existed was as we were wrapping up an analyst summit in Chicago, another analyst who was supposed to share a Taxi with me to the airport forgot her phone and forcibly downloaded Uber on my phone.

And from that moment on, my life was changed.

At the time, I didn't have a driver's license yet so whenever I was deployed to a new project in a new city, I had previously always felt trapped. Compared to my peers who could rent a car, I was constrained to only being able to taxi from the airport to the hotel and vice versa.

I was never able to explore the rest of the city because I was afraid I would not be able to get a ride home if I got too far from the city center.

Case in point: if you were in the outer NYC boroughs pre-Uber, good luck finding any mode of transportation. The taxis exclusively circled downtown Manhattan.

But of course, this all changed after being introduced to Uber.

After that first encounter, I was so obsessed with the product (and saw such a clear user problem that it was solving) that I was lobbying the Canadian government to have Uber legally operate nationwide.

It was only after a few weekends rallying in the heat of summer that I realized that I could probably be more effective if I worked within Uber than outside of it.

At the time, I was in management consulting where I spent my days honing business fundamentals and cutting my teeth on heavy analytics on user segmentation and pricing strategy.

It was clear that Uber saw my analytical chops as useful and this was also at a time where Uber was growing at a pace of a 1 new city a day over the past 6 months and needed a centralization plan.

As a result of that, I was the first external hire on the Regional Ops team based in Santa Monica to develop that strategy.

The Santa Monica Uber office was a strange paradise.

The office itself was beautiful, modeled after the LA Soho club with intricate furnishings, decor and plants.

It felt like I was on vacation while I was at work and if there was ever a learning here for any startup founders (or office managers), how you set up the working environment of your employees really matters. Folks in the Santa Monica office came to work early and left late and many times came into the office on weekends just to be able to think and work in a peaceful and elegant office.

It was also a luxury not lost on me that we were located right across from Santa Monica beach with in-office surf boards so at the crack of dawn I took to starting the day catching beginner waves at the pier before the onslaught of tourists invaded the beach.

At work, I was given the freedom to experiment on rider promotions and driver incentives and pretty much run my region's P&L autonomously. It was insane the amount of responsibility I was given as a 23 year old.

Uber operates on a culture of "best data wins" so all individuals (technical or not) are required to be able to pull their own data. This is probably one of the smartest things any company can and I would argue was the secret sauce that helped it grow so fast. Every new employee had access to all the data old employees knew because they were given the tools to query the data.

The result of this is exponential growth unlock to empower their on-the-ground operators to make decisions fast, and also free up technical staff (engineers/data scientists) to focus on building as opposed to being pinged incessantly by ops on data pulls.

In my role, I owned our fraud process which meant automating queries to identify the first indications of fraud through individual risk vectors – connected phone numbers to other reported fraudulent accounts, % of trips chargebacked or unpaid, accounts linked to stolen credit cards.

I set up our first automated process at Uber to comb over high risk + materiality trips that had been flagged by the scripts I wrote to make sure the trips were legitimate. I wrote multiple internal research papers with recommendations on what is the distribution of “risk” for refunding when it came to customer support inbound that came bowtied with new refund decision matrix for CSRs. This was all while I was the point person running the monthly fraud P&L meetings and setting our regional operational strategy for US & Canada.

I was stretched out of my comfort zone, and I loved it.

The operations job was a good one, it challenged and satisfied me. For a lot of people, working at Uber in 2015 is a no-brainer - a high growth startup that's aiming to make transportation as reliable as running water - that demand side product market fit was unbelievably sticky.

But for me, I was always fascinated by the supply side. I grew up with my immigrant mother who spoke broken English and built our life stringing together a slew of contingent gigs - selling dumplings outside the grocery store, teaching mandarin at the community center, driving new immigrants landing at YVR (Vancouver airport) to their new lodgings, that I wondered what would happen in a world where you can systemize the gigs?

Back in the 2000s, her counterfactual job was not being a barista at Starbucks making minimum wage, it was nothing.

Our life was built on her sweat and hustle that she built up piece by piece like a Jenga Tower.

In my role in operations, a lot of what I was doing was driver facing. Heck, before the existence of the Uber regional operations team, all operations managers were called "DOps - Driver Operations." And every week, I would probably talk to 10-20ish drivers as we launched new incentives locally, or new features to test, or call up drivers stuck in our onboarding funnel.

