Forecasting demand for products and positioning inventory has never been a simple task for retailers, but it has become increasingly difficult in the digital age when shoppers’ purchasing behaviour has become varied and unpredictable.
Supply chain management has gained importance for retailers as the ways in which consumer purchases can now be fulfilled have grown in complexity through services such as home delivery, and click and collect.
Gartner supply chain research director Tom Enright says it is now time for retailers to change their approach to demand forecasting to keep pace with the new industry dynamics and predict better what influences consumers to make a purchase.
Planning well in modern retail requires a fundamental shift in thinking for businesses that have effectively operated in the same way for decades, he says. The traditional method has been to use historical data for statistical forecasting, but that may no longer be sufficient.
“The new way of working is to forecast at the granular level – how consumers will buy rather than simply what they buy,” says Enright.
“That will allow a retailer to say ‘I can see the direction my walk-in shopper is looking like, I can see what is going to happen with click and collect and I’ll see what I may need to do in terms of ship from store’.”
There are other fulfilment methods to consider too, such as ship to locker, dropship, and reserve online.
Enright adds: “The quality of demand forecasting is only as good as your ability to recreate the environment in which historical sales took place.
“What that means is there is a need to have more inputs into the forecasting solution, rather than just historical sales, inventory and price.”
What the public or celebrities are saying about products on social media should be factored in, he suggests, as should information about competitor pricing, the weather and particular shipping offers. For instance, the weather has, and always will, influence fashion sales, while celebrity diet trends will inevitably have an impact on the popularity of certain food ranges.
The thought process is that if this information is analysed, processed and built into the demand forecasting strategies, it will help to shape future business decisions.
Enright says: “What you’re really trying to do with demand forecasting is understand why consumers bought something and what influenced them – and if you understand the influence, you can shape demand and start to understand if any of these influences can be controlled to generate further demand.”
All retailers face their own challenges, but it is difficult to imagine any retailer not wanting to increase efficiencies within their business through better forecasting.
Andrew Rafferty, IT and e-commerce director at UK grocery chain Booths, says: “Even for a company of our size [Booths has 28 stores], ordering just a little bit better and increasing availability or reducing waste by just a shade can make a real difference.
“But it never can be perfect with the variety of ranges Booths offers.”
Towards the end of 2013, Booths realised there were gaps in the company’s demand forecasting capabilities, so it partnered with Finnish technology firm Relex to improve its replenishment forecasting and other supply chain management processes.
Rafferty describes Relex’s technology as utilising “impressive memory processing tech”, adding: “Even for a company of our modest size, there are various permutations to consider. This investment represented our major leap forward.”
The grocer applied the technology over different supply chains and has seen a “definite return on investment” since implementation, he adds.
For Intersport, a sports equipment retailer with about 300 shops in the UK and Ireland, the subject of predicting demand is pertinent. The business is mainly focused on serving small town communities through face-to-face in-store interaction, but it launched e-commerce last year and now completes close to 10% of its sales online.
Laetitia Kotsiopoulos, Associated Independent Stores
Laetitia Kotsiopoulos, head of supply chain and IT at Intersport’s parent group, Associated Independent Stores, says a recent move to stock products in a central warehouse – as opposed to allowing brands to deliver straight to individual stores – has dictated new stock management processes.
“We are switching to an always-available model – some retailers call this never out of stock – and for this, you need to be able to monitor your sales in store,” says Kotsiopoulos.
“But Intersport is a franchise model and we don’t have access to sales information from every store. We are getting there – a new electronic point-of-sale system is being put in place – but it will take some time until we have access to all data.”
Kotsiopoulos adds: “Sometimes so-called ‘killer products’ don’t sell. We conduct the best forecasts we can with the brands, but we end up with stock we can’t get rid of, and we have to mark them down after barely two months in order to ensure we don’t keep a massive level of stock in the warehouse.”
Intersport’s ultimate goal, once it has completed its IT systems roll-out and is in a position to take greater control of its forecasting, is to mount a better challenge to the bigger players in the market, such as JD Sports and Sports Direct.
