How Smarter Forecast is bringing a new dimension to the retail market and organisation

Retail customers expect an engaging personal experience when shopping online or in a store. Retail businesses can do a better job of providing that experience by using data analytics to learn their customers’ needs and habits. As part of the journey towards our vision of becoming Australia’s most trusted retailer, we want to continue to understand and mitigate causes of waste and life deterioration that impact freshness, and maximise availability for our customers. Smarter Forecast helps Coles achieve this goal and enables our business to be driven by one forecast value to optimise our end to end value chain.

A Business Perspective – Brock Newell (Head of Fresh Produce Easy Ordering)

My name is Brock Newell. I am leading a Transformation Program at Coles that is focused on reinventing the forecasting and replenishment solution across the Produce Supply Chain.  Drawing on my previous experience leading the Coles Supply Chain Systems team, I understand the importance of stock flow and product availability for our customers.  We have been actively looking at innovative ways to address and improve the sales forecast and solution for events e.g. weather, customer demand, event planning.  To meet this need, Coles Supply Chain Team partnered with Coles Advanced Analytics Team in 2018 to co-create a new solution called Smarter Forecasting. Smarter Forecasting is born from the concept that external and/or Big Data could produce better forecast adjustments than the Legacy Forecasting in Coles for previous ~12 years. Through running several trials and using Machine Learning we proved the effectiveness of Smarter Forecasting. As a result in the first year we noted forecasting improvements of 4%, and now it is our leading forecast solution. This system has allowed Coles to better replenish, generate improved Supplier Orders and is ensuring we are meeting customer demand. We’ve also expanded the Smarter Forecasting technology to support other Coles initiatives beyond Forecasting and Replenishment, and see this as a core system that will drive immense value across our broader Coles business. Also Smarter Forecasting is now an enabler of a number of initiatives broader than just the Forecasting and Replenishment use cases, that now uses the core Forecasting to drive value across the broader Coles business.

What are my key takeaways?

  • Data is critical, and now more than ever.  Maintaining data accuracy and quality is essential to any successful business
  • This is a journey – “Machine Learning” says it all. The system will improve with time. Put trust in the system to calculate and create outputs of a higher quality free of associated/ unconscious bias.

So how does it work?

Smarter Forecast, is a machine learning sales forecast, that has over 300 model configurations developed, driven by 1500+ data points, with over 30 different algorithms that connect to each over numerous ensembles. It has automated and self-learning model outputs that allows itself to evolve over time.

Our mission is to be the single source of truth for forecasting, enabling AI to automate and drive scientific decisions for Coles.

We capture “facts” from many sources of data (Sales, Promotions, Pricing, Catalogue, Events, Weather) and engineer many features to customise our model (Seasonality, Ends, Substitutions, Promo and pricing groups/clusters, Catalogue multipliers and additional context (e.g. days since) etc.).

Sumith (Principal Data Science Architect): Considering the scale of the Smarter Forecast solution, having the right design and architecture was essential. It is one of the largest scale solutions demonstrating the best-in-class data science and software engineering practices. There have been many reusable components developed that massively speed up the development and production processes.  Smarter Forecast currently provides multiple interfaces to seamlessly integrate with many different systems and applications across the broader business and can be easily extended to support future use cases

Continuous improvement

An important aspect of Smarter Forecast is that its continually “trained” on new data in order to keep it up to date and relevant (e.g. COVID19 impacts). This has a major impact not only from a forecast accuracy perspective, but also ensures the models we build adapt to the ever changing environment we live in.

The evolution of Smarter Forecast is driven by critical analysis on prediction results vs actuals, and is a never ending process that will make the forecast smarter through each cycle:

Sam (Head of AA team): “Smarter Forecast is a fantastic effort by the team to build a world leading demand forecast that understands customer demand at a very granular level and enables the business to make better planning and execution decisions

Building trust in a new forecast: How do we demystify Smarter Forecast to Make Easier Decisions?

We introduced Smarter Explainer, a visualisation app built in R on top of Smarter Forecast, to be able to enable teams in Coles to self-serve, diagnose and generate insights to make informed decisions based on Smarter Forecast:

– Making it as easy and intuitive an experience for all teams

– Providing the right information and at the right level

– Instilling trust and confidence in Smarter Forecast to make those decisions

Alok: “Embedding the forecast into everyday decisions of the business was always going to be a difficult change in mindset and journey. With the Smarter Explainer, it makes it far easier to investigate and explain historical and future predictions by seeing the impact of things like weather, events, promotions, pricing, product placement and many many more features, which enabled us to truly build trust in a machine learning product”

It is more than an opportunity, it is already a reality

Sovann: “A number of key breakthroughs have allowed us to achieve the levels of performance we see today. However there are still multiple areas in the demand forecasting problem which have some way to go before being completely solved

Demand forecasting model captures the impact of price, promotions, space, events, weather etc to give an accurate and rich  forward looking view of sales and business performance.

Taking the advantage of an automated and self-learning model, and driven by 1500+ Demand Drivers for >6yrs history, it is architected to extend and improve over time through iterative development and continuous delivery.

In terms of benefits, it already brought:

1) A unified forecast, reducing the need for multiple disparate forecasts

2) Multiple interfaces to expose to different systems and to enable different use-cases

3) More accuracy compared to existing 3rd party solutions

Key decisions can now be easily simulated to understand customer response. And this will benefit the overall customer journey

Come join the team

* Forefront of forecasting research with real world implementation

* Dynamic and thriving environment

* Join our talented and growing team of over 100 data scientists

* Solve sustainability and customer experience problems together

Sovann Tong
Smarter Forecast Stream Lead,
Advanced Analytics

Alok Joshi
Engagement Partner
Advanced Analytics

Sam Riethmuller
Head of Advanced Analytics

Johan Lequien
Change Manager
Advanced Analytics

Sumith Matharage
Principal Data Science Architect
Advanced Analytics