Updated: Why predictive analytics matters

Predictive analytics has become a buzzword for marketers on the quest to get to know their customers better, but how do CMOs utilise this methodology and the latest technology innovations to maximum benefit?

Predictive analytics is being touted as the best way for marketers to unearth fresh insights into their customer base using the wealth of data, technology tools and data science skills now at their disposal.

In the recent Temkin Group report Prepare for Next Generation VoC Programs, 72 per cent of respondents cited predictive analytics models and open-ended verbatims as increasingly important sources of customer insight over the next three years. Of the large companies with more than 500 staff surveyed, 26 per cent claim to be using predictive analytics software and 36 per cent use text mining software already, with the same ratio of both actively considering investment.

But without the right approach from the CMO and business support, projects can fail to deliver the significant returns they should. So what problems can predictive analytics help the CMO to solve, how do you get started, convince your executive peers of the need for such an investment, and gather the right data to achieve the right outcome?

Big data battle

Predictive analytics is not a marketing-specific methodology or activity, but it is gaining wider application in this sector for a couple of reasons. The first is the availability and cost-effectiveness of technologies delivering the horsepower to churn through immense amounts of data. Increased levels of automation in the marketing function, along with the push to understand consumer behaviour through insights hidden within big data, are also prompting more marketing chiefs to adopt predictive analytic solutions to drive sales and efficiency.

Well-entrenched applications of predictive analytics already exist outside the marketing sphere such as fraud detection and credit risk profiling, while regular users of the methodology include law enforcement, government and pharmaceuticals. Of course like most things, predictive analytics as a concept and intention is not new, and has its roots in data mining, response modelling and statistical regression.

Eric Siegel is founder of Predictive Analytics World and author of the book Predictive Analytics:The power to predict who will click, buy, lie and die. He whittled down predictive analytics in marketing to two main goals: Who is going to buy, and who is going to cancel. The defining characteristic of predictive analytics and what separates it from forecasting is the ability to generate predictions for each individual customer or prospect.

“It’s the holy grail of marketing – to proactively pounce on every individual customer opportunity,” Siegel told CMO. “Making a prediction about each individual is not such a crazy idea. You could have a business rule for customers who fit into a certain segment based on profile, purchases of a particular product, or geography. That’s the ‘if’ part of the rule. The ‘then’ part of the rule might be that these customers are three times more likely than average to cancel their subscription.

“Whatever the business problem is you’re trying to solve, you’re applying these rules to assign a probability to an individual. It’s not about making accurate predictions, it's about predicting outcomes significantly better than guessing.”

Like most marketing innovations today, data lies at the heart of successful predictive analytics projects. All industries are getting excited about data because it represents experience and an aggregate recording of things that happened in the company or brand’s history, Siegel claimed.

“Predictive analytics is technology or methodologies that learn from that experience and works out how to predict,” he explained.

“There is a lot of excitement about big data right now, but the discussion often sidesteps the most salient question, which is: ‘What is the point of that data and where’s the value?' The most actionable thing you can get from data is predicting. These predictions directly inform the action, treatment, recommendation, contact or retention offer on a per-customer basis across millions of customers. It’s the automation of millions of decisions based on millions of predictions.”

What predictive analytics is doing for marketers

Among the applications of predictive analytics in marketing are recommendation engines for cross-selling and upselling, customer churn, and retention programs. John Elder, the founder of data mining specialist consulting group Elder Research, said text analytics is another area gaining popularity, and some companies are even starting to use link analysis to identify connections between customers and account holders.

The US-based company has customers from all industries and describes predictive analytics as a way of addressing the “needle in a haystack”.

“The meaning of predictive analytics is relatively simple: It gives me the ability to help my sales team focus on the best opportunities,” Mindjet global CMO, Jascha Kaykas-Wolff, said. “The complications and complexity of the way you deploy predictive analytics feeds into that equation, but put simply, it’s about providing clarity to the sales organisation about which opportunities they should talk to.”

The US-based company provides collaboration and project management tools and claims 83 per cent of the Fortune 100 as customers. It has already invested in employing a third-party agency to undertake several predictive analytics projects and is now looking to invest in a software solution.

Alongside the technology improvements, Kaykas-Wolff attributed the rise in predictive analytics to a change in the way companies structure their sales capability. “Sales organisations are increasingly looking at traditional enterprise sales as being inside-sales driven,” he claimed.

“The model we had in the past of expensive enterprise sales people in the field, looking for a specific buyer, doesn’t make sense anymore. The dynamic today is to have groups of people inside your company looking at opportunities opening up through your website and digital channels, who then try to mine through that data about those prospects and customers and ultimately convert them.

