Predictive Analytics: Staying Ahead of Your Customers

Page 2 of 2


Charging customers the right amount for subscriptions has proved challenging for digital publishers, says Matthew Shanahan, Scout by ServiceSource's VP for product strategy. Scout provides a "customer lifecycle management solution" that helps digital publishers squeeze more revenue out of existing customers-in some cases, a great deal more. The service functions as a meter, measuring how much someone is using a digital product versus how much he paid for it, according to Shanahan.

For instance, if a subscriber to The New York Times isn't using his subscription enough-relative to payment-Scout will shoot an alert to the Times. The same thing would happen if the user was using it more that they should be. In other words, the service measures product consumption versus payment and is able to suggest price adjustments.

Shanahan says that digital publishers don't have the tools to measure usage on their own, so they lose customers who aren't using the service very often and miss the opportunity to raise subscription rates for heavy users. He says Scout uses "multiple factors" in making predictions, with usage being the key component. The cloud-based software can also predict who's likely to cancel so publishers can try to retain their business.

The company's solution is "highly effective," Shanahan says, and it produces an average of 10% to 15% additional revenue. Some customers have seen revenue increases of more than 30%, according to Shanahan.


Starting a PA project can seem to be a daunting task. There are so many things to consider-the quality of the data, getting executive buy-in, determining the goals of the project-that it's no wonder that most companies haven't tackled a PA project yet, says John Bates, a senior product manager of predictive marketing solutions and data science at Adobe.

PA experts offered some specific advice on how to jump-start a PA project, which is as follows:

Decide ahead of time what's being predicted-A guaranteed way to have a failed PA project is to have undefined goals. "Regardless of what you're looking for on predictive analytics, the first and foremost thing is to say, ‘What are the business objectives I'm going after?'" says Shanahan. "And a lot of times people are confused on that. They get right into technology without understanding what they're trying to achieve."

Elder says it would be frustrating to complete a PA project only to realize the predictions can't be used in a meaningful way. "If you can't do anything about it at the end, if you don't have any way to activate those audiences because all you've invested in is just data collection and analytics, then you're going to be disappointed."

Start small and evangelize-Dutcher says that Apigee provides many services to smaller startups, so size is not a barrier to entry to the world of PA.

"I encourage them to start small, rather than boil the ocean and start big because there's a higher rate of potential failure," says Bates. "Start with something small-a particular area of the site or a particular app, for instance. Get wins and then evangelize it internally. The organization will naturally take note of that, and depending on their level of maturity, they should begin to adopt it more broadly."

Don't wait for more data, as long as your data is good-Experts say that most companies probably have more than enough data to start a PA project, provided the data is accurate and is measuring the right metrics. Bates says that that web analytics data is sufficient and that companies can bring in CRM and any offline data over time. "You spend most of the time prepping the data," Bates says. "It's not fun-the predictive analytics part is considered fun."

Know what you don't know-PA is a science and an art, Elder says, and it could prove more cost-effective and easier in the long term to bring in expert help for the first project.

"Hire us or others to prove the worth of the technology," says Elder. "One thing we do is mentor people into growing into teams. We slowly work ourselves out of a job as we mentor. But there's a lot of work to go around so we don't mind doing that. And clients really love that-they love the ability to call on you to get the work done, but also to start to bring it in house."

Elder says that PA consultants can prove cheaper-even over the longer term-than trying to hire qualified but scarce data scientists and starting from scratch. Once you've had a "win," the company can continue with the consultant, develop or find a budding data scientist on staff, and bring PA projects in-house.

It's vital that the business people and the data scientist work closely together, so the project doesn't end up producing "accurate data on the wrong thing," Elder says. "The tech is not a silver bullet-a lot of it is soft skills, in some ways knowing where the mistakes have tripped people up before, knowing the importance of working closely with the people who understand the domain."

Check your company for hidden PA talent-Bates says that shaking your company's tree could prove fruitful. Perhaps there's a statistician in the finance department who's been itching to try her hands at PA. "Look at everyone who is sitting on the bus within your organization, and see if you might be able to reshuffle it in some way in order to get maximum output from that individual," Bates says. "The first resource you need to be concerned with is human resources and not the software," Siegel adds.


While it's difficult to predict the spread of any technology, PA experts are confident that the technology will continue to be adopted by both large and small companies. "We are past the tipping point where any company of a given size wants to use this proven technology to do better at what they do," says Elder. "And there are enough success stories out there that people are realizing it's kind of silly not to."

Apigee's Dutcher and Scout's Shanahan see a trend away from expert judgment in creating and modifying predictive models to computer-created ones.

"We're starting to see a larger trend toward letting the computer systems or analytics do the heavy lifting," says Dutcher. "What I think we're starting to see now is that the channels that you're interacting with your consumers have exploded exponentially and that's not slowing down; the information that you have on an individual has expanded exponentially and that's not slowing down."

This huge amount of information is starting to overwhelm even expert data scientists, Dutcher says, and a new paradigm will emerge in which the machines will "advise" the experts, as well as vice versa.

Machines can be constantly looking for new trends and patterns that a human may "miss because they don't find the little signal in the terabytes of data that is quite predictive or that has changed in the last three months because consumer behavior has changed," Dutcher says. In other words, standard human-created and maintained models can't react as fast as machine-generated ones can.

Shanahan says that the field is not very far away from self-healing predictive models. Humans won't be modifying models and making changes-computers will. "And that's probably only 2 to 3 years away," he says. "I know of two startups that are doing some very advanced versions of that."

Bates says PA is about to go mainstream. "I see predictive analytics requirements in almost every RFP [request for proposal] we get from clients," he says. "Midmarket or smaller clients are beginning to catch on to that as well; as they grow, their analytics maturity is also growing."

It sounds as if another C-level title-chief analytics officer-could soon become as commonplace as CFO or CTO.

Page 2 of 2