At Lexis Nexis Risk Solutions, predicting the future is the business. Using a combination of big data and scoring analytics, the company helps insurers assess their risk and streamline their underwriting.
Eighteen months ago, CMO Lisa Agona determined it was time to apply predictive analytics to the company’s marketing efforts. “We had been talking about how the effective use of business-to-business data hypotheses could provide us with competitive differentiation,” Agona told CMO.com. “We decided to take that core capability and create a culture of analytics.”
The marketing group piloted a predictive customer attrition model that saw attrition levels drop by a double-digits percentage. In fact, it was so effective that Agona was able to build a business case to invest in predictive analytics enterprise-wide across the customer life cycle. “This is our key focus for the future,” she said.
“The holy grail of marketing is to proactively pounce upon every individual customer opportunity by predicting beforehand who will respond and to pre-emptively intervene each customer loss by predicting who will defect,” said Dr. Eric Siegel, author of "Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie or Die," in an interview with CMOc.om. “The only way to do this is with predictive analytics.”
But the application of predictive analytics to corporate marketing is in its relative infancy. CMOs on the forefront of adoption face a number of challenges: no clear standards, an overabundance of data, lack of analytics understanding, and a field of new, often unproven vendors and technologies.
“A lot of marketers just don’t have the judgment combined with analytic skills to be able to take smart decisions from predictive analytics,” said Pat Spenner, managing director of the Corporate Executive Board’s (CEB) Marketing Leadership Council, in an interview with CMO.com. “Sometimes that leads to big decisions that are misguided. More often, it can lead to analysis paralysis or little course corrections that don’t amount to much.”
But forward-thinking CMOs are overcoming the obstacles and learning valuable lessons about the nature of predictive analytics, including what works—and what doesn’t—in the modern marketing department.
1. Predictions Aren’t Perfect
“Numbers aren’t perfect.” That’s straight from the mouth of Nate Silver, the New York Times blogger who used data analytics to outperform pundits and pollsters alike with his 2012 presidential election predictions.
Indeed, predictive analytics aren’t without fault. “There’s always going to be some level of error and some level of learning,” said John Bates, product manager for predictive marketing solutions for Adobe’s Digital Marketing Cloud. (Adobe is CMO.com's corporate parent.) “If you think you’re going to identify that miracle solution for all your woes and ails, you’re not being realistic.”
With time and effort, predictive model performance can improve, but it will never produce 100 percent accurate prognostication. Early adopters are wise to take that into account and factor in issues that could increase flawed outcomes. “As you are designing your analysis, think of all of the possible factors that might influence a particular outcome ahead of time,” said Jed Jones, president of marketing consultancy MindEcology, in an interview with CMO.com. “This is where real-world business experience comes into play. A power user of analytics packages who lacks marketing knowledge will probably not build a very powerful predictive model for marketing applications.”
2. Data Is Dumb
Without a good hypothesis, any experiment is likely to fail. Brendan Sullivan, CMO of FieldAware, uses predictive analytics for prospect modeling, retention and attrition models, and renewal modeling for customers of the company’s field service management software. But in every case, his team starts “with an understanding of the key drivers of the model,” Sullivan told CMO.com. “You need to let the variables in the model tell the tale and play out.”
For example, when his marketing team wanted to determine which customer prospects were most likely to convert to customers, they realized that one of the most highly correlated variables was going to be how long the potential client had been in business. “Insight from analytics told us that more than six years in business was a key driver of increased conversion,” he said. Sullivan also uses predictive models to determine the highest value targets within a new industry.
CMOs must “understand that the data doesn't know anything, inherently,” MindEcology’s Jones said. “Your job is to make the data tell a story.”
Louis Gagnon, CMO of Yodle, knew he wanted to understand the impact of the price of the company’s software-as-a-service platform on both customer acquisition costs and churn rates. So he developed a model to do precisely that. What he found was that the product’s $700 price tag was actually making it more expensive to get and keep customers. After a number of tests, he found that the ideal price point was actually $149, not exactly music to the not-yet-profitable company’s leadership team. But Gagnon was convinced that the lower price would double the average lifetime of the customer. While it delayed profitability in the short term, it would quadruple revenue in the long-term, Gagnon told the team. And two years later, he was proved right.
3. It’s Not A Marketing Project
At Lexis-Nexis, the key to predictive analytics adoption was involving the rest of the business. “This was not done in someone’s back office. It was an extremely and highly collaborative project across technology, operations, human resources, sales, and marketing with executive backing by the CEO,” Agona said. Indeed, a year before she considered building a pilot predictive analytics model, Agona was out in the business marketing the importance of analytics to the whole company’s future. The partnership with IT, in particular, was key. “Your analytics are only as good as your data,” she said.
