Most retailers are stuck in the past, using obsolete forecasting systems to make decisions. But the models of ten years ago won’t get the job done today.
If you want success in today’s highly competitive retail market, you’ll need to utilize the benefits of big data and machine learning. The problem is that most retailers aren’t technically savvy enough to build these systems on their own. That’s where Evo comes in.
Simply put, we do things retailers struggle to do in-house without bias. Here we discuss four significant specialties we deliver to clients.
Needle in the haystack price signals
Price signals are vital in retail -- they tell companies where demand is and at what price level. But most retailers don’t have the modeling sophistication to identify small price signals buried within high sales volatility. In other words, they suffer from slow reaction times, bleeding cash while they wait for enough data to pile up.
Finding nascent price signals in high-sales-volatility environments is one of Evo’s specialties. Our software can identify price signals as small as a 5% average change even with sales volatility at 30% or higher.
We achieve such sensitive analysis using our proprietary library of models. These range from the ordinary pre- vs post-performance comparison, to our black box forecasting algorithms that compare real performance to a baseline hypothetical of no price changes.
We are also able to tell if variations in sales are due to price changes or other factors. Again, this is the result of studying multiple data streams, particularly stock-outs and changes product lifecycles.
The ultimate result is business intelligence in real time that gives retailers the agility to avoid losses and capture higher profits.
The impact of promotions in real-time
A/B tests, the retail standard for measuring the impact of promotions, are slow and expensive in terms of work, risk and time. With the advent of big data and machine learning, this standard has passed its sell-by date.
Our software automatically measures the impact of promotions in real-time without A/B tests. With Evo, retailers know right away when it’s time to nip an ineffective promotion in the bud or expand a profitable promotion.
Better yet, Evo helps forecast which promotions will work out in the first place. By measuring the results of historical promotions, as well as other relevant data, we can simulate the behavior of new promotions before they’re launched.
Accurately forecasting new-product performance
New products contribute around one-third of sales for the average retailer and up to 70% for industries like fashion and gaming. These figures are set to rise as product lifecycles get shorter and retailers continue to increase product assortment.
So the stakes in forecasting new-product performance are extremely high. If new-product forecasts are too low, retailers face a huge loss in sales; if they are too high, on the other hand, retailers are stuck holding a significant amount of unsellable inventory.
Traditionally, retailers used price and sales volumes to forecast future product performance. They predicted price elasticity and future demand using simple static models based on what happened last year.
The problem with this sort of analysis is that it takes very little data into account. Evo solves this problem by utilizing machine learning algorithms to backtest the impact of a large number of factors on sales volumes.
Evo’s analysis includes elements like seasonality, weather, traffic, competition, promotional intensity and brand loyalty, to name a few. In fact, such machine learning algorithms can incorporate up to 200 discrete factors!
The result is a much improved estimation of future sales at various price points. Moreover, Evo’s black box algorithms work in real time making dynamic pricing recommendations as any of these influential factors change.
Estimating probability of sales despite low volume
Just like with pre-launch products, it’s difficult to decide how to allocate inventory and set prices with newly introduced products. Sales volumes tend to be low for products just introduced to the market, which makes forecasting even harder.
Harder still is forecasting at the micro-level of individual stores. Stock-outs occur at the store-level and pricing varies among stores, so it is necessary to forecast the probability of sales and expected sales volumes at the store level.
Again, relying solely on past sales data won’t work in this situation. That’s why Evo employs a much wider set of factors than just price and sales volume to predict product performance. This allows us to estimate the probability of sales at the item, size or store level even when sales volumes are tiny.
Our bonus is no bias
Businesses tend to be optimistic about new products. This positive outlook often leads to inflated sales forecasts, especially when there are few data points to go on.
Evo’s technology removes the possibility of this human bias. We are able to go by the numbers because we have enough data to make accurate predictions. Our algorithms don’t change whether we like a product or not.
One of our greatest value propositions is the ability to give retailers better information in real time. Let us crunch the numbers and we’ll keep you a few steps ahead of the pack.
About the author
Will Freeman is a content expert at Evo.
He is a former economic journalist and part-time entrepreneur.
His interests include economic development, China, India, cryptocurrency and blockchain, and financial technology in general.