“Imagine being able to tell the management of your company that you know what a customer is going to do in the future,” says Pankaj Arora, senior analytics consultant at online lottery broker Tipp 24.
He is using predictive analytics – that is, looking at historical data to forecast trends showing which customers are at risk of being inactive. Tipp 24, which enables people to play games such as Lotto and Euro Millions online, can also predict which inactive customers have a high likelihood of becoming active again.
During its first year working with predictive analytics specialist KXEN, Tipp24 claims it saw a 300% uplift in the targeting accuracy of its marketing and has cut the size of its target audience for any marketing campaign by 25%.
Use of predictive data to inform marketing campaigns is a burgeoning area, in which the telecommunications, financial and utilities sectors are leading the way.
The energy sector uses predictive data to combat the perennial problem of customer churn. Energy regulator Ofgem says 160,000 domestic consumers change electricity or gas supplier every week, so the need to improve customer retention is paramount.
Clifford Budge, EDF Energy customer insight manager for B2C energy sourcing and customer supply, says: “We can meet customers’ needs better if we learn more about how they think, what they buy, what they use and how they want to interact with their supplier.”
Working with business analytics software provider SAS, EDF Energy uses churn modelling – evaluating the propensity for customers to leave the organisation – by examining metrics such as ‘life on supply’ (older customers are likely to be more loyal than younger ones) and overall interaction levels. Once the customer data has been classified and those most likely to leave have been identified, they are monitored. If they defect, the customer insight team reports this to the EDF marketing department so it can prioritise customer communications based on risk.
EDF Energy has also been focusing on its ‘dual fuel upgrade to reduce churn’ programme – identifying which customers could potentially buy their gas as well as electricity from EDF. The data has predicted that the top 25% of customers (based on revenue) in this category are four times more likely to take dual fuel and so are far less likely to defect to a rival energy company.
By identifying those customers least likely to change supplier, EDF can avoid wasting money on marketing upgrades to them, says Budge. “If you were a supplier that suffered a million customer losses every year, with the top 25% of customers billing an annual average of £1,200 a year, that equates to a total risk from those customers alone of around £300 million a year. Even if you can action just 5-10% of that, it’s a significant saving.”
In the motor industry, customer behaviour can also be predicted in sales and after-sales, although it is a complex art, says Pete Bradley, finance director at new and used Ford dealership Sky Ford. He explains: “Most finance agreements do not reach the end of their terms because customers replace their car before the end of their agreement.”
Sky Ford works with contact centre service provider Call IT Automotive to predict when people are most likely to replace their vehicle. Bradley says: “By looking at the average life period of our finance packages, we can determine the cycle of purchase more accurately than by relying on the contracted terms of the agreement. This enables us to limit customer leakage.”
After purchase, customers return to Sky Ford for servicing. “This allows us to assess their yearly mileage, which is critical to anticipate the time of their next visit,” adds Bradley.
By using predictive analytics, Sky Ford can market relevant offers to customers at the right time. “Our database segmentation determines who, and Call IT Automotive’s predictive model determines when a contact needs to be made,” says Bradley.
The dealership’s marketing is tailored to each customer, depending on the age of the vehicle and on whether Sky Ford can see that the customer has visited another dealer for their previous service or MOT.
But Bradley says consumer behaviour varies depending on the model, version, residual value or age of the vehicle. The predictive model uses customer profiling to establish a contact strategy tailored to the individual.
“The predictive model uses statistical analysis of customer behaviour for each marketing segment. It calculates the prediction criteria to be applied at individual record level and challenges the future contact dates in our database. This allows us to promote targeted marketing offers, knowing when our customers reach the different steps of the ownership cycle,” says Bradley.
The uplift on service and MOT is 10-15% on what the company considers ‘prime customer’ data – those who have visited the dealership for their last scheduled service or MOT, and who are effectively the most loyal customers.
But it is not just industries in which customer behaviour is needs-led that can benefit from predictive analytics. The record industry is also reaping the rewards of access to rich data, either to back up a hunch about a hit record or to identify which album track should be released as a single.
Former Sony Music Label Group boss Don Ienner – who is now principal at IMO Entertainment and an artist manager and consultant to Universal Music – has invested in predictive data for years. He says: “If you know in your heart that the first three songs from an album are going to possibly be the lead tracks, then first, you want to see if you can justify that with some data; second, [data] may point you in a different direction; third, [data] can give you the best sequence of those singles; and four, [data] may just surprise the hell out of you.”
The energy sector uses predictive data to combat the perenial problem of customer churn
Ienner points to his work with Beyoncé on her first solo album as proof that data can be used to predict a hit record: “There was controversy over what the first single would be. I wanted it to be Crazy in Love – I knew in my heart that it was a gigantic, important record for her career. But [Beyoncé] thought something different, and her father wanted something different to that. To me, this was a $100m-plus decision – not only for the company but for [Beyoncé’s] own future.”
Using a company called Hit Predictor, Ienner had the album tracks tested. The company edited each song down to two-and-a-half minutes and sent an MP3 to “a couple of thousand people”, segmented by gender, age and so on. Recipients were asked to listen to a minimum of three songs fully and then vote. They could also make comments.
