The Adelean team is back from the Mix Camp for E-Commerce Search 2023 and shares their takeaways.
MICES stands for MIx Camp E-commerce Search. It is an independent event aimed at product designers, developers, UX designers to share their experience and inspire each other. This single track conference is built around a theme that changes every year, and mixes conferences and an open barcamp. It takes place in the historical Osram Hofe former lightbulb fabric in Berlin, Germany, today hosting TUECHTIG, an inclusive coworking space.
Throughout the years, the organizers have succeeded to create a small, friendly community of passionate members from the biggest e-commerce platforms in the world. We all know and help each other to provide the best search experience. MICES is a very important moment for us in the year.
For more information about the agenda and the edition that is just over: Visit MICES website.
We saw 6 talks, from 30 minutes to one hour, followed by a final panel of experts. The main theme this year was UX for e-commerce search. The quality of each talk was really high. Here are our takeaways, for each of them:
Online tracking does not capture so many things: only the “what” but not the rest. And when customers are buying online, they do it with less senses than offline: no touching, no feeling… After exploring some techniques to improve exploration, i.e. bringing products that would not be normally ranked that high in the system to the view of customers, Andreas’message is the following: don’t only rely on data to improve your revenue. If you have in-house retail knowledge, use it. This point of view full of good sense is rather opposite to the current all-data, all-personalized search experience but shows that relevance in the retail business can also reach greatness with human knowledge.
Otto folks focus on what they did to improve the search before the first click: inside the search bar. They show us various changes they did on their suggestions panel, measuring the impact on order count (using A/B test) and the level of effort. Among others: playing with the space bar, taking in account the cursor moving inside the query, or spell checking. Also they did reranking in suggestions regarding current user’s interest and personalization, powered by machine learning. A full bunch of ideas, and an inspiring “iterate and test” protocol.
Digitec Galaxus is a couple of big e-commerce shops in Switzerland. They have two search teams representing more than 15 people. Their current challenge: make the filter suggestions smarter. With another nice methodology using A/B tests (like many of today’s stories actually), they could iterate from a quick and dirty MVP to taking a lot of data into account and recalculate the options automatically.
This talk was more data oriented, about a problem encountered for every search project that uses machine learning: the cold start problem. What can you do when you add to your e-commerce catalog new products that have no click, no history? With the help of an A/B test strategy, Evgeniia tells us her journey to trying different methods until finding a way to fill the missing past interactions by predicting order and click counts using similarity with existing data. Linking to the first talk of the day, she agrees with the fact that exploration is about these new products (they must be shown to the customers).
Search experts at eBay focused on an important, sometimes forgotten, aspect of search relevancy: spelling the query correctly. They estimate the misspellings in the search bar to 10% of overall search requests. It directly impacts the null and low search results. They began addressing this issue in 2014 following the state-of-the-art methods available, and today they are able to power it up using LLMs (BART) and “automated” fine tuning.
This session is about a dilemma for e-commerce search UX and marketing: how to personnalize the UX to the maximum and be the most relevant possible this way, but also being ethical, human, and respecting the user’s privacy. Marlety explains how they created persona the most diverse possible and match their users to these persona. This is a challenge that shows contradictions, and a promise that will probably seduce a lot of e-commerce platforms product managers.
This final panel brought together three experts to discuss perspectives on e-commerce search. We had Angel Maldonado, CEO of Empathy and advocate of ethical commerce, followed by Aparna Sundar, consumer behavior specialist and UX Research Lead at Opensearch, and René Kriegler, e-commerce director at OpenSource Connections and one of the organizers of MICES, added his thoughts and expertise. A great moment that summarized it all, each one on their own and specific point of view.
The barcamp is a very special moment to conclude MICES: an open forum where everyone can bring their questions to the whiteboard and discuss them in smaller groups. This is when people find solutions, contribute, exchange ideas in a more informal way. Among other themes for this edition, UX for the Search Result Page, Vector Search and Offline testing were discussed.
We always enjoy going to MICES to share with our Search fellows. This year’s theme was insightful, because it resonates with what we explain to our customers everyday. We are concerned by the fact that search is not only data, for us “Search Engine Administrator” is an actual domain of expertise. We also embrace the current trend to use and exploit new technologies such as LLMs to reach the finest exploitation of data. We will remember to take care of exploration over exploitation by adding “salt” to our search results. And always keep balanced between respect of privacy and full personalization; a theme we also had the chance to explore last year at MICES.
We would like to thank our friend René Kriegler, Sebastian Russ and the whole MICES team for bringing such energy and commitment to our e-commerce search specialists community. See you next year!