Meet Khilan Haria. He’s our Head of Product. We’d been grappling with the problem of helping customers discover teas they’d like and onboard with us on their journey to discovering their perfect tea. Khilan and our product engineering team were entrusted with the task of figuring a solution. Three months later, they launched an engine that actually achieves it. Happy with the outcome, the team has gone ahead and applied for a patent!

In a freewheeling, tea-fuelled chat, we asked Khilan how the grey cells went to work.

How did you get started, once you knew what the problem was?
I found a great parallel in my own experience with wine. When I started drinking wine, I was eager to try as many kinds as I could. I was a “wine explorer”, trying recommendations from friends and those who knew wine. But it was the rare wine that came recommended that I actually loved. What I did stumble upon was the discovery of Cabernet Sauvignon as my preferred wine. And then came the frustrating journey of finding a Cabernet Sauvignon I’d greatly enjoy.

I imagine the same complexity applies to the world of tea.

What was your own experience with discovering tea?
I have always been a big chai addict and generally prefer to start my day with a strong, high-caffeine cup of chai. In the world of specialty teas, I was a novice until as recently as 4-5 months ago. And while I am still discovering my perfect cup, our own prediction engine has definitely aided a lot to the process so far and the journey has been fun.

What underlines the design of the machine-learning technology?
The technology offers an objectivity that cannot be offered by a sommelier or an expert. We began by trying to match preferences at a group level but found that we could actually map preferences at the individual level. It’s a classic Data Sciences problem and we used it to create our prediction engine. At the core of it is its inherent objectivity.

Are you saying the technology can make a tea sommelier’s job redundant?
Like with wine, in specialty and premium tea, tea connoisseurs and sommeliers assume the mantle of expert and because they have a fair understanding of the subject, are the ones who choose and order teas. Sommeliers are subjective, base their recommendations on generic inputs.

But teas, like our tastes, are always changing. Every garden produces a tea that’s slightly different each season. And while you can start to expect a certain kind of taste, it’s never the exact same thing as the last year or even the last season.

Every individual’s tastes are unique and determined by too many factors. For instance, I like citrus flavours, an assortment of spices, a hint of bitterness, the smell of rain-soaked mud. I dislike floral and sweet smells and tastes. And yet, this is not set in stone. It’s constantly changing and evolving. I believe that’s the way it is for most of us.

How do you foresee this engine as revolutionising tea retailing?
It will make the discovery of one’s perfect tea fun for the novice drinker. There’s so much to try and discover and the process needn’t be frustrating.

What are our customers saying?
We went live in early July and the response has been phenomenal. Our confidence in the prediction engine is so high that we set out with an aggressive offer to ship a free replacement if a customer did not like the teas picked by our engine. Very few people have asked for this. And listen to what some of the comments have been: “Loved the teas. I love green teas and these were a perfect fit.” “Pleasantly happy with my first box!” “Can’t wait for our next month’s tea to arrive.” “Very impressive, especially for a first timer.”

Have you tried it?
I was constantly trying it as part of our User Acceptance tests. I love it, although that shouldn’t be a surprise. But I must add that it helped me uncover a few different teas. And when I responded with feedback on what I liked and disliked about the teas I tried, it got better and better at making the selections. It works! I doubt if a tea sommelier could have helped me with such accuracy in a short span of time.

Curious? You can try our new prediction engine too. Give it a go right here!

(Visited 1,096 times, 1 visits today)

Leave a Reply