“When you have a lower-cost product and sales organisation, you need a way to make sure you’re filtering out the deals that will take a long time to close and bring the cost of sale to a level that suits the business.”

Read more: How predictive analytics is tackling customer attrition at American Express The top 3 directions predictive analytics will take in 2015
MasterCard buys predictive analytics provider APT for US$600m

Predictive analytics in action

One of the hottest applications of predictive analytics is around customer churn or attrition. A local example is American Express’s global B2B marketing team, which is using predictive analytics to identify at-risk customers who otherwise look healthy in its database. Through a pilot project, the company was able to improve identification of attrition risks by 8.4 times, and arm their marketing and sales staff with more accurate lists of individuals to target with retention campaigns and communication.

Another high-profile and controversial international example of the insights that can be achieved through predictive analytics is from US retailer Target. The company successfully predicted one of its customers was pregnant before her father knew when using predictive analytics to apply a pregnancy prediction score to new parents-to-be.

Siegel estimated organisations using predictive analytics to solve a particular customer problem will increase response and/or decrease costs by 15-30 per cent.

“You can’t afford to give your retention offer to your entire customer base, so it has to be effectively targeted,” he said. “If your top 40 per cent of customers actually includes 80 per cent of those who will respond, then you’re going to cut your costs by 60 per cent, because you’re going to suppress the other 60 per cent of your customer list and only sacrifice 20 per cent of the sale. That makes the bottom line skyrocket.”

At Mindjet, predictive analytics is being applied to inbound trials of its software and activity inside those trials. Resulting data is then fed to the sales team to direct activity.

“The relationship between predictive analytics and the marketing and product teams is pretty important as well,” Kaykas-Wolff said. “Because we are in the SaaS [software-as-a-service] business, the relationship between those two data sources is important to use.”

Page Break

Lattice Engines is a consultancy group which uses predictive analytics to help brands better their sales and customer insights. Clients include Staples, Adobe, Wolters Kluwer, Hewlett-Packard and Microsoft.

According to founder and chief, Brian Kardon, companies employing predictive analytics see an average of 20-40 per cent improvement in revenue. As an example, he pointed to his client, Internet networking vendor Juniper Networks, which looked to better identify suitable prospects for its sales team to pursue.

“There are thousands of variables that can trigger people to buy switches and routers, such as a change in the company’s size, or a new CIO,” Kardon said. “Using predictive analytics, we identified one variable which accounted for 60 per cent of an account buy, and that’s companies that have signed a real estate lease within the past 30 days. These companies are 10 times more likely to close than someone who hasn’t moved offices.

“Could a sales rep find that out by themselves? No, because every organisation and territory has a different way of filing leasing information. We also identified five products out of Juniper’s 500 SKUs that are most likely to be purchased by those signing leases. So instead of focusing broadly on prospects, we enabled Juniper’s sales team to focus on signing up key accounts that were almost certain to convert.”

The result was quicker sales and conversion rates, as well as an improved performance right across the sales cycle, Kardon said.

In another example, Lattice Engines worked with the life sciences division of GE to find out who its sales team should target first to purchase its specialised lab equipment. “What we discovered was that those that had just received a government grant were eight times more likely to purchase equipment,” Kardon explained. “But when it came to labs that worked on a particular type of protein, we found they were 50 times more likely to buy GE’s lab products.”

By using historical data to learn from a company’s previous customer transactions and history, Lattice Engines was able to apply the knowledge to an Internet and academic search to find new prospects for GE. For both Juniper and GE, the process involved looking at thousands of purchases and different attributes of its customers over a two-year period.

In the case of its work for Dell, Lattice Engines helped the sales team understand what products are best pitched at which existing accounts in order to improve the technology manufacturer’s upsell rates. This involved looking for patterns in the purchase history, as well as researching existing customers looking for new job listings, recent venture capital injections and office relocations.

“Many companies know the business triggers but can’t prove it, or have this kind of ‘tribal’ knowledge within their sales division,” Kardon said. “By using big data, we can prove and predict where best to position efforts for maximum return.”

How CMOs make it happen

Anametrix CEO and former CMO, Pelin Thorogood, claimed marketers are in “spring training” right now when it comes to interpreting data and utilising predictive analytics. The company produces business intelligence and data analytics technology tools to help companies drive better data insights.

