Super article on predictive analytics. From marketingweek.
As big data has become embedded in brands and marketing departments, gradually the talk has become focused on the predictive – even prescriptive – power of data.
With companies recording interactions with customers at an ever more granular level, the onus is on marketers to use this data not only to understand past trends but also to predict future behaviour, and to specify automated responses to these predictions.
Pete Markey, CMO at the Post Office, points out that predictive analytics can unlock significant commercial opportunities. It was for this very reason that, early last year, the Post Office created an ‘analytics centre of excellence’.
So far, the organisation’s approach to predictive analytics has been centred on uncovering ‘next best action’ scenarios, using customer data to indicate the best offer or communication to send to a given customer when they interact with the brand. “We’re at early stages with this but we are seeing encouraging results,” says Markey, pointing also to work done with mortgages, which has shown where to deploy mortgage advisers to service customers who are “predisposed to [consider using] the brand”.
Markey also notes that data scientists are increasingly being sought out at university careers fairs.
New capabilities required
Mike Whitelegge, head of big data solutions at Marks & Spencer, agrees that there is an acute hunger for data-related skills. “If we look forward five years, the types of jobs we’ll be needing to fill have not even been thought of yet,” he says. “I feel the universities could do more than they are now.”
Whitelegge says the introduction of Marks & Spencer’s Sparks loyalty scheme late last year has been helpful as a tool to unlock the power of predictive analytics. He says that the “triangulation” of data sets is particularly interesting as it enables marketers to link behaviour with psychology around factors such as the weather.
“Every day our shopping experiences are subconsciously filtered,” he explains. “I find working in retail interesting for this very reason. We’re all consumers and [in this industry] it’s easy to humanise it.”
Yet for Ben Kay, a senior consultant in analytics at IBM, organisational cultures will need to change significantly in order to exploit the power of predictive analytics. “Data and insight is often ‘owned’ by a department,” he warns “There needs to be a democratisation of data.”
Kay believes that many brands do not understand how to use predictive analytics, and neither are they aware of the potentially huge impact it can have.
“We’re moving from reactive social listening to looking at the ‘why’ behind it,” he says, claiming that effective analysis of social data can uncover personality traits which, from a marketing perspective, can enable brands to tailor communications “to a segment of one”.
Léonard Gaya, head of digital at French publisher Editialis, has also witnessed the power of engaging people in a targeted way.
He admits that in the past the publisher tended to deliver newsletters or email marketing campaigns in an ‘indiscriminate’ fashion: “It was a case of cross your fingers and hope for a response.”
But he says that, thanks to the implemention of predictive intelligence tools from Sailthru, Editialis is able to anticipate engagement at an individual level and has increased click-through rates ‘dramatically’.
AutoTrader, meanwhile, has set out to develop its predictive targeting capability this year. Lara Izlan, director of programmatic trading and innovation, says that with more than 40 million cross-platform visits per month AutoTrader has a lot of data. “But behavioural and profile data can be static,” she says. “We want to get under the skin of car buyers. The process of car buying takes a few months. We want to know where they are on this journey.”
Izlan says that AutoTrader plans to leverage algorithmic models and lookalike modeling, targeting individuals with similar online behaviour to customers, to grow its user base. “By using propensity models based on search behaviour, we can target more effectively,” she says. “It’s a competitive market. We simply have to evolve this way.”
Izlan also points out that AutoTrader is moving away from thinking about the traditional marketing ‘funnel’. “We’re trying instead to think about it as a cycle,” she says.
Similarly, Daryl West, social media insights lead at O2, says that the company is looking to monitor the conversations that are happening constantly across social channels, and has been working with social media analytics company Crimson Hexagon. By monitoring words such as ‘tariff’ or ‘network’, O2 can forecast social satisfaction scores, he says, and thereby use social media as a type of early warning system.
However, he warns that it takes time to learn how consumers talk on social media.
In doing so – and in using data to predict customers’ behaviour in general – it is important that brands do not “cross the creepy line”, warns Jason Gordon, partner at Deloitte. Yet Gordon claims that computers will be able to do most of this work in the future – even when attempting to achieve this delicate balance.
