I’m already an investor in retail tech, so I’ve got a pretty keen eye for what to look for in the sector. Recently, I saw a firm whose product just blew me away.
The problem is that most non-grocery retailers don’t really know what their customers want to buy. Should they stock more red shirts or blue dresses? The answers to questions like these make or break the fortunes of firms in the fashion and homewares sectors.
These factors don’t only apply to retailers. Everyone who’s in the business of personal taste has to place bets. Currently, they do this on little more than a hunch about what’s going to be popular and by using vague “trend forecasts”. Record labels, homewares manufacturers and many other types of firm are all beholden to the whims of the public – but they have no reliable way to predict changing tastes.
But what if there was a scientific way of understanding what people will buy – right down to the individual products? A secret “mind control” formula that means you’ve always got just what the customer is looking for, when they come into your store?
The firm I’m talking with today has got just such a product. And I’m going to be stepping you through exactly why this is such a game-changer for its customers.
The firm is called SoundOut, founded by David Courtier-Dutton in 2007. It’s a UK firm, with global ambitions. Already claiming to be the world leader in prediction in the music industry, it’s now branching out into retail – and the initial signs are that a revolution in prediction is on the way.
(For transparency: some time after this interview was written, SoundOut agreed to promote its Seedrs crowdfunding using a marketing firm I’m financially linked to. As always, the production and scheduling of this interview was wholly independent of this deal. No money is ever taken from firms featured in Exponential Investor for our coverage. If you have any questions, please do get in touch.)
AL: What is your firm trying to do?
DCD: When you have no data, SoundOut predicts the future. It fuses advanced crowdsourcing, and data science, to predict how new products will perform when they become available to consumers. Our current clients include all the major record labels, and an increasing number of fashion and homeware retailers across the UK and US – as well as FMCG [fast moving consumer goods] manufacturers and distributors.
AL: What is the exact problem that SoundOut addresses?
DCD: When retailers are choosing completely new products, they’ve got no historic sales data to work with – because the product hasn’t been sold before. Their buyers and merchandisers rely heavily on gut and experience to estimate what volume to buy and where to set the price point.
They know that customer demand is principally driven by a combination of the appeal of the item and the perceived value. However, they can only estimate both of these factors. A combination of high appeal and low price is a recipe for a bestseller, while low appeal and high price will end up on the sale rail, or in landfill. Not only does this directly waste money, but it also means that cash is tied up on unprofitable stock – instead of being put to good use.
AL: How exactly do you pull off this feat?
DCD: SoundOut has its own panel of 2.5 million reviewers to precisely measure the appeal and optimum price point for any given item – be it a T-shirt, kettle or sofa. That in itself is gold dust to buyers and merchandisers. However, the panel is only half the story. SoundOut goes much further – by working on the data. We can backtest our methods, to show that they would have worked if our clients had deployed them before. We do this by testing just 50-100 items in any given category against historic sales data. Then, we can use our machine learning technologies to predict sales in this category for any retailer. Using just the proposed price, and the SoundOut test data, it is then possible to predict sales for any new item being considered. That’s a far more reliable process than buying on “gut feel” – and when the product might not be easily available mid-season, running out can be an expensive mistake.
SoundOut can typically predict with over 80% accuracy which will be the strongest and weakest sellers for a retailer. Furthermore, we can also estimate the likely sales for each product. This enables a buyer to drop bad products – and reinvest the money to buy lots of the more promising products. Using the SoundOut models, the buyer can also simulate different price points – predicting the level of demand at each price. This enables them to understand what pricing strategy will deliver maximum overall profit, for any given product or range.
AL: That sounds useful, but is this really a significant development?
DCD: Absolutely – because no one has ever done this at scale before! It’s an approach that has near-universal application across retail, and related industries. For retail it could be a real breakthrough moment. Everyone knows there is a link between perceived value, product appeal and price. However, nobody has previously been able to measure this relationship accurately enough to build customised prediction models, using machine learning.
AL: So is this data-driven approach the future of retail?
