A Product Recommendations Framework – Where Does My Team Even Start??

You’ve been tasked by your Executive Leadership Team to increase orders and average order value (AOV) over the next 12 months. You have a solid plan in place, as you’ve worked on it for months. Yet there’s something nagging at you, like there could be something important missing in your strategy and you’re not quite sure what is lacking?

You take a quick work break to clear your mind. You remembered you need to order dog toothpaste on Amazon. And then….BAM!!! It hits you right between the eyes. Did you really need to buy the dog toothbrush and bully stick with the toothpaste? No, but you did!! Then you take a closer look at their site and give it some more thought. Product recommendations are everywhere. Not only that, Amazon (and many other sites) are leveraging various algorithms and personalization techniques as well. Then you think to yourself, this is it, this is what I am missing in my strategy!

How do you even get started here? What guidance can you give your Director of Optimization to kick-start this initiative and make sure your team gets off on the right foot? Let’s dive into that now.

My goal here is share my experience working with product recommendations with clients over the past decade. And in turn, I hope you are able to take something away to help you achieve some of your goals.

Phase I – Data Strategy

If you’ve read my post on Freeing up your Data Teams so they can Provide Actionable Business Insights, you won’t be surprised to hear that the first thing to do is task your Director of Optimization to have the team dig into the data. You will want the team to focus on these three aspects of the data, shown below. As you can see, there is a combination of quantitative and qualitative data here. The qualitative part will require interviews with key business leaders who can answer the questions for your team.

💡Opportunity sizing – quantitative

💡Primary KPIs – quantitative

💡Page functionality – quantitative and qualitative

OPPORTUNITY SIZING

 Opportunity sizing is fairly straightforward, and this data is instrumental in driving prioritization discussions when the time comes to map out a schedule of releases. The goal is to determine how many visitors will have the opportunity to see your product recommendations placement(s) throughout your site. You may be surprised at what the data shows you.

For example, when working with a client recently we were all surprised to learn the percentage of visitors that hit a 404 error page on the site. Further, the error page was not a great user experience in that there was really no where for the visitor to go but to either hit the back button in the browser or exit the site. What a perfect spot to place product recommendations and keep the visitor engaged on your site. And even better if you can personalize those product recommendations!

I recommend going beyond using only Visits (% of traffic) for the sizing opportunity. You can also look at metrics such as page conversion rate and exit rates. These additional metrics will give you more insight into the potential product recommendations placements on the site. For example, you may find that two of the site pages have nearly identical shares of overall site traffic. However, one of the pages may have a much higher exit rate as compared to the other page. Data and insights like these will make prioritization much easier. You’ll soon be on your way to developing a long term product recommendations roadmap.

The last part of the data strategy is both quantitative and qualitative in nature. This is where you will need to put some strategic thinking time into the following questions:

❓Why are visitors coming to this page?

❓What do the majority of visitors do when they are on this page?

❓What is missing on this page?

❓What clutter can be removed from this page?

Having insight into why visitors are on a certain page will help you in the Algorithm Strategy phase of this framework. For example, it seems obvious but visitors land on a search results page because they are looking for something specific. However, what happens if the visitor can’t find anything helpful related to the search term, or if that visitor mistypes the keyword? Most of us know this is called a null search results page. But, do you know how often this happens on your site? Further, what is the experience for the visitor when this does happen? Are they given any other options to easily continue looking for items? Or are they given a dead end?  

Analyzing all of your key site pages like above will quickly help you identify low hanging fruit for product recommendations placements. Once you have a list of prioritized pages, you can now start thinking about what products to actually show on those pages, and how those products are determined.

Phase II – Algorithm Strategy

Now that the team has identified and prioritized pages to place product recommendations on your site, it’s time to start planning out the algorithm strategy. It only takes a few seconds on Amazon’s site to see various strategies in place. There are many to choose from, but not every strategy is the right one for the visitor or page.

I like to break down the strategy into (1) known visitors and (2) unknown visitors. Have your team dig into not only what attributes are collected on the site for known visitors, but also what is not being collected and should be collected.

