Check it out here: https://skifinder.aussieskier.com
As a passionate online, but reluctant physical retailer, one of my main observations has always been about scale. When we do a really good job of writing a product description or filming a YouTube review, we do it once and it goes out to thousands or tens of thousands of people.
But when we have a quality one-on-one interaction with a customer in-person or over the phone, all that effort only gets to be used once - and it's a shame, because in the context of helping customers choose skis, I think we're really good at it, and I've often been looking for ways to 'bottle it'.
So we're clearly in the AI era, I've been using it heavily for all sorts of things, and the thought came to me in the shower, as all good thoughts do, maybe training an AI is how we scale our one-on-one interactions to a wider audience.
Now we haven't recorded our in-store or phone recommendations, but inside our email ticket system we have thousands of conversations that have resulted in hundreds of ski purchases.
So I decided to see how I could leverage this data in an AI world to help inform our online customers, and really distill down our philosophies for how we help our customers select their next pair of skis - and in terms of training AI this turned out to be a data goldmine.
I didn't want to upload our data directly into AI for privacy reasons, so I used AI to write a script to filter through a bulk export of our tickets to identify the particular conversations that were recommending products, and filter out things like shipping/tracking and other spam. The script also pulled in order data to see which conversations resulted in a ski purchase, and finally anonymised this data so it could be uploaded into AI for analysis & form the basis of training data for the tool.
This churned away overnight and came back with a consolidation of insights that really helped understand what factors informed our ski selection process and at this point I had some clarity that I could build an interactive ski finding tool.
Now at this point I need to stress that while I am a very technical person I don't know how to code. At all.
I've been using AI to write Shopify front end code and some Python code to do backend tasks on our website but I was definitely going to need to take the next step to write a full app.
We decided that we should take the AI insights on how we recommend skis and create a ski ranking algorithm, which we then combined with our own overall philosophies on how we assist in ski selection, the interplay between height, weight, skill, style & location - and after much tweaking we came up with a version that mimics our personal process.
Once a skier enters their details, the algorithm takes 9 key factors - height, weight, age, gender, skill level, preferred terrain, skiing style, turn shape preference, and location - to create a personalised ski ranking before our AI delivers the final recommendations with detailed explanations.
It also queries our online store for the ranked skis, extracts all the specifications - we are absolutely meticulous with our metadata which informs the 'Specifications' tab for all the Skis listed on our site - and passes them on to AI for inclusion.
But it doesn't stop there, I also realised that our YouTube reviews are a strong source of our direct opinions on skis, and given YouTube auto-creates text captions for all our videos, we built a connector from the tool to see if we have a review video for that model, and if so it incorporates our review wording into the data sent to the AI for consideration.
The end result is a recommendation of 3 skis with an explanation that incorporates the skier's attributes and intentions, the specs of the ranked skis & how they relate back to the skier, the tone & method is informed by our ticket training data and any insight from our YouTube reviews is also included, with online store previews, pricing, recommended length and of course a link.
But while the tech is amazing and we love the recommendations we see when we add in different skier types, we still think that the human touch is as necessary as ever - the tool can't cover all eventualities, and it's well known that AI can go off-script from time to time, so we're more than happy to still add the human touch.
So after the skis are recommended you can add any more information you see relevant and press the button down below to request an expert consultation. All the data will be forwarded to our customer service system and one of our staff will come back via email with a bespoke recommendation.
Try it out: https://skifinder.aussieskier.com
We've really enjoyed creating the tool and we think it is going to be another step towards being the top online retailer of Skis in Australia.