And it became clear that as I was working in a company that intersected between real people and best in class software, for me to make the most impact to build the best gig marketplace for drivers is to not just launch products but to build products.

I wanted to create a broader impact not just from the business perspective and not just from short term operational fixes, I wanted to ship long term solutions, and I was ready to throw myself at the next challenge – I was ready to move to HQ.

The first meeting I attended when I moved to Uber HQ at 1455 Market St was a “war room” meeting on rolling out the pillars of driver forward – Uber’s ironclad effort to prove to ourselves and the world around us that we were "driver obsessed". Being slotted into the Driver Pricing team would mean that the product I was working on would be a very big part of this effort to regain trust with drivers.

I didn’t know anyone in the building, let alone anyone in the meeting I was attending – but leading the discussion was a woman with straight jet black hair, perfect posture, clear eyes, and golden hoops. She was actively listening, and while other folks around her were clicking away on their laptops, she was staring intently at the speaker, patiently waiting for him to finish his soliloquy while she has the data in hand to counter him.

She was the least senior person in the room, but from the way she held herself, you would have never been able to tell that. She was a blend of charismatic and empathic while being extremely articulate.

Throughout the meeting, she asked insightful questions and her position was to defend the plan on reducing earnings variance by reimagining our trip pricing levers to compensate drivers for “trip defects” – where the time of the driver’s time to pickup is longer than the actual trip itself, or times where Uber dispatches the driver from a high earnings area (that is surging) to a location that has a low earnings areas.

This made a lot of sense to me, but she was getting attacked left and right by colleagues who argued that this decoupling of rider pricing and driver pricing will unravel the core of what is sacrosanct of Uber’s marketplace.

She held herself with an air of patience as she rebutted each of her contenders – an engineering VP here, a data science director there, in such articulate fashion that I thought - "she should be running all of Marketplace at Uber."

To this day, a lot of the quirks I have are mere emulations of her and she set the path for me and many other female PMs at Uber to not be afraid to speak our minds and to really embody the Uber cultural value of #toestepping. (More on that later)

At that point, whatever doubts I had in my mind of moving up to San Francisco was quashed – I was going to join the product team on driver incentives and I was going to learn the skills of the trade.

The learning curve was steep- and I was eternally grateful for my rather haphazard decision back in college to be an economics research assistant.

Random forest experimentation, switchbacks, synthetic controls. Meaningless words to me but a few weeks ago are now my life.

Our driver incentives team, composed of some of the most decorated PhDs in computer and data science, were hellbent on the mission to intelligently and efficiently spend Uber’s $2B annual promotions and incentives budget in an automated fashion.

This was a noble mission, especially in such decentralized company where each city GM (there are 721 cities) owned his or her own budget.

In this highly matrixed org, there were many clashes between ops and tech where we (with our "perfected" machine-learning algorithm) would be put on a A/B test against the city GM's own methodology on what was the most efficient way to spend their incentive budget.

I knew I was the least technically decorated person on the team, and what I lacked in credentialing, I was going to make it up in good old fashion hustle. I memorized our tech stack front to back and read every engineering memo and RFCs that lived in our repository.

I did not want to be a liability when we had product discussions on implementation or feature extensibility when we were talking to other product teams.

The eventual payoff for all that effort was that I was the only person on the team that truly understood the end to end “incentive creation, generation, payment” work flow – and it was a lot less automated than we would have ever liked to have admitted.

For all our boasting to city GMs that we had a fully automated ML incentive generation system, we were:

1. not very good at adjusting for holidays or events (and when you have hundreds of cities on your systems, there’s always some holiday every day) because our algorithm assumes every week would behave the same as the previous week

2. do not take in as an input the competitive pressures (Lyft, Ola, Didi, 99, Carem, etc) and when they change their incentive schemes weekly, throws our prediction model into a tizzy

3. not very great at spending to budget because of the above competitive pressures, when budgets change drastically, our algorithm has a tough time spending to budget, so the projected overspend and underspend swung like high and low tide surf charts.

All that means although our tech leadership would like to believe we have an “intelligent” driver incentives system, it is far from automated. I logged 18 inbound requests daily from city teams and 36 manually interventions every week to override the system, changing offer allocations, weighing different times more with a hacky coefficient weighting so we can produce the end results the city GMs wanted.

