The opportunity
In 2019, I was the first designer at Etsy to work on our Recommendations team, which had historically been Engineering-only. I observed a great deal of miscommunication, misaligned goals, and lack of shared language and understanding between our ML teams and our Product teams, so I set out to bridge this gap.
The solution
I identified key ML engineering partners who shared (or were amenable) to a more user-centric approach to ML-driven product development, created internal buy-in from leadership that this gap was worth bridging (namely ML eng managers and Directors, along with key Product Directors), and then worked closely with 1 ML engineer and 1 Product engineer to create a 2-part workshop:
Part 1 of this workshop was focused on introducing core ML concepts and Etsy-specific context to non-technical folks who worked on or adjacent to Product, namely product managers, product designers, and managers who were unfamiliar with ML. The primary goal of Part 1 was to broaden general understanding across as many teams within Etsy as possible.
Part 2 of this workshop was more granular and outlined a step-by-step collaboration process for Product teams working with ML engineers for the first time. The intent of Part 2 was to give Product teams who were interested in leveraging ML models more guidance around important technical and user-centric considerations, how to align on goals and metrics, and how to iterate collaboratively on ML models in a user-centric way.
Then, together with my primary ML engineering partner, we went on an internal roadshow across core organizations within Etsy over a series of 3-4 months.
The feedback
“I feel a lot more educated on how to talk about ML and how to understand it here at Etsy. You have a fantastic presentation style and broken down concepts so that they were easy to grasp.”
“Honestly this presentation helped me work through some of the tension I’ve felt in working with the ML teams. They don’t operate like Product Engineering, and it was a good reminder of how they work and of how much work goes into creating a model.”
“I just wanted to express how much I enjoyed your ML/Recs presentation yesterday! It was so well written / said and succinct. I shared it with my PM and we’re planning to reach out to the ML team to talk about some future possibilities. Thanks so much for sharing this knowledge.”
“This is really, REALLY good. It explains recommendations in a really easy to understand way. I shared the deck with my ML eng team to help them understand how Product folks view recommendations in the broader ML landscape.”
“This was excellent. So helpful, especially the slide with the “easiest to hardest” methods. :chefs-kiss: Also, making the distinction between personalized and non-personalized recs. Tons of great content, thanks so much for this.”
“That was great — the concept of using heuristics vs ML was really helpful. I’ve found myself wondering how ML could be applied to my work at Etsy... solving problems with ML is pretty “sexy” but now I realize we sometimes look for ways to do it when, actually, in a lot of cases, a heuristic approach might be cheaper and just as good. Thank you for the effective presentation!”