Why should retailers include price optimisation and customer demand forecasting at the core of their digital transformation? Mo Joueid, Google Cloud Platform Practice Lead at Kin + Carta, shares some insights on this topic and how the Google Cloud team at Kin + Carta are supporting supply chain optimisation journeys.
This story begins with a pair of jeans. Like many of us, I used the recent holiday period to take advantage of the retail sales that normally take place around the holiday season. Call it a wardrobe refresh, should I get the opportunity to get back into the office and meet clients face to face again, or some well deserved retail therapy. But one of the items I have had my eye on for a few months, were a pair of jeans. But not just any jeans, but I seem to have picked up a soft spot for well fitted designer jeans over the years. And there’s perhaps nothing better than seeing that item of designer clothing you’ve had your eye on come down in price. And if I manage to need a smaller sized waist then all the better.
But as my luck would have it, when the price did come down, as did their inventory and I could no longer find the jeans in my size. I then watched as the days went by in hope that the situation would improve for the better. But what happened next had me puzzled. Aside from the sizing issue, the price for the item began fluctuating. Did I miss the memo that said that seasonal sales mean that prices come down but may also go up and down several times over a period of days/weeks?
I was soon to learn that my ‘preferred’ retailer was indeed a Google Cloud customer and I, of all people, should have realised that they were most likely making use of the many Artificial Intelligence/Machine Learning (AI/ML) features that the platform offers, including, but not limited to:
- Price Optimisation, and
- Demand Forecasting
So what could have been influencing the price of my pair of jeans?
A ML model running on Google Cloud can take any number of data points to make a decision. At the very minimum, these tend to include previous sales and inventory related data. However, this may branch out and also incorporate environmental factors e.g. milder weather affecting cotton production, higher transport and logistics costs, trend setting that has sparked interest in a particular line of clothing or, as we witnessed more recently, the uptake in casual clothing as a result of home-working and less socialising. The event feeds can be broad and wide, but also need to take into account the subsequent implications of dynamic pricing.
In the case of my pair of jeans, how would the customer demand for size 34 fluctuate if the price were reduced by 5% versus 10%? How is the profit margin affected in each case? What impact would this have on inventory level? These are questions all modern retailers should look to answer, but deriving and scaling this kind of insight across such a diverse set of data points is challenging to say the least.
There is no one-size-fits-all model for retailers. Some fast-fashion, low margin retailers may find themselves price optimising on fewer data points, applying very minor price fluctuation, and placing more emphasis on customer demand and inventory management. Whereas a designer fashion retailer may rely on a completely different set of data points and factors, focusing on a niche clientele, and only needing to resort to price optimisation for a smaller number of line items.
In this blog, Part 1 of our Supply Chain Optimisation Series, we look specifically at how Kin + Carta leverages Google Cloud to help retailers optimise their pricing strategy to combat these very challenges.
How do we deliver price optimisation to Google Cloud clients?
We will now begin to look at the technical aspects of how we approach this with Google Cloud. The journey starts by working with clients to understand their business’ operating model. The initial objective will be to identify, explore, and assess data (quality) from source systems that will be integral to the ML modeling. Having identified these data sources, we begin the process of designing and building a robust and flexible Foundational Data Platform on Google Cloud that provides opinionated defaults, while allowing retailers to build and scale out data pipelines quickly and reliably. This will typically comprise multiple, segregated logical zones that provide elasticity:
- Capture/ Landing Zone: This will serve as the inbound point for the data platform. It will ingest and persist product details, list prices, order/sales history, supply chain management (raw material, sourcing, transportation, cost of returns, marketing), and any market data which may help the retailer remain competitive. This raw data lands here, before being processed. Whilst customer data is not a requirement for price optimisation, if there is any customer related data, this zone (or data therein) will have restricted access before this customer related data is masked/tokenised (see Google Cloud Data Loss Prevention — DLP —).
Note: Whilst the above doesn’t address the specifics of EU GDPR compliance, where this is a requirement for retailers, there are additional considerations given on which data is captured, processed, and stored.
- Processing Zone: Once ingested, this acts as the temporary staging area for in-flight data during transformation. The data required for price optimisation - as it originates from multiple different sources - will be disparate by nature. As such, this is where the data is subsequently curated and transformed through controlled processes to arrive at a Common Data Model (CDM). This provides a consistent definition for metrics, attributes, and rules for normalised price reporting and predictive modeling.
- Store: We always seek the separation of storage (landing/curated data), from compute (processing). This zone serves the general data storage needs of the retailer, but may also become the basis of a broader program of works geared towards a Data Lake or Warehousing. A topic that we intend to cover later in this series.
- Analysis/ Curated Zone: These are the final curated datasets available for external systems. We combine all the classes of data into one consistent, normalised data structure that meets the needs of the price optimisation model, but may also serve broader ML needs of the retailer. It may also be where you have data aligned based on customer segments, in cases where this is a requirement.
- Use/ Consumption Zone: In this zone, we start to unlock the potential of Google Cloud ML, and specifically Kin+Carta’s recent acquisition, Octain, working alongside Vertex AI. We begin with training and tuning the ML learning models, taking into account various different possible scenarios, and outcomes. With each cycle of training that is performed, the model is tuned and optimised before resuming, maturing and improving at each turn. We then proceed to model serving through predictions, delivering valuable dynamic pricing to the retailer, which adjust in real-time.
With higher clothing and footwear prices pushing up living costs in many countries recently, and retail inflation reaching 6-7% in some cases, this is likely to have some impact on consumer behaviour. The Google Cloud ML feature covered here, supported by Kin+Carta’s very own Octain solution, can yield endless value to a retailer. It can be used by retail leaders to tackle the retail challenges ahead in real-time, manage customer expectations, maximise margins, as well as drive towards supply chain planning and optimisation. The next time you are shopping for that next retail item, take a minute to question how that retailer arrived at the listed price, and why they are confident that you would buy at the listed price.