The adage that city operations provides the inputs and tech spits out the outputs was rarely abided by and many weeks it was me who just manually intervened to produce the outputs.

Ultimately this led to an intervention I had with tech leadership after doing this for a few months. I felt burnt out, and I didn’t appreciate that they were going to sign our team up for another wave of city launches when they did not truly understand how much additional manual work that would cause the on call engineer and myself.

It is worth calling out- while engineers had the luxury of being on rotation while they are “on call”, PMs did not. Every week I was paired with a new engineer who would be tied to the hip with me as we battled through the different bugs we had from upstream service providers and work through the different business requests for manual interventions for the upcoming week’s incentive offers.

So in this intervention, I drew out our holistic incentive architecture and identified all the failure points and pressed that as a team we can’t ignore maintenance work.

We shouldn’t be chasing the next shiny object (our team was pivoting to building the next generation of session pricing and multi session pricing levers: Consecutive Trips (CT) and Build Your Own Quests (BYOQ) while it seemed that our current system was just being put to decay. This forced me to kick off an effort to pause new build as we went back to basics to build touchless experience of our current system.

There was a major lesson in this endeavor – to speak out when you feel burned out.

While I learned to speak out to my team, it was about this time in the spring of 2017 when another colleague learned to speak out. Her name was Susan Fowler.

The one word I could describe the whole situation was – flabbergasted.

As saddened as I was at what happened to her, it was very off-based from what my experience was as a woman working at Uber. I was also confused why netizens who didn't understand the full story were lashing out against my myself and coworkers, and was tongue tied as I tried to explain the whole situation to friends and family.

It was a really trying time - a time where many people I looked up to left the company, where I looked at our category position metrics have a sharp hockey stick curve in the opposite direction, where numerous friends started boycotting Uber and I was caught having to defend my company, my coworkers, Travis Kalanick and even myself at every social event I was at.

I felt like a tape recording repeating that Susan Fowler’s experience was not representative of every person at Uber, and especially as a minority female, never once had I felt discriminated against or sexually assaulted.

I know at the time, many of my female coworkers (myself included) took to social media to tell our stories and provide perspective on the other side to neutralize the somewhat hyperbolic narrative of Uber being a "bro culture" when our female to male employee ratio was pretty much on par with every other tech company in Silicon Valley.

While Uber was imploding internally, I was actually starting to really get into the swing of my job.

I was feeling good about tying up the loose threads of our incentive generation system thinking “Wow. Look how I'm managing, does everyone see I'm pulling this off??...." to my audience of.... no one.

I was unaware that another storm was brewing on the other end of driver pricing – one of our new products – Consecutive Trip Boosts (CTB) just got our user research usability test back, and it does not look pretty.

What happened was our flagship driver incentives- CTB - which at initial A/B test demonstrated to be highly efficient, causing our leadership to pump funding and resources to building it up, is highly inefficient.

The “efficiencies” of Consecutive Trips came from the fact that drivers didn’t get the incentive – it was too complex, so in the short term you have a lot of drivers trying it out (unsuccessfully) so Uber wouldn’t need to pay out the drivers on the trips they had driven while trying to complete the incentive. This gave this incentive a very low "cost per incremental trip". Basically we got these "incremental" trips from the driver for free (since they didn't complete the incentive but drive to try and get the incentive.)

But over time, you will find that drivers will stop trusting the product and switch to the competitor.

The report cut through the most basic elements of this product, which was for drivers to “drive 3 trips consecutively” at peak time and core geofences by only needing to start the first trip in an activation geo (usually downtown of the city) and having the freedom to drive the next 2 trips anywhere, but between the 3 trips not rejecting, cancelling or breaking the 3 trip series.

Sounds easy enough right? Wrong. We had drivers left and right in the research misquote how to achieve the incentive, a large portion of them thought all 3 trips needed to happen within the activation geo – seemingly “impossible to achieve”, or that POOL trips didn’t count (they did!), and other basic misinformation that forced us as a team to come into recognizing that we have over-indexed on efficiency and in the midst of it, created an unnecessarily complex product that made sense to PhD economists and computer scientists, but not to our everyday drivers.

Not to mention, localization was a nightmare- “consecutive trip boost” was a hard concept to translate in other languages, let alone describe in less than 2 sentences to increase glanceability for users.

We were in an all out design crisis.