“Part of it is data scientists and the people manipulating the data don’t understand the business side, and the people understanding the business side don’t really know how to manipulate the data,” she said. “It’s important for people to speak each other’s language so the analysis is done in a much more holistic manner and they can truly answer the right question. There needs to be a lot more conversation between the analytics/data science people and the business people – the general and category managers, plus the brand managers who own the product category and its P&L.”

There are a host of technology products, third-party agencies and data scientists available on the market today that actually do the hard work of predicting but none of these will be successful without certain elements and processes in place.

Firstly, having a person to champion your predictive analytics project is vital if you’re to gain a measurable return on investment. “I often recommend my customers watch the movie Moneyball because it’s a great example of success in this field,” Elder said. “Brad Pitt plays a real-life person who used analytics in baseball to great effect. He wasn’t the analyst himself, but he believed in it, protected the work being done and made decisions using the data. He was the hero of the story and earned rewards as a result.”

Having accessible, historical data in the right format is another must. Whatever predictive analytics tools, third-party consultant or in-house staff member you employ to do the predicting for you, they will require data and snapshots of customers and information at various points in time if they are to learn how to predict.

“Predictive analytics is never going to solve bad hygiene in sales processes and product marketing,” Kaykas-Wolff pointed out, adding CMOs need to have a deep comprehension of how predictive analytics works at least in principle if they’re going to get significant returns. “You have to be clear on what your targets are, your pain points, value proposition and on top of that, the price has to match the market needs.”

“You also have to understand your business,” Elder continued. “We could be great at the data science component, but the best results come from being teamed up with someone who knows their industry, domain, data and the problem so we can learn from each other.”

Thirdly, CMOs must seek senior executive support, as well as access to plenty of data, Kardon added. “We often find data is kept in siloes where marketing has website information, sales are holding onto CRM and the IT department has purchase history. No one person can make this effective,” he said.

“You must first build the case with the CEO. Then team up with sales, IT and your support desk. A CMO can choose the predictive analytics technology, vendor or solution, but when it comes to big data it’s about collaboration and data sharing.”

It’s also important to remember predictive analytics is only useful if you act on the information generated. CMOs need to build a plan for what is going to be predicted and what they are trying to achieve. In addition, they must look at how best to give the results of their predictive marketing efforts to the sales team. In this case, Kardon advised integrating the results into the company’s CRM, and rank customer lists in order to those more likely to buy.

“You have to flesh out what the business value is going to be and how to act on that prediction, or you’re just putting the cart before the horse,” Siegel said. “Is the marketing staff willing to effectively throw away the bottom 40 per cent of the prospect list by supressing it? As a marketing chief, you must do whatever it takes to make that operational change.”

If you’re not 100 per cent clear on what the output will look like and aren’t willing to incentivise behaviour in the sales organisation or marketing team, it’s not worth making the investment, Kaykas-Wolff argued.

“The increases in customer conversion will pay for the technology multiple times over, so often the CEO or CFO will see it as a good investment,” he claimed. “It’s really the behaviours and changes where the rubber hits the road, and that is the sales organisation. Are they going to accept the data? Where are they going to see it? Will it sit inside of Salesforce, attached to documentation, is it going to change the business process or change the sales pipeline?”

For Thorogood, next-generation analytics is about finding new information, not proving your own suppositions. “The key is using the data and analysis to eliminate things you may have not even thought about before, instead of using data to support existing premises, which is really using it as a crutch,” she said. “Keep an open mind and leverage data through the edge cases, rather than the cases in the middle of the delta. And really look at it with a holistic understanding of the business, and optimise the forest, don’t optimise the tree.”

Whatever approach CMOs decide to take, predictive analytics needs to be treated as a business initiative, not an IT one. Siegel believed most companies have the potential to do more and better act on the data, while Elder saw further opportunities for predictive analytics around sales attribution across channels and has started pioneering this approach with one of his clients.

“One of the nice things about data mining is that it usually has a direct bottom line affect,” Elder said. “But what I would point out is that better companies do things they know won’t make them money but will educate them about their customers. Learning what the response rates are in different groups, even if it won't generate a direct return next week, still pays off long term.”

Predictive analytics gets closer to mainstream

Predictive analytics for the masses is getting ever-closer, if recent technology vendors moves and investments are any indication. In more recent months, US-based predictive applications vendor, Infer, secured US$25 million in a series B funding round led by inside investor, Redpoint Ventures. The latest investment brings the company’s total financing to $35m.

