Being able to predict what will sell (and avoid what won’t) is the dream of all buyers. If you can avoid stocking the poor sellers – then you can also cut down on waste, and that means profits will increase.
Over the past few years there has been considerable excitement over the potential of big data for retailers, and there have been great advances in demand forecasting by extrapolating today’s sales and using this to predict what will sell tomorrow (and move stock to the right stores accordingly). Add a sprinkle of data science and you can even account for the weather, day of the week it is and whether it’s just after payday. Companies such as Blue Yonder excel in this space but according to David Courtier-Dutton, CEO of SoundOut, this is too little, too late; they are simply reducing the negative impact of the original buying mistakes.
Courtier-Dutton believes the elephant in the room remains the inability of buyers to predict what will and won’t sell – if they got that right they could transform their bottom line profits. Zara is perhaps the closest to achieving this via its ‘just in time’ approach to production where actual sales in store trigger in season production runs – and as a result Zara’s net margins are almost double that of the competition.
But a new age may be dawning – and it is coming from a surprising place – the music industry…
“For a number of years the industry has been pre-testing songs with online panels of millions of reviewers to ensure they release the strongest singles for their artists. This is very similar to creating, for example, a fashion range – making sure the best sellers make it into production and the dogs don’t. As the market leader in music testing we’re now moving into retail and seeing some dramatic results.” says Courtier-Dutton of SoundOut.
SoundOut, based in Reading, has a ‘hit’ rate of over 80% when predicting where a new track will peak in the charts. And the 2million strong online reviewer base are just as good at predicting retail. Recent testing with a major accessories distributor saw the reviewers correctly categorising best and worst sellers with over 80% accuracy, leading to a dramatic reduction in unsold stock over a six month period.
But assuming retailers can optimize their range selection, buyers are still left with the decision of where to price each item and how deep to buy. This is the true holy grail of retailing.
“To address this challenge, at SoundOut we pulled in a team of data scientists – PhD and Masters students from the University of London and have, for the past nine months, been applying advanced machine learning technologies to train predictive models on historic data for a major UK retailer. By feeding in SoundOut review data, reviewer price expectations, the actual retail price and historic sales data for groups of items, they trained prediction models to forecast how many units of any new style would sell at any given price. Once the models are trained, the retailer simply tests samples with the SoundOut review panel (a three day process), and the prediction model then tells them what and how deep to buy.” explains Courtier-Dutton.
The system periodically retrains itself as new sales data becomes available and gets smarter and smarter over time. It is still early days, but the model works as well with lamps and bedding as it does with fashion. Even the earliest models show a 15% improvement in forecasting accuracy – and this is just scratching the surface of what will be possible.
People who write Christmas lists are rarely disappointed on Christmas day – similarly asking your customers what they want before you buy it for them beats guessing what they might like.