Request a Demo

Close Form PopupLink to Close the Form Popup

Demo Form

Learn how we use your personal data in our privacy policy and about our country/region options

Data Science and Look-Alike Modeling with Canadian Tire

Written by Alida

Published December 03, 2018

Data science. Look-alike modeling. Training predictive models.

Sounds complicated.

But these are hot topics especially for progressive insight professionals who are looking to deliver even more value to their businesses. With great examples and advice from Canadian Tire, I hope to demystify those topics and show how insight professionals can support marketing and product decisions around customization and personalization that today’s consumers demand.

Before that, I’d like to extend a big thank you to Cedric Painvin, Associate Vice President Consumer Research, and Colin Huggard, Senior Research Consultant, at Canadian Tire, for sharing their data science journey and advice for success. For context, the Canadian Tire Corporation is Canada’s largest retailer with 1700 stores, owns multiple brands including Sport Chek, Marks and Canadian Tire, and generate over $12.6 billion in annual revenue.


Getting started: bridging silos

Canadian Tire’s insight community, the Canadian Tire Customer Panel (CTCP), is a key component to bridging data silos. Canadian Tire is in the lucky position of having millions of people in their loyalty program and an engaged insight community of close to 200,000 members. The loyalty program’s unique identifier forms the bridge between:

  • Purchase behavior collected via loyalty member transactions
  • Demographic, psychographic and attitudinal data collected via CTCP

With this connection between the loyalty program and CTCP, the insight team is able to develop unified customer profiles that are deep and granular enough to inform the level of personalization that customers demand, and can be leveraged by decision-makers across the business.


Assembling a can-do team

Being stewards of the voice of the customer and connecting the dots to transactional systems, Canadian Tire’s Triangle loyalty program requires a can-do team comprised of insight professionals and an embedded data scientist. The insight team follows a test and learn approach to deliver wins to the business and experiment with things that haven’t been possible before. No one on the team was initially an expert on how to use community data to enhance machine learning and data modeling – they had to learn their way there.

Combining information from the loyalty program to build on the growing knowledge gained about community members can really bring niche product buyer profiles to life which enables better targeting for marketing and special offers. With this paired knowledge, the team can often deliver value without intrusion – finding insight without having to reach out to their community members at all. For example, the insight team has profiled the demographics of brand buyers versus non-brand buyers for brands that Canadian Tire carries which include Motomaster, Mastercraft, NOMA, Canvas and Woods. This helped the brand teams have a better picture of who their customers are in order to help develop their long-term strategic plans.


Training Models and Finding Look-Alikes

Pushing data out of the community and into the transactional system makes two things possible:

  1. Smarter machine learning and an anchor for the predictive model
  2. Look-alike modeling, where results from a small sample of community members are extrapolated to like-type customers in the wider customer database

Retailers are trying to personalize their messaging and offers to the right people at the right time. Here are three questions Canadian Tire’s insight team was able to answer by building on what they already know from their transactional loyalty program data and CTCP:


Are we sending targeted messaging to the right person?

Trade Pros, including plumbers, electricians and contractors, are Canadian Tire’s high-value customers. It’s hard to determine whether someone is a trade pro based just on their transactional information. Through an iterative process, the insight team asked a hypothesized group of trade pros about their profession through the community. The results enabled them to remove the people who were inaccurately categorized, with profession becoming the anchor to train and improve the model. They were able to successfully identify a sizable trade pro segment in their loyalty data. These people spend 3.7 times more and visit the store 3.6 times more often, so knowing who these customers are and improving targeted messages will have a sizable direct business impact.


What’s the right time to send a “just moved” offer?

Data scientists had the anchor they needed to train their data based on inferred geolocation and translations data but needed to validate their hypothesis that a customer had recently moved. They simply asked people who they thought might have moved if they actually had moved. The result is an algorithm that will deliver more relevant offers at the right time (after recently moving) to the right people (recent movers). They’ve also started experimenting with predicting when someone will be moving, based on what they’re buying and when.


How could we customize messaging with the right message using look-alike modeling?

Colin shared two great examples of look-alike modeling:

  • Customizing marketing communications about the benefits of Canadian Tire’s loyalty program and credit cards. They talked to customers in the community about messages and benefits that were relevant to them. Then they found like-type customers in their wider loyalty database and sent them improved marketing messages through digital channels. The blend between survey results and loyalty data resulted in more impactful and effective communications. They’ve seen double the conversion rate, 30% savings in cost per thousand impressions and half the cost per acquisition.
  • Delivering customized value-based digital marketing for automotive messaging. Canadian Tire wanted to personalize automotive messaging so they worked with a research vendor to do consumer segmentation based on attitudes, behaviors and mindsets for the automotive category. They used the typing tool (key questions that drive categorization) to recreate the segments within their community. Since each member of the community is also part of Canadian Tire’s loyalty program, their data scientists can layer the transactional information on top of the segmentation to identify the behavior patterns that are most likely to occur within each segment. Once they connect the dots between the segment and the behavior, they can project the segments out to the entire loyalty program universe. Added bonus: the progressive profiling of community members over time will continue to build and improve the predictive model. So within the loyalty program, Canadian Tire can then identify who is the best fit and what is the most profitable segment for specific products and campaigns. Their targeted offers again become more relevant versus simply basing offers on sales data alone.

These projects have had direct and tangible results through more effective marketing, a clear impact on revenue and a growing appreciation throughout the organization of the community as a strategic resource.


Advice for starting out on the machine learning and look-alike modeling journey

Colin shared his advice for getting started:

Start simple. Results that might seem obvious or dull to researchers are likely to be mind-blowing to other stakeholders in the business. Either because the answers are helpful or because it fires their imagination for what else is possible.

Track everything. Institutionalize tribal knowledge about the process, learning and metrics. This helps to demonstrate learning and quantify what improvements were made during the process, especially when things don’t go quite right or as expected.

A/B test everything: Take any opportunity to A/B test or see what pulling on different levers can do. Leave no stone unturned. It’s all about learning and uncovering new insight.

Socialize: Share the results widely. Drawing attention to our successes isn’t just an ego thing, it will wake people up to the potential of customer panels and big data, and you will likely end up with very actionable requests coming from unexpected places.


Wrapping up

Translating the insights into a revised marketing communications strategy is the vision of where Cedric and Colin want to go to support Canadian Tire – developing better-targeted offers, briefs, messaging, content, imagery, copy and more. The CTCP paired with innovative machine learning and look-alike predictive modeling tactics are helping to bring this closer to reality. It’s great to see two typically siloed teams starting to work more together as partners for delivering customer insight to their business.