For this to happen, says Kotsiopoulos, the retailer’s systems need to ensure there is enough stock in its warehouse to fulfil repetitive orders and allow for new product ranges to arrive on time in each store, so there are no delays in launching the latest fashions.
The situation at Booths and Intersport – which, respectively, represent the small and medium-sized retail communities – shows how stock management processes are viewed as fundamental to retailers’ success.
But what must it be like for the big grocers, food-on-the-go players and fashion houses that operate hundreds of sites and/or sizeable omni-channel operations?
Machine learning matters
That is where machine learning should be considered, says Enright, who argues that it can help retailers to find links that traditional statistical forecasting processes would not be looking for.
Some of the UK’s bigger retail names are at the early stages of using this sophisticated technology.
Dave Potts, CEO of the UK’s fourth-largest supermarket, Morrisons, recently announced the introduction of a new cloud-based automated ordering system as the retailer’s “biggest new initiative”.
Using Blue Yonder’s Replenishment Optimisation system, which incorporates machine learning technology, Morrisons is crunching complex store-specific data to make accurate ordering decisions. Launched in 2016 across the retailer’s 491 stores, the system automates more than 13 million ordering decisions a day, based on vast amounts of internal data and external information such as weather forecasts and public holidays.
The grocer says it has seen the benefits of optimising ambient and long-life product replenishment, and has since been exploring how this technology can assist the management of its fresh food.
Read more about retail technology
- Intu has partnered with L Marks to launch a retail startup incubator programme to help advance UK retail and cater to modern consumer habits.
- Consumers are now more trusting of other customers and product reviews than they are of advice given by retailers.
- Online retailer Made.com has partnered with Hero so showroom staff can chat live with online shoppers to help with queries and shopping decisions.
Other tech suppliers touting their artificial intelligence and machine learning capability in retail include IBM, Microsoft and Ocado Technology, but on many occasions this technology is being applied to customer segmentation and e-commerce performance. Either way, the retail industry finds itself at the start of a new wave of disruption and opportunity because of such sophisticated tech development.
Costa Express, the unmanned coffee machine arm of Whitbread’s Costa Coffee chain, was ahead of the game when, in 2011, it started using integrated telemetry to provide real-time reporting on the performance of its machines and drink sales. A year later, the system was enhanced so that real-time data was used for automated replenishment.
Using software from ToolsGroup, data from the unmanned coffee stations is now used to forecast demand, optimise inventory and generate replenishment proposals for Costa Express’ distribution and procurement operations.
What was once a time-consuming manual process has been automated and embedded into central supply chain operations, providing various cost savings.
Booths’ Rafferty can see the wider retail industry benefits of using machine learning for predicting demand in separate sales channels in the way Gartner’s Enright recommends.
“When you are looking at the amount of data you can get on your customer online – behaviour trends that you don’t have in bricks and mortar – I can see how the channel selection could become more and more crucial,” he says.
“In the grocery world, online is still around 6% of sales, but in fashion, where stores are often just showrooms to support e-commerce, I can see this being much more crucial.”
Deloitte predicts that retailers’ operating margins will drop by 3-5% in 2017. Combined with more specific customer demand around product availability, rising property and staff costs and the social pressures of reducing waste, retailers are looking for efficient systems and working methods to grow their businesses.
For that reason, better stock accuracy from the point of sourcing right through to the point of sale is becoming ever more critical, and there are so many factors to consider when forecasting demand.
Enright believes retailers’ success in this field will rest on finding ways of systematically monitoring all variables and creating the historical environment a consumer was in when they decided whether to buy or not.
It might be the weather, competitor positioning, pricing, shipping and returns offers or customer reviews that had an influence, while certain fulfilment methods will be more lucrative than others. All this information can be taken into account, measured and used in future planning decisions.
“When you think about the amount of data that would involve, today’s world of price inventory and historical sales looks pretty primitive – and that’s because it really hasn’t changed at all in the best part of a decade or more,” says Enright. “You now have to approach demand forecasting at a more granular level.”
How will AI for demand forecasting affect its accuracy?