TDWI Upside - Where Data Means Business

Q&A: The Insights Behavioral Analytics Offers (Part 2 of 2)

Advanced behavioral analytics can help organizations analyze consumer behaviors and then alter applications to drive more profitable transactions.

In the first part of this discussion based on a recent TDWI Webinar, we explored definitions and use cases for behavioral analytics. In this part, industry analyst and data expert David Loshin discusses different methods in advanced behavioral analytics and considers what factors into a buy versus build decision for rapidly deploying an advanced behavioral analytics system.

Loshin, president of Knowledge Integrity, Inc, is a recognized thought leader and expert consultant in the areas of data quality, master data management, and business intelligence and a regular contributor to Upside.

Let's talk about specific methods in advanced behavioral analytics. First, what is cohort analysis?

Cohort analysis is an approach in which we look at subjects grouped by common characteristics, common experiences, and common time frames. We then look at sequences of events for these groups in order to understand whether they are doing what we want them to be doing on a regular basis. We monitor the continuity of that profitable action. Essentially, we want to identify points of attrition among similar subject groups.

Let's take a healthcare example. Imagine that we want to look at whether patients are continuing to refill their prescriptions at the expected rate. The doctor prescribes a hypertension medication that needs to be refilled monthly. If we're monitoring to make sure those patients are continuing to refill the prescription, we can see whether they are behaving as desired. We are monitoring the continuity of the action and looking for points of attrition so we can determine if there are behavior patterns.

In another example, a company might be emailing a marketing newsletter monthly. It wants to monitor continued reading of the newsletter, including which items the target audience is reading, for how long, and whether they are reading them at the same rate they were initially. That can all be examined in terms of how the subjects of the analysis -- the newsletter recipients -- are behaving when grouped by common characteristics and experiences.

Finally, an example of cohort analysis in automation is monitoring whether IoT -- Internet of Things -- devices are streaming communication as expected. You might monitor to make sure that they are continuing to perform updates as planned, looking at the values and making sure they are meeting expectations.

So that's cohort analysis. What about funnel analysis?

Funnel analysis, as the name suggests, looks at how participation narrows as people move through events along a particular desired sequence.

Going back to the marketing campaign example with the newsletter -- maybe the request in the email is to register for an event. Maybe during the event there is a paper that is promoted for download; we might look at who actually takes that action. As you move along the stages of the process, you see a reduction in the number of individuals who have taken each action.

We do funnel analysis to figure out the rate at which individuals follow the set sequence of defined tasks. Essentially, you want to assess the conversion overall -- who are the people moving through each stage? Who is falling out of the sequence at different stages along the way? How many people read the newsletter but didn't register for the event? How many people who registered actually attended the event?

We can use cohort analysis to look at other paths as well. We might look at people who registered for the event but didn't attend, or people who at some point downloaded the paper, or downloaded the paper but didn't attend. We also want to look at people's profiles and compare them to paths taken; doing that allows us to find opportunities for conversion to that final sequence, the one we really want people to follow.

We can also understand where the biggest drop-off is and how we can change our process to reduce fallout and drop-off rates. Overall, we want to improve the number of individuals who make it to the end state.

Funnel analysis lets you examine situations where people start out along one sequence and then find out when and who -- and eventually why -- they do not finish the way you expect.

What about path analysis?

That's looking at the actual paths taken by different parties through different stages of the process. Take an e-commerce example. When you visit an e-commerce site, you might first search, then look at one product, then one recommendation. You might then go back to your original search results and view another product and another recommendation.

As we examine this path, we can see that over a sequence of events -- or a session that an individual might participate in -- there are many stages or actions that the person takes before reaching the end of the session. We want to find out at what point on the path people are reaching a desired state, and at what point people are reaching an undesired state.

We want to collect all the paths and decision points that individuals are taking and look at the experience overall to figure out which paths are faster, or lead to a desired result in a more efficient way. We want to identify where the streamlined paths are and where the barriers are.

Maybe there's a stage that everyone who fails to complete the path goes through, so there's something about that stage or action that is discouraging individuals from moving forward. Path analysis means looking at all that.

These are all different aspects of behavioral analytics?

Behavioral analytics, again, goes beyond just doing customer profiling. It can give us insights into why people are doing what they are doing and at what point they diverge from the path you've laid out. It's definitely valuable to your organization. The question really isn't whether you want to do behavioral analytics, it's how can you integrate it more quickly into your organization.

Can you discuss the build-versus-buy decision when it comes to a behavioral analytics platform?

In any situation where you want to integrate this kind of capability, you have the opportunity to choose between building it in-house and buying it.

If you want to build it yourself, you may need to redesign your existing platform to accommodate a new type of data capture. For example, you might have a system that logs all of your transactions in a relational database but doesn't give you the visibility into the sequence of steps an individual takes that you need for path analysis. In that case, you would need to redesign your existing platform to accommodate path data management.

Also, your existing platform may be constrained in terms of storage space and execution performance, especially if you're talking about millions of individuals performing hundreds of millions of actions on an ongoing streaming basis. That can be a significant resource hit.

Third, you may need to design and develop the algorithms and processes for behavioral analytics. Finally, you are going to have to do project and resource planning before putting everything into production, and that creates a lag in time to value.

Buying a solution, on the other hand, can allow you to leverage newer platform technologies, especially if it's hosted in a cloud-based system. The vendor has probably done similar behavioral analytics deployments a number of times, so you can benefit from that experience; they can essentially give you a jump-start.

Typically, the vendor will have an optimized path for data storage and behavioral algorithms. You won't be penalized in performance from having to store and analyze millions of different transaction sequences.

Also, the solution can allow you to leverage access to a variety of external sources, especially when you're trying to combine analytics and profiles with your behavioral analytics. That means you can look at streaming data feeds from social media sources, for example, or data acquired from third parties, or perhaps your own data in different cloud-based systems.

Finally, as with just about any buy-versus-build solution where you elect to buy, you get faster time to value. If you're not an expert in behavioral analytics, it may make sense for you to purchase a solution rather than build it.

What are some things to consider with any behavioral analysis platform?

Whether you decide to engage a vendor or put it together yourself, you'll need a platform that provides scalable storage and performance. You also must have behavioral analytics technology -- the tools and the models -- fully integrated into the system. You'll need a data organization to represent paths for the sequences of activities, actions, and transactions, and you'll need the ability to manage a wide variety of source information.

The conventional on-premises data warehouse is basically not optimized for behavioral analytics, so you may want to consider a cloud-based or hosted solution.

A cloud-based solution is going to combine the cloud-based platform with the service provider's knowledge and experience, and that's going to increase your benefits. It will lower or perhaps even eliminate your costs for system management -- your system can be customized not just for behavioral analytics in general but for doing the types of analytics common within your industry or type of business.

That kind of solution is also going to support both structured and unstructured data. You'll be able to leverage the vendor's capabilities in connecting to external data sources. With a cloud-based environment that allows for it, you also have the benefit of storage and computational elasticity so that as your needs change you don't have to continually go out and acquire new hardware and software. It gives you rapid deployment, relatively low cost, and is easy to launch.

TDWI Membership

Accelerate Your Projects,
and Your Career

TDWI Members have access to exclusive research reports, publications, communities and training.

Individual, Student, & Team memberships available.