For Yodle’s Gagnon, who hypothesized that an adjustment in product price could provide a more sustainable route to profitability, working closely with finance was important. If he didn’t have them on board, then it’s unlikely anyone would have paid attention to what the models were telling him.
4. Use It Or Lose It
One of the biggest mistakes marketing leaders make in the field of predictive analytics, Siegel said, is generating a predictive model that is never put to use. “The organization needs to be prepared to change existing operations according to the predictive scores,” he said. “For example, if direct mail has normally been sent to an entire list, there needs to be buy-in to now send not to the whole list, but to only a portion of the list, suppressing from it those customers predicted as less likely to respond. If the per-person predictions are not going to be acted on by driving per-customer decisions in this way, then the predictive models ultimately serve no purpose.”
It’s important not only that there is high-level support to act on the results of predictive analytics, but processes in place to do so are also needed. “You have to ensure that the actual output of these models ties into execution points,” Adobe’s Bates said.
5. Correlation Is Not Causation
“Most marketers took that one stats class in college, and they’d prefer to forget that,” Bates said. And, for the most part, that may be OK thanks to analytics solutions optimized for those who may not be data scientists. But if there’s one piece of information worth remembering from the quantitative marketing class it’s that correlation is not causation.
“All these new tools based on correlation analysis,” said Kevin Clancy, CEO of Copernicus and author of “Your Gut Is Still Not Smarter than Your Head,” in an interview with CMO.com. You may see a correlation between increasing the marketing budget from $18 million to $19.5 million and a market share increase of half a point, but is it a causal relationship? “Only time will tell,” said Clancy, whose company recently invested in a predictive analytics company.
6. Small Wins Are Big Hits
Most CMOs believe that they must compete on analytics if they are to be successful, Adobe’s Bates said. “One of the biggest issues is the siloed nature of marketing once you get further down into the organization,” he said Too many CIOs see those silos as an obstacle to implementing predictive analytics. “You can still leverage it within a single silo,” he added. “Take that portion of marketing and get wins with that, and you can further evangelize analytics.”
For example, when Agona wanted to begin a small, controlled predictive analytics pilot, she picked an area of marketing in which there was a foundation of data and knowledge. Lexis Nexis had built some attrition models in the past. As a result, the pilot took just six months. “It was important that it did not drag on a long time,” she said. “We wanted to find the low-hanging fruit, prove the concept, and sell it.”
Added FieldAware’s Sullivan: “We haven’t waited for a full 360-degree view of the customer that is perfect. The insights you can gain along the way of developing the models can be immense, and it continues to sharpen your approach and thinking. It also begins to frame better execution.”
Over time, data gets better, models get stronger, and analysis gets richer. “We’ve also realized that we can use predictive data to test out concepts and approaches quickly,” Sullivan said. “Rather than waiting for the perfect view of the world, the models get us close, and through our execution we gain even further insights that feed back into the models.”
At Yodle, Gagnon said he practices agile marketing. “We test things—a lot. We test price points and value propositions and go to market as fast as we can to gather enough data to have an opinion,” he said. “Test and test and test. And test right.”
7. Import Some Experts
There’s no denying that a few data scientists with a feel for marketing may be hard to find. “You need people that are good in the quantitative areas, like statistics, but can also connect the dots with the marketing components, like advertising campaigns and sales efforts,” said Alexander Edsel, director of the marketing master programs at the University of Texas at Dallas’ Naveen Jindal School of Management. “Consumer behavior is foreign to the typical statistician.”
The best are “well-employed,” Gagnon said. But those considering outsourcing the whole of predictive analytics to a third party may be making a mistake. “On the analytical level, it is important to have seasoned practitioners at your side for the hands-on technical execution,” Siegel said.
One of Agona’s first steps was hiring a strong analytics professional to lead the pilot project. “They do exist,” she said. She wanted someone not only skilled in data science, but also in communication skills. “The other challenge was that we were building this capability from scratch in marketing, so we needed someone who was OK with being the first—a self-starter willing to take on the risk.
CEB’s Spenner also advises CMOs to introduce training and mentoring programs that will “bring big judgment to the table with big data.” Make sure everyone understand the statistical limits of what can be concluded from the data, for example, or the business dynamics that are at play when bringing predictive analytics to a decision.
8. Be Promiscuous
A whole host of vendors, technologies, consultants, and methods are out there when it comes to applying predictive analytics in the marketing function. “Everyone is doing their own thing,” Edsel said. “And if they find something that gives them a competitive advantage, they’re keeping it to themselves.”
The best approach right now is to experiment—a lot. “As with any new development in marketing, everyone and their mother jumps into the business. Some are really smart and others don’t have a clue,” Clancy said. “Hopefully as corporate America continues to invest in these technologies, some standards will emerge.”