Being on the team for more than two years, I had never truly interacted with our designers or UX researchers…ever. They were almost like oxygen in the room – I just took them for granted. My world revolved around the engineers and data scientists who “built” the system, I figured the app surface wireframes was a secondary concern. Boy was I wrong.

I attended multiple usability studies across Latin America, Europe and the Asia and was exposed first hand what good user research consisted of; part human psychology, part scientific documentation – it showed to me how new users would want to interact with our app in a perfect world – and every time they “mistakenly” swiped to a page that did not have the information they were looking for, or double tapped a map marker hoping something would happen and nothing happens, was not a mistake on their end – it was ours.

We let our users down. We did not take in consideration the use cases that they would like to use the products for, and we did not provide enough clarity to give assurance to the end user that what they intended to do actually resulted in the actions that they wanted.

It was eye opening – and ultimately the inflection point in my life that I understood what separates good PMs from great PMs is the user empathy you have to be able to transcribe user stories and problems into technical requirements.

But at the end of the day, what I learned the most from Uber is the idea that to build a world class business, you need to be moving forward. This means most times when we have imperfect data and context, the default is to not just wait for top-down orders but to experiment and learn.

Granted, you should spend the majority of your time hiring for the absolute best people who are smart, high-integrity and default to action, but once you hire them, you really should let them free to explore all the knooks and crannies of the business.

Don't gatekeep them with hierarchy, give them full autonomy to explore.

The mantra that is believed in Uber is that “perfect is the enemy of shipping.”

We live in a world where time is limited so we need to be intentional with how our team’s time is spent. Is the marginal benefit worth the marginal cost?

If you’re 85% of the way to delivering a product, the best use of resources, time and energy is not to perfect the remaining 15% in an echo chamber of approvals on a powerpoint deck, it’s to launch the product to a small submarket and A/B test the results.

Is there a probability that you break the market?

100%.

And is there a probability that once it fails, you are now in a better position to build an alternative solution that works better?

100%.

When you’re launching, testing and monitoring in real time - you have the luxury to roll it back ASAP when the metrics don’t go as planned and you don’t live in a world of “what-if-isms”.

You live in a world of data.

I’ve seen this at work multiple times in a working sessions where a senior VP might make a truism or a generalized statement like “let’s launch the next tranche of uberPOOL cities today” and a junior person would push back after crunching the numbers to say “no, we don’t have the density for those markets to be sustainable,” and instead of being ignored (or worse, chastised) by the VP, the more senior people in the room would give the green light to A/B test it across a few tier 3 cities to get real results vs any need to flex hierarchy or authority.

In a few week's time, if the results were contrary to what the VP predicted, s/he will be the first to acknowledge the results and pivot the launch plan.

This is the definition of building a culture of no ego.

It is a culture of meritocracy - best data wins.

There are many operations managers at Uber I know came up originally as customer support agents or data scientists who started off as a local marketing analyst in a satellite office - the opportunity to grow within Uber was endless. Heck, I started off as a lowly operations manager myself.

Even when the company grew to be over 20,000 employees, it ran lean and fast like a much smaller startup.

Our teams were intentionally set up to be autonomous units, with high authority to make decisions at the edge. Coordination and collaboration across teams was tricky, but we trusted each other, and made it work.

As the PM of driving pricing, once I align with my five regional operational leads for pricing across US/CAN, LATAM, India, EMEA, and APAC - the launch of a new feature is pretty much locked and loaded.

Sure, I need to give a head’s up to my peer PMs on other product teams, but generally the expectation is you own your swim lane and are responsible for its execution - success or failure.

Not only that, but the whole culture in every fiber of its being believes in the concept that all employees are #ownersnotrenters.

Uber imbues the idea that the next big idea can come from anyone in the company by giving everyone ample opportunity to solve problems regardless of their function or location.

The most debilitating place to be is stagnancy and at Uber, we eschewed a culture of seeking permission, we truly believed in our cultural values of #toestepping and #alwaysbehustling to not be a an environment of bro-culture, but instead an ecosystem where anyone, who has done the research and has the data, can influence any decision.

And it’s beautiful. In the 3 years at Uber, it felt like I got more done in a year than I had in the last two decades.

Travis often says, “Fear is the disease, hustle is the antidote.”

To this day, I live by this motto and will carry it with me to every new challenge I am given.