‘What sets Dunnhumby apart from the big data players? Offline modelling and brick-and-mortar data’, says the organisation's personalisation expert, Matt Cresswell.
West London based Dunnhumby, notable for its Tesco clubcard scheme, is reportedly being considered for purchase by Google now Tesco has put the firm up for sale, after financial problems with the retail giant reported earlier this year.
Making use of big data and streaming to provide insight is a valuable proposition for any firm. But Google’s interest lies in Dunnhumby offline, historical customer data analysis capabilities, which differentiates it from other personalisation services.
“Our unique selling point is our offline modelling. A lot of our competitors will bring in clickstream or web data to try and identify what a customer’s “mission” [the goal of a shopping trip] is,” he says.
Typically, data savvy e-commerce retailers or personalisation services firms will use around 10 seconds of customer’s browsing history, and continue to serve that as a recommendation on a website - even if the customer has bought the item in a shop. A pair of trousers may stalk customers while they browse online, even if they are already wearing them, for example.
“Dunnhumby is trying to to attack this from a different approach,” Cresswell says. “We model a customer’s behaviour and really understand what they are shopping for and how often they purchase these items offline. Then we delight these customers online, by showing them recommendations we know they like, but closing the loop so that when they do buy something offline they don’t get bombarded with messages that aren’t relevant anymore.”
The 35-strong personalisation team is just one part of Dunnhumby's operations, Creswell explains, and is focused on providing push notifications that might remind a shopper if they haven’t bought something they usually need, or if it is likely that they are running low on a certain item at home.
It also runs email campaigns and real-time healthy alternative notifications for grocers. To do this it uses the bespoke algorithms created by Dunnhumby’s data science team.
It takes masses of data from its retail clients, like Tesco or Coca Cola, and processes it on-site. This is published for its algorithms to predict what customers want overnight in an offline model, on top of its distributed database platform for data warehousing. It anonymises this data and sends it to the Microsoft Azure cloud, so when customers go online to shop it can offer recommendations and apply “real-time context within a couple of milliseconds,” Cresswell says.
Dunnhumby does incorporate the clickstream and web data bought by most e-commerce sites too, but only uses it to help identify a customer’s browsing preferences, he adds.
If Dunnhumby were to join forces with Google, a combination of the search giant’s algorithms and Dunnhumby’s historical data could be a perfect match.
A new buyer could inject the funds Dunnhumby needs to hire top talent and continue to innovate, Tesco CEO Dave Lewis told analysts, after announcing it had sold its broadband and Netflix-style film service BlinkBox to Talk Talk in January.
In a financial statement it states it will “explore strategic options for the Dunnhumby business”, with Goldman Sachs on hand.
When asked how the imminent sale was affecting employees' motivation, Cresswell declines to comment. He says: “On the whole, Dunnhumby is always a positive company and we’re so focused in what we have to do on personalisation.”
Despite the rumours, it’s business as usual for Cresswell’s team, and their focus is on testing new big data technologies on the market, including HD Insights: a Hadoop platform distributed by Microsoft, Azure Fabric and testing in Spark, he says.
But these new technologies bring challenges, including hiring the right talent who can work with experimental new tools to make shopping more personal.
“Previously, if you built a system you would use Oracle or SQL and you had two platforms you could choose between to learn and train in. Now there are so many diverse platforms to choose between it is hard to find talent that has coverage in all of these specialities and further, people that know what technology you need and when,” Creswell says
Politics can also get in the way of data science architecture too, he admits.
“Sometimes we have to be a bit open minded about what we should use, and what we need to use.”
The personalisation team have been using the Microsoft Azure cloud to provide customers with personalisation and recommendations since 2013. Dunnhumby also use Oracle's cloud-based Fusion apps to run their back-office IT functions.