Q&A with Luís Trindade, Principal Product Manager of Experimentation
With over thirteen years of founding start-ups, mentoring, building ProductTank Lisbon, and leading data-driven product development, Luís Trindade has a wealth of knowledge that he’ll be sharing live on May 20th at Optimizely’s Test and Learn, a virtual summit with leading experts on progressive delivery and experimentation. In his session, Luís will explain how to instrument product experiments with faster and better results by identifying the metrics for product success.
Recently, Luís and I discussed the outlook for designer fashion amid the global pandemic. To preview his session for the conference, we talked about how to align product priorities and move fast.
Perri Bronson, Optimizely: How is everything going in your world? How has COVID-19 changed things for your business?
Luís Trindade, Farfetch: We are building the global platform for luxury. Traditionally, a sector dominated by brick and mortar, you can imagine the impact store closures and tourism has on our network of boutiques. As an e-commerce marketplace connecting consumers to designer inventory online, we are in a strong position. We are helping boutiques who have closed their doors to sell online. In many cases, creating a digital revenue stream for these high-end designer shops is critical to their survival. Everybody is adapting, and operating out of necessity is making people innovate.
Perri: I’ve heard that the ‘data behind the scenes’ is your competitive differentiator at Farfetch. How did data become a focal point for how your platform is architected?
Luís: We are bridging the online and offline worlds to revolutionize the customer experience, and the only way to do this is by connecting the user and product information through the entire journey. Data is essential for making better business decisions. Without it, we would be gambling with the future of the product based on guesswork.
But data doesn’t just inform our strategy. It’s the foundation of our products. For example, data powers our recommendation engine (Inspire), and search and ranking systems, visual merchandising tooling, and pricing recommendations, just to name a few. This focus on a data-driven strategy and platform laid the perfect foundation for a great experimentation culture from the beginning.
We gain insights from both quantitative (behavioral tracking and metrics) and qualitative (user research and interviews) sources to generate hypotheses for how we can learn what our users want. By testing the hypotheses as experiments, we attain faster learnings that lead to product iterations that result in more meaningful development cycles and features with higher potential to impact success. It’s all about testing assumptions and making decisions based on the facts.
Perri: Everyone wants to go faster without losing precision. How do you achieve this?
Luís: In short, we fail fast and learn faster. Experimentation is not all about winning, it helps you quickly identify what does not work, and this is an incredibly powerful loss prevention tool. By accelerating insights, you make better decisions, including where you should focus your product development efforts. Taking a test and learn approach allows you to have more space to evaluate bolder hypotheses in a controlled environment. Safe risk-taking leads to more potential for greater rewards.
Your bolder ideas are also the ones that converge sooner, so you will be able to control them better so long as you have accurate information. Experimentation is a science of probabilities, and this means ranges of data, rather than one, final number. For example, there is a confidence interval in any test metric. A finance department prefers one “final” set of numbers. But the only way to build a reliable product recommendation in such a variable situation is to consider a more dynamic data set.
At Farfetch, we begin estimation on our initiative investments already accounting for that uncertainty. And then it’s about how we share all our testing results. We present the expected outcomes as ranges of probabilities, and we make decisions only when the results indicate a higher certainty of success. Next, we outline estimations for business cases generated by intervals of the estimated financial impact. Then we launch all tests using holdback groups rather than taking the sum of the cumulative results, which would include natural overlaps of concurrent tests. This final run of experiments is our victory lap at the end of each quarter, where we validate the cumulative impact on the business.
While this is not an easy task indeed, it’s the most accurate method we’ve got. It allows us to move quickly on high-risk, high-reward opportunities, and discern the impact from a more thorough perspective.
Want to learn more proven experimentation practices from Luís? Tune in to his talk from the May 20th Test & Learn, the Product Experimentation Virtual Summit.