Any e-commerce app can easily benefit from crowd knowledge. With the possibility of streaming vast amounts of behavioral data to the cloud, we as data scientists continuously see new approaches to utilize this precious information within innovative products and services. One, at least in the realm of mobile apps, so far rarely explored approach is unique product offerings which many know from established e-commerce web platforms such as Amazon, Zalando, or eBay.

fashiongram recommended items

While large teams usually have the capabilities to employ their own data science crew for the aforementioned scenario, we have made it our mission to offer a generic service to take over this task for you and to merge this it into our marketing stack. To exemplify how this works, this blog post will feature individual steps utilizing data from the shopping app FashionGram which is regularly fed with popular fashion from Instagram. The setup is straightforward: the SDK is implemented within the app to forward customized events to our servers where we apply collaborative filtering. We embed the resulting insights within our messaging framework and target highly profitable users in an automated fashion by sending tailored and personalized content. So let’s take a look at this in practice and taking one step at a time.

FashionGram makes first contact with segments of users who have a healthy engagement within the app in the first week after the install. To interact with each of these users, push notifications are used which are filled with individualized content.'s User Interface for Push Notifications

Within the target field, one is able to utilize the results from our recommender system which comprises precomputed items with a high likelihood of matching an individual user’s taste. As mentioned earlier, we embed a collaborative filtering approach to find similar items or even categories of products. The results are ordered according to their matching score which is described by means of the Pearson correlation.

Ranking of Recommended Items

The next question to be answered is at what time of the day one should get in touch with the customers. A full time employee will most likely feature an activity pattern in the mornings and evenings, whereas students are typically using their smartphone during the day as well. Once again, we can statistically derive this information from a user’s activity history. The histogram features a bar chart of 24 hours according to each user’s hourly engagement translated to UTC time. By simply clicking the “perfect moment” button, our system will automatically detect the optimal time over the next 24 hours where the user is most susceptible for engagement. For example, the graph shown below features a user quite active in the mornings and evenings:

User Activity Histogram

In this post, we wanted to show you, how you can benefit from your own crowd’s data and utilize this knowledge in our marketing stack to cover key questions in regards to when to interact with users and what to send them. Up next, we will be showing you how this can be automated for whole segments of users with the help of’s smart tags.

If you can’t wait to read more about our upcoming features, sign up for our newsletter or follow us on Twitter, and we will notify you in time. In the meantime, here are some further posts you might be interested in: