Endeavour Group builds an AI-powered personalisation engine

Woolworths’ liquor offshoot Endeavour Group has built an AI-powered personalised marketing engine that it hopes will drive revenue growth and improve customer experience across its brands.

The group was created last year and is home to brands like Dan Murphy’s, BWS and ALH Hotels. It is expected to separate from Woolworths and become a large ASX-listed company in its own right.

Appearing at the Data+AI Summit 2021, Endeavour Group data science lead Xi Liang described the group’s efforts since November last year to build a personalised marketing engine.

“Our data science team, together with another cross-function team, is building a personalised marketing engine that performs segmentation for different groups of people with targeted messages,” Liang said.

“It was one of the most important strategic investments in our company and we [need] to bring a significant increase in revenue and improve our customer experience.”

Though the architecture was not revealed, it is at least partially run on a Databricks platform.

Liang said the actual engine comprised “around 20 different modules to make business decisions”. 

She said there were four main problems to solve - the theme of the email, what product should be in it, who to send the email to and when to send it. 

Touting products a customer had previously purchased was relatively simple.

 “We built an XGBoost model to rank the likelihood of people purchasing a product,” Liang said.

“We look at their past history, demographics … [and] at other sources of information, for example the marketing engagement, whether the product is on sale or any other promotion that is about to happen in the next few weeks.

“We built this model before getting into personalisation to help us to target the customer on the product they have purchased before.”

The company similarly takes clues from past behaviour when trying to recommend products a customer may be interested in, but has not purchased before.

“If we want to introduce a product that the customer has not purchased before this becomes much harder, because if you introduce a new product, especially liquor, it can be a little bit harder to make a decision,” she said.

“Do I like the flavour or not? Especially for beer, people tend to stick with the same beer, whereas for spirits or wine, they are more open to new options. 

“So when we solved this problem, we looked at marketing responses - for example, whether a customer has clicked on this product but has not purchased it, or whether this product is in their wishlist, or was ever in their cart but they never paid for it. 

“Those are a very good signal for us to capture in making the recommendations.”

Liang said the hardest decision made by the personalisation engine is the email theme.

“We have about 2000 different types of emails for customers to choose from, for the engine to push to the customer,” she said.

Themes included mystery gifts, products that the customer used to buy that are on sale, and new products to explore.

“We’re using a multi-armed bandit algorithm to solve that,” she said.

“There are two layers of information we look at - one is how likely you’re going to click the product in the email, and how likely you are to buy ther product. 

“Obviously those two are correlated but not necessarily [in the way that] you’ll click and you’ll buy, so we have additional logic to enhance whether this person will buy the product after clicking it.”

On the question of when to send emails, Liang said the company is yet to use AI to determine this, although it intended to do so to find the “optimal frequency” with which to send messages.

Liang noted the personalisation drive is already resulting in new experiences being put in front of the group’s 4.5 million customers.

She showed a before-and-after of a member email for a Dan Murphy’s customer. Where before the customer would receive a generic list of liquor products with no personalisation, the customer is now displayed options based on past purchases, regional preferences, product features and preferred price point.

Liang added that Endeavour Group had “a lot of plans on our to-do list in terms of personalisation.”

A major effort underway is the use of personalisation to aid in customer retention in the highly-competitive drinks market.

The group is also looking at how it can send push notifications to people that come within a certain distance of its outlets.

“This is something we’re already working on,” she said.