When I first got to Walgreens I was instructed to work on showing personalized products on the category pages.  In the end, I was able to reframe the problem, and learned how a basic algo can work better than a complex one.

🔹 Sitewide (ecommerce) revenue per session up 3.4% ($ millions/yr)
🔹 96% confidence based on A/B testing software
🔹 Two sprints of discovery work
🔹 One sprint of development

From management's perspective, the problem was a lack of personalization on the e-commerce pages of the site. Most e-commerce sites were implementing some form of it, and it seemed to be what shoppers online wanted.

After the initial briefing, my first plan was to test product carousels with personalized content on the category pages by creating a new area for the carousels to reside in:

We needed an algorithm that would use existing data to make recommendations. To communicate my proposed algorithm visually with the development team, I made the following graphic:

During the process, I had my doubts that any content on the category pages could be successfully personalized. Not to drop the "A" bomb, but even Amazon did not appear to try and personalize their category pages, and they personalize everything.

Still, I completed a handoff as desired by stakeholders, and used the remaining time and budget to pivot.

Why rely on complicated personalization algorithms when we have room for simple-yet-targeted solutions? 

On the product pages (as opposed to the category pages) we know more about what the user wants, so even a simple algorithm can be powerful.

Like this carousel that pulls similar products with high ratings:

Here is where reality struck, as I was unable(in my first month on the job) to convince the visual team, fresh from the creation of a new style guide, to back the cutesy mini carousel. This also would have needed logic to handle the concept of 'highly rated'. Still too complicated.

At this point in the process, I noticed that the carousels currently being used on the product pages had some room for improvement. For instance here is a carousel found on the page for a hair coloring product:

For a user who is looking at hair color, the products populated in this carousel might seem random.

So my proposal was simple: let's show more products from the same product type the user is looking at currently.

The design I ended up delivering really is not much of a design at all. It's only a product carousel that contains the top 15 products ( by views ) from the same category as the product which the user is looking at. Yet it's also not difficult to see why it tested well across several metrics.  

 For the user looking at hair color, which carousel appears more relevant?:

This project was a textbook example of everything I was taught in grad school, about deconstructing the ask, reframing the problem, and combing through potential solves without blinders on.