Hi, everybody, my name is Samuel Jolibois. I work at Petit Bateau for more than seven years now as a technical project manager, and I would like to share with you our experience with RichRelevance(now Algonomy). So first of all I will introduce who is Petit Bateau and then how we implemented RichRelevance(now Algonomy) So far, and what are our projects?
So, first of all, who is Petit Bateau? It’s all brand, which was born in 1893 in France, and at first, it was selling only clothes for babies, and then kids, and now 80% of products are for babies and kids and 20% for adults. And that was around 400 stores globally and started its ecommerce presence in 2006. So how did the story with RichRelevance(now Algonomy) started. So at first in 2010, we use the recommendation system called Avail, which was bought by a RichRelevance. And in 2014, we have decided to migrate from Avail to RichRelevance(now Algonomy). So we’ve Avail, we got only one block on the Add to Cart page. And we’ve RichRelevance(now Algonomy), we extend, we extended the blocks everywhere on the website. So last year, in 2015, we started many projects. And we would like to extend RichRelevance(now Algonomy), not only on the website, but on over devices and over channels, such as emailing.
So first, first of all, how did we use RichRelevance(now Algonomy) until now, on the website, so we use RichRelevance(now Algonomy) on seven placements on the website. And as you may see, the results are very different from placements to another. So the best placement is the Add to Cart page. Because this page is very prominent through the checkout process. And the recommendation blocks are very visible. So it works pretty well. It works well on the category page as well, because it’s on the top of the page. And another interesting result is on the search page, when we have no results, we display a recommended block, and it works pretty well with 8% click through. Beyond the strategies used on the placements we have setup many rules or filters to exclude some products, for instance, we would like to exclude low value products below 10 euros. And for some special categories, for instance categories of products. With partnership, we don’t want to push recommended products because we would like to stay compliant with the with the partner.
And also we use a special rules during events. For instance, during the sale period. During the first two weeks, we would like to push product on sale. And during the three last weeks we prefer to push products from the new collection.
So what is the work in progress in add Petit Bateau with RichRelevance(now Algonomy). So we have decided last year to extend ritual events in emailing in mailing in a mobile app and use your projects or started using be wise. So first of all, I will present you how we have started to work on emailing.
So the idea is to personalize these engaging touchpoints because emailing represent one period of or turnover. So it’s very important. And we would like to make it more powerful. And another reason why we would like to do this is to leverage the emailing process because our merchandising teams spends a lot of time on selecting products to put on the newsletter. And sometimes when the customers received the newsletter, some products are out of stock. So we would like to avoid these issues.
So how did we proceed? First of all, we list to all emails which would benefit from RichRelevance(now Algonomy). So, I will show you two tables, one regarding the promotional emails, and another one regarding to transactional emails. So we have listed on the on the left side, all campaign names, and then we, we could target offline or online clients. We could also define if these clients are our belonging to the loyalty programme or not. And for each email, we would like to push one strategy or maybe several if it’s possible, and also some filters to exclude products. We don’t want to show up on emails. So we did the same exercise with transactional emails. Because we would like to avoid any dead end points on emails, especially with order cancellation or returning products, we would like to tell the customer okay, you didn’t get the product you want it but we can propose you over products you may like.
So, we would like as well to do the same thing with print mailing because at Petit Bateau We keep going and sending mails to our best customers around four or five nails barrier. And one idea we would like to explore is to for instance, to send a postcard to celebrate the birthday of the customer. And under on the front side of the postcard we would show the last product purchased offline or online and on the reverse side, we will use the recommendation to push some products you may like.
So, it would be more or less the same reason and same benefits as with emailing. So, in 2016, we also have the project of launching mobile app. So for Petit Bateau basically it will be very simple app to help our customers to check out their rewards if they belong to the loyalty programme and besides that, we would like to, to propose a special module on the under application to sell products as a second end, because as you know, Petit Bateau sells products for children. And when the clothes doesn’t fit the kids maybe the customers would like to sell products to over 30 Petit Bateau customers. So on this application, we would like to start very, very basically, at pushing products on the homepage based on the previous behavior on offline or online. And if the mobile app works well, maybe we could extend the recommendation within the user the user page, if the customers have filled in for instance some information about their kids, we would like to be able to push different product for the little girl and for the older boy for instance.
So now comes the biggest project we have started in 2015 using BYOS module, so the use case for Petit Bateau was to follow and to predict the purchase of a customer based on the age of his or her previous purchase. For instance, if a mother bought a pair of boots for a little boy size six months, six months later, we would like to push him over products, but this time in size 12 months and three months later, we will push products in size 18 months. So, the idea behind this is to have a recommendation more accurate for the final customer, because we got a lot of complaints about customer who said okay your recommendation is well, but when I went to the homepage, the product was not available in the size I would like. So, it could be disappointing and we would like to avoid this.
So, how did we proceed for doing visa use case. So, we started a kind of predictive recommendation process based on five steps. No four steps. The first one is the data collection. The second one is the data analysis. And the last one segmentation and defining rules.
So for doing this Petit Bateau has conducted a survey on 900,000 customers, both offline and online with at least one purchase during the last 18 months. So, you should know that online customers at Petit Bateau represent only 10% of our customers. So, it’s a huge benefit for Petit Bateau to inject inside RichRelevance(now Algonomy) offline data. So, the results of this survey are quite interesting. And we defined it defined five micro segments from best customer to sleepy clients based on activity, affinity with products and so on, and this is a total of 50 Micro segments and all these segments are identified by your segment ID and for each segment ID their survey was able to set up a separate section of products on SKU level and the initial segmentation script was built and developed in ss we can reuse it with other languages and we are starting to implement it in BYOS.
So, on this slide, I will present you focus on the size issue. So, the survey analyzed the customer behavior. So, online we can see the age of products both those so 0 to 3 months or three to six months and so on and on columns, you can see the date of purchase and this table was used to set up rules based on the ageing process for instance, three to six months after the first purchase of the product for a newborn, we have push product in sizes six to 12 months.
So, this is the this is the main die data we would like to introduce in BYOS is to do a personalized or recommendation and it started the first test started just a couple of days ago and maybe we can tell you more on next year if the results are fine. So which challenges do we have to take hold within these projects? So the first challenge is organic organizational, because for doing so, we had to build a cross functional team, including the marketing the merchandising, and data analyst as well as the technical team.
Another main challenge is cultural. Because we’ve been pretty battles, there are some resistance to the to the machine recommendation. So we have to initiate mind shift mindset shift to trust the recommendation produced by the machine, and latest challenges are technical, because our implementation at Petit Bateau is based on the reference colo level. And at that time, we didn’t get the recommendation on the size level. So this is a big technical challenge we had to face.
So this is a big technical challenge we had to face. So now what’s next for party by toe? So first of all, we would like to finish the projects started in 2015. So we hope to do this before the end of the first half of the year. And we also would like to improve our recommendation rules used on the on the website. So maybe an audit will be useful for doing this. And we plan to migrate that your size level in Rex in order to leverage the ageing process.
Another project is to use offline and online customer data everywhere in, in mobile in emailing in mailing. And we have to think about, which is the best recommendation does it based on offline data only or do we have to mix offline and online data, so we have to think about it more. And we would like to enrich our catalogue feed with new attributes. So an example could be stock level, like low level high level, so thanks to this attribute, we will be able to push high level high stock level attributes, products. And maybe we could add over attributes to it extend similarity, similar similarities or product recommendations. So we are thinking about colour, colour type, sleeves, sleeve attribute and, and many others. And the last thing is maybe to use the advanced merchandising to propose looks on the product page instead of simple recommendation. And maybe we would be using the Engage module to distil some pieces of contents throughout the pages. And that’s it is still a lot.
So thank you for your attention, and if you have any questions.
