This week we will be discussing how you can get started with Product Recommendations.

A lot of websites already use product recommendations, having seen what Amazon and other large retailers are doing, but have never really thought about what else can be done with them and the power they can have.

This blog is for anyone who would like to know a little bit about how it works and how to get started.

Examples of Product Recommendations:

  • Here’s what you may be interested in: based on your viewing/purchase habits
  • Other people who viewed this product also viewed: based on others’ viewing/purchase habits
  • Other similar items: associating products with each other based on their attributes
  • What you previously viewed: your direct historical activity
Product Recommendation examples

Why use it?

Product recommendations act almost like another form of navigation. They guide the user to showcase other products they may have not viewed or even considered before. There is also the more overt case of it being a tool to upsell or cross sell other products.

Know what you’re getting yourself into:

You could copy the ‘out of the box’ examples detailed above, which will enable you to get started straight away. However, is this beneficial for your users and your site?

For example, are all product recommendations good, have you thought about the pros and cons of having them on your site could do for your business?

Are your ‘out of the box’ product recommendations showing cheaper options, meaning AOV is going down despite sales potentially increasing? And, what does this do to your bottom line? Are you causing confusing for customers, with too many options causing to customers abandoning their purchasing decision altogether?

To start your journey into Recommendations, you really need to consider what behaviour your trying to encourage. Understand what your main KPIs are and realise this will always need constant review and monitoring. Do you have the product data and sales feeds to keep your product recommendations current and up to date? For example, if your using a sales feed with data from the last 6 months, what was selling at Christmas time may not work in the summer months.

Top Tips for getting going:

  1. Identify a problem that you believe a solution such as product recommendations will help the segment of customers who are being affected.
  2. Work out how you will feed your recommendations, what feed will you use? If your selling products with different stock levels how up to date is it? Is this feed setup automated or manual?
  3. Work out what the logic will be, these will be the rules for your algorithm.

For example, you may decide: anyone that is currently viewing a sports rucksack will be shown, 4 other rucksacks, within the same price range as the currently viewed product but must be a different colour. You therefore need to make sure your feed currently contains the following attributes:

Product Category: Sports Bag

Product Type: Rucksack

Price: £x.xx

Colour:

  1. Know the logic for your negative cases – what to show if there are not 4 products to retrieve back within your product recommendation, and so on and so forth. It may feel like you are opening a can of worms here, but the clearer your logic is the less hiccups you will have at the implementation stage.
  2. Build out the logic and test all scenarios to make sure it works.
  3. Make sure you are able to monitor and track your results.
  4. This final comment shouldn’t be last point to think about but, you will need to consider the design, placement and content of your product recommendation.

What Next?

Once you’re happy with your product recommendations they can be set live and then monitored.

What can I do with it once it’s set live?

This is just the start! This is when you can evolve your product recommendations. You don’t just have to showcase one type of product recommendation on your site ether.

  • You could showcase ‘what’s new’ for new users
  • Spend an extra £x.xx to get free delivery
  • Similar products with the same characteristics
  • Top reviewed items in this category
  • Here are other ‘red trainers’ that you haven’t yet viewed.

Not forgetting, to test

  • Placement of your product recommendations
  • What page should they be featured on?
  • What location on the page?
  • Should they be shown after a specific behaviour?
  • What content should be within each recommendation (Picture, CTA, Price, Review, Colour, Rating).

The list is endless!

Should I consider Machine Learning?

Machine Learning or Neural Networks are great for inferring associations within data sets. The right systems have the ability to pick up on trends that you might either not know, or not have the time to drill down into. There are, however, several challenges associated with doing this:

  • Learning Time: It can take large amounts of data to draw accurate connections with Machine Learning. Sometimes, you should expect weeks or even months of experimentation before “the machine” begins to generate reliable predictions/recommendations.
  • Misfires: What is the impact of recommending the wrong thing? How closely can you control what “the machine” selects as a match?
  • Black-boxes: The best algorithms are the hardest to explain – often known as Black-box algorithms. If you’re unable to explain what you’re showing, you’re not gaining any knowledge about what your users like. Whilst the additional revenue is great, the real gold is in data.
  • Skillset: Adopting AI technologies often require strong expertise in engineering, dev ops and data science – skills which are not always easy to acquire.
  • Cost: Servers to run AI models can be amongst the most costly machines you run – the depth of calculation and performance required are often extremely resource-hungry, and this means needing powerful machines to scale your service well.

None of this is to say you shouldn’t adopt an ML-based approach – at it’s best they’re incredibly powerful and low-maintenance ways of generating additional revenue. It just often takes a lot of work to get to that place.

If you would like to discuss Recommendation Engines in more detail or talk through what would be feasible on your website, please get in touch!