DCD: It’s certainly a part of the future.
The current model is: take a best guess at what the customer might want – then spend the next three to six months trying to minimise the impact of all your buying errors.
The future will be: only buy items you know you can sell, price them right, and buy just enough to satisfy demand.
SoundOut’s recommendations aren’t perfect, as factors out of our control will always play a part. Things like the weather, competitor activity, celebrity wardrobes, etc, all have an effect. But now retailers can have a quantitative measure of value and appeal – and they can accurately forecast demand for every item in a range before they buy. Why would anyone not want that?
AL: Will we see numerous companies appear in this space?
DCD: It’s unlikely – because there are three significant barriers to entry.
Firstly, SoundOut has over 2.5 million reviewers, with over 2,000 new signups every day. This gives us the ability to generate 100,000s of reviews a week in a completely controlled environment.
Secondly, the methodology behind how we aggregate, weight and calculate appeal, etc, is protected by a US patent.
Thirdly, we’ve also got the data science skills to statistically normalise and weight every opinion and rating we get. It’s far from a simple exercise – we’ve spent years honing our systems. Currently, we refine what we’re doing by testing around 10,000 items per month. We now operate at a scale, speed, accuracy and cost that few could hope to replicate.
We also think we run the firm pretty well. It’s a very lean organisation, with just 14 employees. We’ve always worked on the assumption that we can only make big profits if the team stays small and focused. The way to make this approach work is to relentlessly automate. We let the crowd, and the technology, do all the heavy lifting.
AL: Is there a risk that competitors can join your panel and pinch the designs being tested?
DCD: That’s always a risk of market research – but the chances of this are slim. Over the years we’ve tested lots of pre-release music from famous artists such as One Direction, Lady Gaga and Katy Perry. But because the reviewers don’t choose what they’re listening to, and can’t tell who the artist is, they rarely realise that it is the real artist. We often get comments like “You sound a bit like One Direction – you need to find your own sound” and “It’s like a bad Lady Gaga”, etc.
With 10,000 songs and designs running through our system each month, the chance of any individual reviewer finding and copying something sensitive is very remote. Add to that the fact that none of the reviewers know how well the item is being rated, and you end up with a very good level of protection for the client.
AL: How do you compare to competitors in this space?
DCD: Companies such as Dunnhumby, Aimia, 84.51° and even Kantar have had great success finding patterns in existing data sets and powering personalised marketing campaigns – but SoundOut is targeting an altogether bigger problem. For the first time, we’d like to think we’re offing savvy retailers access to a genuine “crystal ball” approach to predicting consumer demand. The only company in the same space we are aware of is a US business called First Insight who have recently raised $14m venture capital. Their methodology for collecting consumer insight is very different from ours – as they do not own their own reviewer panel. Furthermore, they are not making use of artificial intelligence and machine learning, but the customer proposition is similar.
AL: How easy is it to sell this product?
DCD: SoundOut is disruptive – so the sale is typically made at board level. Our platform is used daily to help buyers make decisions, so management must be 100% satisfied that it does what it says on the tin. As a result, the typical sales cycle is six to nine months. Part of this process is for us to prove beyond doubt that we add significant value, by consistently identifying the bestsellers and dogs across multiple product lines. However, once we get customers on board, we normally see very rapid take-up internally.
Going forward we will increasingly sell through channel partners, such as retail consultants and other retail technology businesses. We already have two such relationships in place, and they are bringing in new clients on a regular basis. Right now, we have a strong sales pipeline.
AL: Finally, can you please tell me about your funding – past and present?
DCD: We’ve raised over £7m from over 70 friends and family investors to get the business to where it is now. Right now, we’re currently looking for a further £2m.
So, do you think that SoundOut represents a good investment? I certainly think that it’s a great technology play, and it also benefits from significant barriers to entry. However, the big challenge is to selling to a whole range of customers quickly and cheaply. If it can streamline its sales process, and get some good channel partners, I think it’ll do really well. However, I’d love to hear what you think, so please do send your feedback to firstname.lastname@example.org.