Discover more from Applied ML | Recommender systems
Joel Truher : the opportunities ahead
Applied ML #12
… second part of the conversation with Joel Truher, who until recently led the recommendations team for Google Shopping. He has held leadership roles in search and e-commerce in the past like CTO at Lycos and Chief Architect at shopping.com / Ebay Ads to name a few. Thanks to Joel for sharing his learnings and vision with us.
G: Joel, thanks for educating us about the e-commerce space in the previous article. You have been in the search and e-commerce space for a while. Could you please share how it started?
Joel: Well, I did stuff on the web from about the time when web was. For a while I worked on strategic defense. However, I was curious about the sustainability of unpopular ideas. I joined Mother Jones magazine… I was their computer guy. I was curious about what was happening elsewhere in tech. I learned about search. I worked on a search engine called Hotbot. I worked on the user experience and ads.
After the acquisition by Lycos I started looking at all aspects of search there. I ended up being the CTO of Lycos.
G: Lycos search was dominant then (around 1998), the most visited web portal I believe. What worked well?
Joel: That’s an interesting question actually. I can’t really put my finger on it, and I think that is one of the problems. We were doing a bunch of things but nobody really knew what was working well. There were debates about whether search itself would be some sort of a commodity. I think the reason that Lycos was popular was that it was early and investing in growth and quality. For a while Lycos had good market share with the users. However with Google’s growth things took a downturn.
(G: Readers who want to follow up on how to understand the growth of your service and engineer for future growth, I recommend this video (notes):
G: What could you have done differently at Lycos?
Joel: Looking back I feel Lycos had the opportunity to do things that were truly great. I should have pushed harder on improving the quality of the user experience and on scaling the search engine. Perhaps I failed to make a case to the CEO to make the investment that was needed for it. In hindsight, the investment was not that much! Perhaps my skill in making the case was low. Some of the traffic acquisitions like Tripod didn’t work out. Also, Lycos was a public company whereas Google was a private company.
G: You spent some time in Google around 2006. Can you tell me about Google in that period?
Joel: When I joined Google in 2006, it was an amazing place. There were lots of people trying out new ideas. Innovation was nurtured and even celebrated. As a company, Google was trying to make many small bets that had the potential to change the world. I was leading the team building Froogle, which morphed into Product Search and eventually Google Shopping.
G: What do you think about the role of different levers in e-commerce: Search vs Recommendations?
Joel: I think about it differently. I feel users have come to expect a long standing relationship with a platform. If you have been telling a platform like Google / Amazon about yourself then you expect the e-commerce platform to recommend better stuff. It is reasonable to expect this quality.
If you are asking what is more important between them? I think it’s neither. It is getting good product data. This is different from "user behavior" data, and figuring out the user problem they are looking to solve. The domain of product data, and the domain of "user problems to solve" are both super complex and hard to capture.
The product data problem is hard because there are so many "authors" of it and the best authors (manufacturers) have the least incentive to participate in a robust way.
Detecting the user-problem-to-solve is hard because there's no "author" at all, every user's state of knowledge of their own problem-space is different, and every exploration path is different. It is hard for the user to convey their problem to e-commerce platforms as they would to a human. If you are approaching consumer needs as a pure ranking problem , you are doomed, same for recommendation. Users’ needs are a lot more nuanced. Ecommerce platforms should be able to converse with people at the same level. There is an effort to dance around this by providing refinements by facets etc. These have been shockingly unpopular. Not because people don’t understand it… but it does not work very well.
E-commerce platforms have to understand the user in their language. The complexity increases because why users like something is not really known to them. “I like this item but I don’t know why I like it.” That's what Pinterest does well. It is able to identify what you like without you having to spell it out, but taking you through many engaging exploratory journeys and observing you.
The challenge ahead : How do you bridge the super nerd specs driven approach, and the super nuanced thing that Instagram or Pinterest do.
G: What are you excited about in the e-commerce space? What’s the opportunity that has not been tapped?
Joel: I am excited about the growth of direct to consumer (D2C), and it will be interesting to see how and which platforms figure out how to work with the new generation of D2C folks.
There is a possible symbiotic relationship there. It is not practical for every DTC brand to market to every user on their own, and I don’t see it as an either-or as well.
G: Can you share a little about what challenges you are passionate about these days?
Joel: I am passionate about empowering authentic voices and factually correct information. More generally, I am interested in bias and fairness in our digital conversation, about building accountability and responsibility in the system. The problem I see with our digital conversation today is that it is dominated by a few loud voices. Also, it tends to focus on larger issues and does not properly represent issues that affect a local / small community. I am interested in building power through information networks i.e. the countervailing force to provide normal people more agency.
Towards that end, I am trying to apply optimized media methods (which we covered in the previous article) to the problem of scale within community organizing for power, as conceived by folks like Saul Alinsky, or more recently, Hahrie Han (I highly recommend her books).
G: Thanks Joel. If people want to reach out to you, how should they?
Disclaimer: These are our personal opinions only. Any assumptions, opinions stated here are ours and not representative of our current or any prior employer(s).