This blog will investigate the shortcomings of typical mobile commerce sites, in the context of specialist apps such as Spotify, Instagram, Netflix and Pinterest.
At the heart of these apps is a smoother user experience coupled with an AI that can really flourish in that environment.
Here’s a video to explain:
Training your AI
AI’s aren’t born smart, they have to learn. You do this by taking a sample of known data, and asking the AI to predict ‘what is this’ or ‘what does this mean’? You then compare this predicted outcome to the known outcome and see how close you get. The process is then repeated with different samples of the same data set to see how close you get.
To paraphrase a much more in-depth blog from Roel Decoene, this AI model needs three attributes to succeed:
- A good dataset to train the AI, which comprehensively captures the variation in the phenomenon you are trying to capture. Lots of variations in your data mean you need a big dataset.
- A set of algorithms where the performance has demonstrably been shown to scale with data—if new data pops up, your model needs to be able to handle it
- The training data needs to reflect what’s happening in the real world—to bring it back to Qubit, our AI has been trained on fashion and beauty specific data sets, to name just two.
The cats and dogs example serves us well, as the differences in cats and dogs are known, but there is plenty of variation between types of cat and dog. Therefore, it’s clear to see that an AI model would need a load of data to ‘get smart’.
In this article, WIRED report that Instagram has been able to predict what’s in a photo (cats, dogs… and stuff like wheels and sunsets) with 85.4% accuracy. The size of the dataset? 3.5 billion public Instagram photos—carrying 17,000 hashtags. Do you have enough data to train your AI?
Allowing your AI to flex its muscles on mobile
The human condition on mobile is to browse with social media-like efficiency. We scroll, tap, like, we share, and we do it at pace. One thing we don’t like: slow page loads.
For your AI to show off its smarts, it needs a more free flowing environment. Sometimes this is the case on mobile commerce sites (on a category page for example), but it’s when you get on a product page that things start to slow down.
Qubit Aura tackles this problem by reflowing your site into an instagram-like feed, with a twist: Much of the information you might on a product or category feed is stripped out, such as price, size and reviews to make the browsing experience as seamless as possible. It’s designed to show the maximum amount of products with the minimum cognitive load.
Is this relevant for ecommerce?
Consider this: Accord to this article, over a third of Instagram users have used their mobile to purchase a product online—making them 70% more likely to do so than non-instagram users. This instagram generation are used to living on mobile with easy going, free flowing interfaces, and being relevant to them matters.
If you’d like a look at how some of Qubit’s clients are conquering this challenge, take a look at Qubit Aura. Super smart AI, trained on a tonne of data, with a super cool UX to boot.