Working out how to set and amend prices for products – to ensure that they are optimally balanced – has always been a tall order. It’s been a time-consuming process which hasn’t been easy to scale up to the boom of the ecommerce world. With ecommerce everything moves faster, and identifying optimal product prices using old methods is impractical and cumbersome, and will be outdated the moment it is set.
The good news is that the very tools (automation, the internet, and smart technologies) which have driven the speed and momentum of ecommerce, can also help solve the problems. As supply chains gradually incorporate more artificial intelligence (AI) and machine learning in their processes, they can better utilise the data available which in turn can feed through automated processes to set optimal prices.
Price Optimisation – Past
Previously the ‘luxury’ of price optimisation was fairly restricted to certain industries with relatively limited inventory. For example, old-school price optimisation was generally a fairly simple process in hospitality, car manufacturing and sales, and airlines. The act itself was complex, but the relatively low inventory made it possible to complete. It also required human experience and intuition governing insight in to competitor analysis. It required pulling information from different sources manually. But it was possible on this scale.
The Power of Data
This type of price optimisation becomes near impossible for industries with much greater levels of inventory. The required competitor analysis is impossible on a human scale.
Due to technological advances, however, we are now gaining data insights in to every facet of the supply chain, and customer demand, that we simply couldn’t benefit from before. Everything from AI to machine learning to computer speed, all comes together to provide us data – and the means to use it – on a totally different scale.
This has led to some fascinating research from MIT, who have succeeded in developing a precise model and method for setting optimal prices when there is a broad inventory base – in real time – and on a constantly updating basis. In fact, this machine learning algorithm, developed by Simchi-Levi, is thought to be such a breakthrough that it won the INFORMS Revenue Management and Pricing Section Practice Award.
MIT Price Optimisation
MIT have conducted trials looking at three ecommerce businesses: Rue La La, Groupon, and B2W Digital. In each example, MIT discovered that they could “increase each retailer’s revenue, market share, and profits for selected products by double digits.” That’s incredibly powerful and something that all businesses could benefit from. Bizarrely, it’s not actually what they set out to do. Their goal was to establish a way of reducing inventory. In fact, they found a way to more effectively optimise prices instead.
This model of price optimisation requires a three-step process:
- Improved Forecasting: The process involves identifying a ‘cluster’ of products with similar factors and product characteristics. With this, they involve a machine learning technique called a regression tree using if-then statements in order to create a prediction. Within this we see pricing principles of old, such as looking at historical sales data, but adding a technological twist which isn’t possible by human hand. The machine algorithms can predict up to 20 if-then statements which in turn predict the interaction between demand and price.
- Learn: The next step in the process is to conduct test runs of the prices matched against actual sales. Here, the actual results are used to amend the machine generated pricing curve. This stage is one of refinement.
- Optimise: Lastly, the new price curve is applied across a multitude of products (with similar sales characteristics) thereby optimising the price across the board and across time periods.
It should be noted that not every step needs to be implemented all of the time. They give examples of Rue La La Inc. who didn’t want to alter prices during a sales period, so they missed out the learning step for the purpose of consistency.
Does This Method Always Work?
MIT are keen to point out that “innovation isn’t only a matter of inventing a new tool; it’s also about people using the tool.” It’s important to note that humans are not removed from the equation of price optimisation, but in fact are integral to making the 3-step process actually work in practice.
At the outset, there is a huge need for change management. Managers and supply chain leaders need to pave the way for the approach to be successful, and to create the right environment for it to flourish. Data is only ever as useful as the actions taken. This means that supply chains and their leaders need to be willing to: analyse such information as competitor data (and therefore in many ways be reactive, rather than the proactive approach they are used to); break down the different silos so that there is true collaboration, particularly in terms of data; and, get specialist data analysers on board.
This is inherently difficult as generally people are resistant to change – especially if they feel jobs or roles will therefore become superfluous. However, it is a tool, not a means of replacing people. This is particularly important to realise when viewed within the context that in the vast majority of businesses it is just a select, and relatively small, range of products which drive the business, and indeed pricing itself.
Price Optimisation – Future
There’s still a long way to go. We’re still reacting to the monumental change that ecommerce has brought to the world of retail and to supply chains. However, what this research does point towards is the importance of utilising a combination of huge quantities of historical data, machine power and learning, and data insight, all combined. By doing so, and using the tools available to us to do so, we come ever closer to price optimisation.
Given how tight margins often are in the ecommerce market place, that only makes sense.