KNOWN VISITORS

You may not always have a large percentage of visitors that are known on your site. But when you do have that information, it is very powerful in helping craft a positive experience. Aim towards showing visitors more of what they want without making them do any additional work.

Here are some questions to help prime the pump along with a suggested algorithm.

QUESTION

What is the visitor’s favorite category?

Which item(s) did they purchase recently?

Do they have a most viewed item?

Does the visitor have nay items in the cart from a previous visit?

ALGORITHM

Most popular in this category

Bought this bought that

Similar products

Email with similar product recommendations

We won’t get into this topic today, but there is a ton more to talk about with regards to third party data and customer data platforms and the integration with product recommendations.

Here’s an example of product recommendations for a known visitor. There are a few different algorithms on this row of the home page. “Pick up where you left off” is using Recently Viewed and so is “Keep shopping for.” “Buy Again” is showing items that I have purchased over the past several months, making it very simple for me to repurchase these items. Have your team think of similar strategies that are applicable for your known site visitors.

AMAZON KNOWN VISITOR PRODUCT RECOMMENDATIONS

UNKNOWN VISITORS

 It can be a bit more difficult to personalize product recommendations for unknown visitors. You don’t have past purchase and browsing data. However, this should not preclude your team from exploring this area. Your team can leverage what they know about the visitor during the current visit. This includes what product(s) they are viewing and adding to cart, as well what they are searching for on the site.

It could also include products that are added to Saved lists if that functionality is on your site. Finally, some clients that I work with have visitors that purchase different products depending upon where they are located. For example, visitors in cold weather states are more likely to view and purchase skiing equipment as compared to someone from a warmer weather state.

Here’s an example from Amazon when I visit the site as an unknown visitor. The product recommendations contain top products in both Movies & TV as well as Content Creators. Your team can do something similar with popular categories on your site.

AMAZON UNKNOWN VISITOR PRODUCT RECOMMENDATIONS

Lastly, here are some questions to help prime the pump for your team when thinking about how to implement product recommendations on the site for unknown visitors:

QUESTION

What are top selling products in their area?

What other products are bought with this product?

Can we change content within the visit based on what unknowns are browsing?

What products are similar to the one being viewed?

What are the highest margin categories?

What are the highest volume categories?

ALGORITHM

Top sellers with geo-targeting

Bought this bought that

Custom profile script

Viewed this viewed that

Best sellers

Most viewed

Phase III – Measurement Strategy

Almost there! Before launching your product recommendation placements on your site, you will want to think through and document a measurement strategy. I’ve been in too many executive level meetings where questions were asked about the usefulness of the data from product recommendations. This typically comes down to a measurement strategy issue.

Here’s an example. A product recommendations carousel was placed on an item page for a larger retail ecommerce company. The recommendations algorithm was spot on and the design was beautiful. Further, a really cool A/B test was run testing algorithms against each other. The data was gathering and the insights were developed. During the executive readout, one VP asked a few simple questions:

“Do these results capture visitors that viewed, clicked and bought the product that was recommended to them in the carousel? Or, did we capture visitors that bought a similar product after clicking a product recommendation?”

There was a brief moment of silence in the room. Sadly, the team had to answer ‘no’ to that question, and offer to dig into the data more and get back to this group.

Here are some questions to help guide your team as they carefully think through the measurement strategy:

❓Do we want to measure just products that are clicked AND bought?

❓Do we want to tie product recommendation IMPRESSIONS to other products bought by that visitor showing influence?

❓Are we interested in knowing what products get the most/least clicks? Purchases?

❓How engaged are visitors with each product placement in the carousel? Do the 1st and 2nd products get the majority of the clicks? If so, how would this influence future designs and placements on the site?

So, I think we have covered a lot of material here. I would even say each section could be broken down further. I wanted to give you high level guidance that you can discuss with your Director of Optimization. One last note. Check out what other sites are doing with product recommendations, and not just Amazon! There are so many good (and bad!) examples of strategies and designs that your team can leverage to help you achieve your growth goals. Hopefully this information was helpful, and good luck on your continued journey of improving your customer experience as well as helping your company achieve its goals!