Infer is using the funds to accelerate plans to bring predictive analytics to all types of sales and marketing organisations. The company, which was established in 2010, has a host of technology customers including AdRoll, Cloudera, New Relic, Optimizely and SurveyMonkey, and claims to have grown revenue bookings by 150 per cent quarter over quarter. Other investors include Andreessen Horowitz, Social + Capital Partnership and Nexus Venture Partners.

Framed Data, another startup in the predictive analytics technology sector launched in August 2013, has also raised $2m from Google Ventures, Innovation Works, Jotter and NYU Innovation Venture Fund as seed funding for its operations. In a blog post, the company said the capital will be used to build out its “prescriptive arm”, which aims to automatically convert predictive insights delivered through machine learning and data science into actions for businesses to use.

The company claims to have hundreds of companies using its predictive analytics engine. According to a story on Techcrunch, Framed Data doesn’t intend to compete with data analytics companies, but instead, runs their data through its machine learning models to generate predictions about user behaviour.

Its founder, Thomson Nguyen, said the product has been expanded from looking at users who are about to leave a brand, to now predicting users who are likely to upsell or pay for a premium service.

US-based big data and predictive analytics specialist, Cognilytics, was also in the news in late 2014 after being snapped up for an undisclosed sum by US telecom, CenturyLink. The 200-employee firm provides mid-size and large enterprises with big data deployment, management, advanced decision making, predictive analytics and data visualisation capabilities and is an SAP Hana partner.

And in June 2015, AOL made public its acquisition of predictive analytics vendor, Velos. AOL promptly shut down the startup service, and has yet to specify how it plans to bring the capabilities into its wider offering.

Other established enterprise vendors, including SAP, Software AG and Razorsight, also announced expanded predictive analytics capabilities in June.

Ovum senior analyst for information management, Surya Mukherjee, told CMO that predictive analytics had always been an “arduous goal” for data analysts, and has failed to impact the entire enterprise so far because it has been kept in the hands of power users and analysts.

“It really surprises me, even after years in business intelligence, how little of predictive technology has actually flowed beyond data geeks and statistics majors,” he commented.

“Newer approaches to predictive focus on bringing predictive to the masses, and that is where I see a majority of capital injection coming from in this space. Predictive has to transcend its ivory towers and flow down to the business user to be widely adopted.”

One acquisition designed to deliver predictive analytics to the masses was SAP’s acquisition of KXEN in September 2013, Mukherjee said. Another emerging example of more accessible predictive analytics capabilities is IBM’s Watson analytics

Infer and Framed Data are both about solving specific predictive problems with automation, allowing marketers the opportunity to better utilise these analytics processes, Mukherjee said.

TDWI Research analytics analyst, Fern Halper, also saw growing interest in predictive analytics to understand customer behaviour led by sales and marketing areas.

“Companies clearly want to predict customer response to direct marketing campaigns and be able to upsell or cross-sell a customer,” she said. “They also want to be able to stem customer attrition. The use cases for predictive analytics in sales and marketing are growing.”

As the market becomes increasingly aware of the power of predictive analytics, Halper said vendors are trying to make predictive analytics easier to use and offer it in a way that is consumable by a variety of end users.

“Many vendors have tried to make predictive analytics more ‘user friendly’ by automating some model-building capabilities,” she said. “They are including better visualisation capabilities to aid in pattern detection. They have also introduced ways to operationalise predictive analytics in business processes, which has opened up the technology to more end users.

“Predictive analytics is also being offered more frequently in the cloud. New startups are leveraging the power of the cloud to offer specific kinds of solutions – such as in sales and marketing.”

Mukherjee agreed vendors now have a focus on greater automation for predictive tasks and improving user experience, and added there’s also an emphasis on integrating big and small data as well as achieving near-time analysis.

“We believe that true democratisation of predictive analytics will happen from both the demand and supply ends; meaning, on the one hand, that automation of predictive technologies will make it easier for a larger group of users to take advantage of predictive,” he said. “On the other hand, there will be an increase in the number of people with data analysis skills.”

So are you ready to predict?

3 simple steps to start predicting:
  1. Apply predictive modelling to analytically learn to predict;
  2. Apply what has been learned to assign a predictive score for each customer;
  3. Act on these scores. It could be as simple as sorting a list and putting a yes or no next to each prospect for your email marketing campaign.

More on analytics:

5 big ideas to profit from analytics and big data
How predictive analytics is targeting customer attrition at AMEX
The keys to smarter data analytics
4 barriers stand between you and big data insight

Follow CMO on Twitter: @CMOAustralia or take part in the CMO Australia conversation on LinkedIn: CMO Australia.