If we want to improve the world of e-commerce, we need to start at the beginning. We need to start at the way products are categorized.
15 second preview: One of the key problems in e-commerce is a failure to deliver contextual product & usage information to the user, leading to a lack of conversion. In this 3-part series, I use the example of buying a pair of headphones to discuss the underlying problem of product curation. I then explore machine learning as a possible solution to this problem.
I’m setting out to buy a new pair of headphones. My requirements: I should be able to run with them outside and they should offer good sound for playing my electric piano. I go to Amazon and type in “headphones”: over 100,000 products available. In an attempt to reduce this overwhelming amount, I check out the filter options:
As you can see, Amazon offers me some helpful, clear options. I just want to look for headphones, so should I go for Audio Headphones? Over-Ear Headphones? In-Ear Headphones? Audio Headphones seems the most straight-forward filter offered here, so I decided to go for that. Great, now only over 10,000 products are available! (How did I filter out about 90,000 products with the most superficial filter thinkable?) What remains is selecting a headphone suitable for both running and playing the piano. Some new headphone filters are available:
I select “Sports & Exercise” and “Wireless” — but what about my piano sound? I let that one go for now; to begin with, let’s see what this search offers me. The first viable product I come across is the “Active Noise Cancelling Bluetooth Headphones Wireless Canceling Microphone Low Bass Response APTX Hi-Fi Audio Over Ear Protein Ear Pads Foldable Hard Case Travel Work Computer TV Sports Steel Black”-headphone. Catchy.
So, would this headphone really be suitable for running and playing the piano? Let’s check the product page. I see some product images and a short description, followed by two sliders featuring related items (sponsored and non-sponsored). Next are a bunch of hard product specs from which I can infer the best use — but only if I have enough knowledge of what these specs mean and which specs I need.
If e-commerce shops actually want to help us shop, they need to change the kind of product information that’s available. Specs should be there, but only in a supplementary way. First and foremost, a shop should be able to show me whether a product is suitable for me — and why that’s (not) the case. After all, that’s the answer I’m looking for when shopping online.
The difficulty here is that such information — of course — differs per person. I need headphones for running and piano playing; my mom needs them for audiobooks and plane rides. That’s why I believe that we need to move away from functional product specs and towards what I’ll call “soft specs”.
You can think of soft specs as things you can’t directly measure in a product: you can directly measure a headphone’s weight, but not whether it’s suitable for running. One way to determine soft specs is to combine certain hard specs; another is to use expert knowledge based on experience. In fact, soft specs could be a key reason you sometimes still visit a “real-world” store, to ask questions like “is this headphone suitable for running?”. In this way, a soft spec if very much like a use case.
“Is this webshop suitable for offering me soft specs?”
So, we’ve determined that webshops need to make the translation from expert knowledge to soft specs in order to offer users helpful filters to quickly find the product they need. But how to go about this? Often, it comes down to a three-part process called product curation.
First: gathering the knowledge needed to create and categorize soft specs. This could be done by having a product expert designing a certain rule, e.g. “if a headphone is wireless and less than 200 gram, then it’s suitable for running”. However, this requires perfect rules that precisely select the products you want to select — and I’d question whether perfect rules exist for every kind of soft spec.
Second: spreadsheets, spreadsheets, spreadsheets… The second step isn’t a fun one. The newly made soft specs need to be saved, to know which product belongs to which soft spec(s). So: let’s update the database! 😄 Unfortunately, most product updating still happens manually, in the form of adding an extra feature for each product in the database. So, for the audio headphones on Amazon, we would have to categorize over 10,000 products manually… Nobody wants to do that.
Third: happy users! If you have decided on useful soft specs, users should find it easy to quickly find the right product for them. In my case, I’d want to filter on ‘suitable for running’ and ‘suitable for piano-sound’, and find the perfect headphone in only two steps.
The problem is clear. If we want to add soft specs to a product we have to come up with perfect rules that select precisely the product we want, and we have to update our database — usually manually.
So, how can we fix this problem? How can we make product curation easier and more efficient, in order to ultimately improve the online sales journey?
Two words: machine learning.
In this 3-part blog series I describe how I solved this problem, what my steps in thinking were and how I designed a viable solution:
- Introduction: The Problem of Product Curation (you are here)
- E-Commerce & Machine Learning
- Opening the Black Box of Machine Learning: let’s see what’s happening
If you enjoyed this article, please hit those little 👏 below (you can “clap” multiple times!). Want to know more about our curation system? Find us at meetfeli.com.