“It came back so overwhelmingly in support of Crazy in Love, there was no getting around it,” says Ienner.
While gut feeling is important, predictive data is hugely valuable in today’s cut-throat music industry, he adds: “You have to make sure the decision you make has been vetted as much as it possibly can be. That means hairs on the back of your neck standing up, along with some real research that tells you what’s going on.”
When Ienner was working with the American pop rock band Hot Chelle Rae, which recently won Best New Artist at the American Music Awards, he called on Sound Out – which provides predictive insight, testing and analytics for the music industry – to test the third single for release.
“I was convinced – and actually the head of RCA was also convinced – that one particular song was going to be the next single, and it was our favourite song on the record. But it didn’t come back as well from testing as a few of the others, so we released a different one.”
Testing songs to predict audience response is also important when creating film soundtracks. Using a well-known song is not always a good idea, reveals Ienner: “There are times when something has been played so many times and no-one wants to hear it any more, or it may not fit the scene – you have to respect that.”
While the nature of the music industry means predictive analytics can be carried out by testing tracks on a willing audience, companies in other sectors work in different ways. The nature of online means brands necessarily capture a lot of valuable data relating to customer information and buying habits, but some supplement this with additional data sources and initiatives.
EDF Energy buys in third-party data sets, such as attitudinal data, to better understand customer attitudes, and lifestyle data with demographic details such as location and household arrangements.
Other organisations have implemented loyalty schemes, which can be an effective way to capture valuable data. Sky Ford has launched a loyalty scheme that rewards clients with discounts on future purchases. This data supplements its database information.
“It requires our customers to log in to a web portal and disclose a minimum amount of information about themselves and their driving patterns,” says Sky Ford’s Bradley. “Throughout the vehicle ownership, each point of contact is an opportunity to gather information which can be used to update the customer and vehicle records. This permanent flow of information is the basis for a refined prediction of future events.”
Of course, there is still an element of chance when it comes to predicting what any individual is likely to do. As Bradley says: “A one-size predictive model does not fit all, and never will do. We accept that a small percentage of contact is made at the wrong time.
“However, we see this as an opportunity to make adjustments to our database and potentially reschedule a conversation at an agreed date.”
Whether a brand is making predictions based on individual or generic customer behaviour, marketing is undoubtedly reaping results by looking into the future.
Case study: E.ON UK
Energy supplier E.ON UK has been tapping into the predictable behaviour of consumers to increase marketing effectiveness. Working with KXEN and its InfiniteInsight product, E.ON has been able to exclude 70% of its initial prospect list and focus on the 30% most likely to buy. Ryan Cotton, head of CRM at E.ON UK, says the technology “lets us target the right prospects with the right products, increasing sales effectiveness”.
E.ON has been able to create lots of predictive models in a short time. “We’re able to build quality models in less than a week and, if necessary, in a day or within hours,” says Lee Thomas, decision analytics manager at E.On.
In one campaign, predictive modelling led to a 20% uplift in sales and direct savings of £150,000. The models have also been used to optimise customer channels including email, direct mail, telemarketing and face-to-face sales.
In an effort to reduce churn, E.ON has also been able to boost the accuracy of its predictive models using a ‘text coding’ feature. This allows additional data to be introduced to the analysis based on commonly used words and acronyms in conversations between customers and call centre agents. For example, the frequency of acronyms like CED, which stands for ‘contract end date’, can be highly indicative of a customer’s intent to switch providers.
Case study: The Royal Shakespeare Company (RSC)
The Royal Shakespeare Company (RSC), in Stratford upon Avon, works with Accenture on behavioural segmentation to predict customer behaviour and tailor its marketing accordingly. The number of active bookers on its database has increased by 37% between 2006 and 2011.
RSC head of audience insight Becky Loftus says: “Anyone can have large numbers of people on their database but having an increase in active ticket bookers shows that using past customer behaviour to focus our marketing effort has been successful.”
The RSC’s segmentation includes genres people have attended in the past. “Whether it’s the big Shakespeare productions or the work of new writers, we can predict what people might like in the future,” adds Loftus.
“One segmentation category is golden geese, who are high spenders; then we have the regulars, who attend regularly but don’t necessarily spend as much and are probably a bit wiser to when the best prices are on offer.
“The golden geese might be more interested in the offers we have around dining in the RSC’s rooftop restaurant, so we will tailor that accordingly.
“Our regulars might be more interested in block booking shows, so we tailor our marketing based on what they have done in the past.”
Another of the RSC’s segments is ‘late bookers’. “If we have a show coming up and numbers are low, we send emails to late bookers who live within an hour’s drive of Stratford, perhaps with an offer,” says Loftus.
The RSC captures data via people registering on its website when they buy tickets, and by collecting postcodes when people buy tickets by phone. It also proactively collects information through data capture cards, which ask people what they thought of an exhibition and whether they would be interested in hearing from the RSC in future, for example.
It has recently started collecting data from people who eat in its restaurant in an effort to attract a broader audience.