Good AI knows what you smell like

Article by Natalie Cramp, first featured on DataIQ.com .

What does a decent digital customer experience currently look like? For example, if your browsing or previous buying behaviour indicates you like polkadot skirts, are you offered a polkadot dress? I mean, who doesn’t want to copy Minnie Mouse’s fashion sense?

Maybe a more advanced algorithm can recommend a range of complimentary tops to go with the skirts you like? Many engines currently used by retailers do a passable job of inspiring customers to make purchases. However, if passable was good enough six months ago, it certainly isn’t now.

The unrelenting movement from bricks and mortar to online has been turbocharged by the pandemic. Buying behaviour may have permanently changed and retailers face perhaps the most challenging environment on record. Enhancing the online experience should be a priority and one area where there is a lot of room for improvement is within “try before you buy”.

Subjective, one-off, or complex purchases require something a bit special.

Many recommendation engines, especially those used within the fashion industry, only use the most basic data science techniques. They are generally fed by limited demographic data, past purchasing behaviour and, of course, browsing data. This is usually enough to get recommendations within the general ballpark.

However, what happens if you’re a perfume, wine, make up or car retailer? Exceptionally subjective, one-off, or complex purchases require something a bit special. This is where a highly-engaging customer journey underpinned by machine learning can play a huge role.

So how do we approach this challenge? Well, I can speak to this with first-hand knowledge. We were tasked by a major beauty company to build a bespoke recommendation engine for perfume. In my opinion, this is one of the most highly personal choices an individual can make – what smell do you like or, more specifically, what do you want to smell like?

Crucially, how do you convince the consumer that a recommendation they are viewing online is going to smell the way they want it to? We’ve all had the well-meaning friend or family member buy us something which is may be the same brand as what we wear. Then we put it on and smell like we’ve rolled in a bath of aftershave – there seems no rhyme or reason why one thing smells good on us and the other doesn’t.

The most predictive question for perfume is whether you believe in the death penalty.

If you or others are going to spend significant money on it, you want to be pretty sure of delight not disgust. That’s where data science comes in.

As with everything, we need to start with the data. This challenge requires more than just basic demographic, purchasing and marketing data. We really need to know who it is we are talking to and what they really like. Our preferred method was the direct approach. We used machine learning to formulate 18 simple questions that could be used to build clusters.

Some marketers may consider this strategy sacrilege. Too often, brands seem to believe that consumers need to be duped into providing their personal data. This is obviously not the case. If you can show your customer that letting go of some personal information provides a clear benefit, they are often more than happy to do so.

In this particular example, the questions that were formulated did throw up a few surprises. Unsurprisingly, some didn’t make it into the final customer facing version, eg, that the most predictive question for perfume or aftershave preference whether or not you believe in the death penalty.

The answers to these questions allowed us to cluster the population into twelve personas, define rating rules to quantify how much a customer likes a product, and to identify key ingredient combinations. Thereafter, six algorithms were created to work in symbiosis to power the recommendation engine.

Of course, data science is really only half the battle. The reality is that you can make the most accurate recommendations in the world, but if you can’t convince the consumer these are accurate, compelling or even inspiring, then they are unlikely really to help a business. As a result, working hand-in-hand with the marketing function every step of the way is essential to generate a holistic customer experience. Everything from the website or app copy, payment system to the UI needs to be tailored towards both the product and the customer.

There is an economic incentive to create experiential stores.

This is really the crux of the matter. If businesses want to differentiate their online offering, or even supplement their in-store experience, they need to ensure that they a) use data science to get the deep customer insights they need, and b) integrate and use these insights into all parts of their business, especially sales and marketing.

A recommendation engine is also only one of many customer experiences data science can underpin. The best retailers will undertake a two-pronged approach. They will make the online experience mimic the best aspects of in-store – specifically, it’s personal and visual appeal – while also making their in-store experience more engaging.

Creating experiential stores in person has been talked about for a number of years. However, now there is a pressing economic incentive to do so both in-person and online due to the pandemic. Achieving this goal ultimately requires retailers to go beyond the basics of data analysis and become real innovators who are willing to take chances on cutting edge data science techniques.

It’s clear that if we can use maths to understand what smells people like and then encourage them to buy perfume, we can, with a little imagination, tackle many of the other challenges facing the retail industry.

[mr_rating_form]
Comments
(0)

View more recent postings

+ Add comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Loading...