Mecca has established its advanced analytics capabilities by working with AWS and Servian to clean up its data and build personalisation models for its two million customers.
Mecca is one of the first Australian companies to deploy Amazon Personalize, a machine learning service that provides a simple framework to build and run personalisation models.
The personalisation project is built on a wider data transformation which began in 2018 when Mecca built a new data warehouse on the AWS platform and selected Tableau as its BI reporting tool.
After building the data foundations, the business wanted to progress from the ability to access simple historical data to more predictive features, and began working with Servian and AWS to personalise the communications it sends to customers.
Lauren Shepard, head of CRM and loyalty at Mecca, said the brand needed to build digital capability throughout the business to bring the online experience closer to the store experience.
“When you go into a Mecca store, the personalised service and recommendations you get from a host is something that we wanted to be able to translate into the online world,” Shepard said.
Speaking during a recent AWS presentation, Shepard said there was a “reasonably large amount of chaos” across the retailer's data and systems.
“We had lots of information on our customers but we weren’t really using any of that information to optimise their experience,” she said.
On the communications front, Mecca was still sending mass emails so all opt-in customers “would receive the same beautifully designed content piece.”
Shepard said achieving 1:1 communications required simultaneously cleansing the company's data and restructuring it while trying to build out predictive models, as well as upskilling the e-marketing team to enable lifecycle marketing.
Once the business was confident in the outputs from the personalisation model, Mecca ran an initial proof of concept that pulled suggested products into a marketing campaign for mascara.
“By including personalised recommendations in those emails, it drove up all of the engagement measures for the email but most importantly it actually increased the average order value by more than 50 percent,” Shepard said.
“Our original hypothesis was that by showing people products that they are more likely to purchase or which are specifically relevant to them, we could not only get them to replenish their mascara but we could get them to add other items to the basket.”
Using product recommendations across Mecca's entire catalogue, as well as a long short-term memory (LSTM) propensity model to identify the best timing for product replenishment, Mecca's email click-through rates increased by 65 percent.
Following a successful proof of concept, Mecca is now running its personalisation model weekly for all its active customers, generating more than 10 million recommendations.
Initially, Mecca implemented product recommendations using native capabilities from its email management system, however the AWS recommendations have outperformed the email platform.
“We tested the Amazon Personalize recommendations against the system recommendations from our email provider," Shepard said.
"In theory, the recommendations coming through the email platform are also based on purchase history, but they don't take into account as many measures as the model that we were using through Personalize.”
Shepard said Mecca is working to continually improve the underlying data set to build more predictive models.
“We have really proven out the case that by actually showing products that are relevant to a particular customer, their life stage, their journey, and their purchase history, they're far more likely to convert.”