“It’s been hard for computers of the last decade to process the unstructured text and images on social media data, but language processing has moved on and we’re going to see immeasurably more accuracy,” he explains.
“There is no sector that won’t benefit from predictive analytics,” Gordon adds. “And not just predictive but prescriptive, too.” He describes ‘prescriptive’ analytics as that which delivers very specific recommendations around factors such as stock levels.
“Millions of decisions get made every year that would benefit from a predictive modeling capability,” he says. “It can help optimise pricing processes and promotions. Often it’s impossible for a human to keep on top of all of this, so averages get applied and specificity gets lost.”
Gordon believes that we will see marketers moving beyond ‘rudimentary’ loyalty schemes: “Rather than starting with a list of coupons, you can start instead with the customer and deliver really compelling offers.”
The need for strategy
This does not mean that marketers of the future will be able to rely entirely on computers. There will always be a need for a clear marketing strategy and a methodical approach to testing various options.
“You can start with a hypothesis but you must move into testing and validation,” saysPhuong Nguyen, director at eBay Advertising. “Some hypotheses stare you in the face and make sense while others are more speculative.”
Nguyen points to a piece of work eBay did last year on new mothers’ past and future shopping behaviours. The fact that a user had purchased maternity clothing was taken as an indication that in three to five months they may be in the market for newborn products such as baby formula. “We worked with a leading high street pharmacy and it doubled ad effectiveness,” he says, adding that “all targeting is based on predictive elements in this way.”
Rupert Naylor, EMEA vice-president of Applied Predictive Technologies, which is part of MasterCard, similarly suggests that acting on insights from predictive analytics can generate positive outcomes that may seem counterintuitive to marketers. Adopting a test-and-learn approach can help identify the benefits while reducing the risks. He gives the example of when sandwich chain Subway introduced the ‘$5 Footlong’ sandwich. A franchisee in Florida came up with the idea, he explains, but the company was nervous of losing too much margin. Yet thanks to a very successful test, which revealed how the offer could drive more customers in the store who may then spend on other drinks and snacks, the concept was eventually rolled out internationally.
With the threat of more stringent data protection legislation looming large, Naylor also points out that predictive analytics does not necessarily require any personally identifiable data. “It’s based on statistical models,” he says. “What can be really transformative is the scale.”
How powerful can predictive analytics be for marketing purposes?
Very powerful. A lot of direct marketing activity is retrospective; it’s about trying to replicate the past. With predictive analytics, you’re being proactive. It can turn the tables.
Have you done a lot of this type of activity?
When I worked at Zurich, it was about understanding who the customers are. That’s the first thing – identifying who your customers are and where they are, so as to be able to deliver targeted communications to segments, and find similar ones. When I worked at Sky, we did propensity models with predictive elements, so we could identify if somebody looked likely to leave. It’s about looking at a grid of products and developing stories [to determine what the next best action should be].
Are marketers still at the early stages of understanding predictive analytics as a tool?
We’re at the tipping point. Traditional database marketing and analytics are changing. Technology can process so much more data now and people are waking up to this. Twenty years ago, people fell into data jobs but now there’s a focus on data in Silicon Valley too.
What will see in the future?
We have self-learning technology now and ability to hone in one-to-one [on individual customers]. That’s game-changing. But whereas lots of startups have the luxury of no legacy [technology], many other organisations are burdened with legacy systems. Also the thinking from marketing and sales is very constrained to current capability. Who knows what the jobs will be in 20 or 30 years’ time. It’s no longer about delivering a report or piece of analytics – technology is moving so fast that it will all be automated.
In the future, companies that don’t use predictive analytics to optimise their marketing spend and marketing actions will not be able to compete effectively with the companies that do.
This type of analytics is geared towards identifying triggers that signal changes in a customer’s likelihood to need or buy a product. One example could be identifying when a car company’s potential customer has a daughter that is turning 16 and might be in the market for a car.
Another example is a telecom company that may realise that the percentage of 3G calls on a 4G phone combined with the percentage dropped calls is a key driver into the likelihood that a customer might leave.
One of the biggest challenges with applying predictive analytics to marketing is the ability to quickly test various actions and analyse the implications on sales and revenue. This is where the concept of machine learning or cognitive computing can come in.