A Big Data Q&A with Carter Mills

More and more companies are asking us how to leverage vast quantities of data to influence tactical and strategic decision making. It’s undeniable that there is great value in the many internal and external sources available. However, integrating, transforming and making that data actionable seems to be harder than ever.

As big data analytics have emerged, the skill sets and practices governing it have changed. What I’ve learned is this: The data scientist’s ability to look at things differently is the deciding factor between being overwhelmed by everything obtainable and being able to zero in on exactly what needs to be done.

Since big data analytics has evolved organically from traditional data analysis, the natural output is frequently more of the spreadsheets, crosstabs and PowerPoint decks that have traditionally been created—a lot more. In fact, many times a standard approach is to point a revised set of programs at a new, larger dataset and then append the results to an existing executive report. Very soon the audience, wowed by the effort that must have gone into such a voluminous analysis, finds the results too much to digest and too confusing to make any tactical decisions.

Whether you are working with transactions, social media, free-form text or smart meter data, there are three truths in getting the most out of big data:

1. You must approach problems from a different angle, often from the ground up.
2. These projects often have significantly longer timelines than traditional analytic projects. You must be prepared to commit the time and resources necessary to reap the full benefits possible.
3. Even though the input, process and analytics may be complex, your deliverable should not be.

Here are some of the questions we’re hearing from our clients:

Q: There are a lot of data out there, and it seems easier to collect but harder to pull together in a concise, visually appealing manner. What are some best practices for compiling actionable output from big data and for making big data results visually appealing?

A: The answers to both of these questions are found in what any good analyst would do if they weren’t busy trying to use big data to fit their organization’s historical mold. To steal a line from Stephen Covey, “Begin with the end in mind.” Many seasoned analysts approach a problem by focusing first on the statistical technique that should be applied to achieve the outcome. When working with big data, much of the time and effort is shifted to the beginning of the project. Before you think of tools, techniques and output, ask yourself these questions:

1. What problem are we trying to solve, and what is the desired business result?
2. What infrastructure and resources do we need to address this problem?
3. How should the data be integrated and aggregated?

I know this sounds a lot like project management and not statistical analysis. But, a good design will make your overall effort faster, more flexible and more actionable over time. One of the problems with big data is that it’s, well, big. Design a project poorly and when a C-suite executive asks to see your results a little differently, it can mean rebuilding the entire analysis. Furthermore, this could mean a restatement of what you’ve already presented, which can undermine credibility.

As far as compiling visually appealing data, think “less is more.” Just because you can show a customer’s activity in 15-minute increments doesn’t mean you should. Consider your audience and your original business objective. In addition, much of big data analysis is about examining things over time. As questions arise, resist the urge to “bolt on” another set of results into an existing report or dashboard. Since the data available are so vast, one can quickly marginalize the value of the output by making it confusing and unfocused. Finally, create and communicate metrics that don’t require an advanced mathematics degree to understand and take action. Remember, the end users rarely have an analyst background.

Q: It seems like clients always have to sit through 200-page presentations of meaningless data, waiting for the actionable nugget. With big data this can only get worse. How can vendors use the more data available to get to what the client really cares about?

A: This question made me laugh out loud because it can be so true. However, it’s not just market research firms—any analytics agency can fall into this trap. Over a decade ago, an agency I worked for created a huge analysis binder for a large, not-to-be-named company that used to be good at renting movies to folks. It was filled with every imaginable crosstab and statistical test that could be performed on a set of tables. I asked my boss, “Why such a thick binder when only the last two pages matter?” He responded, “When a client pays this much money for an analysis, they expect a huge binder they can point to on their bookshelf.” Now that big data has arrived, we need a cultural shift.

Granted, a thick presentation can be used as a reference document for all sorts of unasked questions. But, there is no denying it is perceived as a tangible validation that the effort matched the dollars spent. In today’s world, the delivery hasn’t changed with the times or the technology. We must acknowledge that most big data projects require time to design, set up, integrate, compile, explore, aggregate and socialize before the original analytical endpoint can be addressed. That’s a lot of money spent before the first deliverable. Agencies and clients must collaborate to set expectations early and then follow a few simple rules:

1. Review and agree on the projected timelines and milestones for the project.
2. Create a solution that has value beyond its initial purpose.
3. Understand there might not be any immediate “silver bullets” (see point 2).
4. Give the client access to the data, if they want it.
5. Tell a good story, get to the point and move all that other stuff to the appendix.

While big data projects require a lot of exploration, the best designed ones answer a specific business objective. As long as both parties keep this in mind, long presentations that simply regurgitate facts can be kept to a minimum.

Q: What skill sets should our internal and external specialists have to get the most out of our data?

A: In a word, curiosity. Yes, you need to hire people with degrees in econometrics, statistics, mathematics and database architecture. Those skill sets are essential. However, if the analysts approach your business problems by finding the quickest route to an answer, you will never get the most out of your data, no matter how advanced the analytical technique.

I recommend hiring “Knowledge Miners.” These are individuals with the technical skills, business acumen and curiosity to look at things differently. They generally don’t care for the way things have been done before or use pre-aggregated data. They combine things, create things and pull from many different sources. They enhance your existing data instead of rolling it up so they can build a model. The one thing they need is time. If you have traditionally built an attrition model in two weeks, bringing a Knowledge Miner into the equation may double or triple that time. However, in the end, you will have a better model and many new variables you can repurpose for other efforts such as trigger programs or research studies. The trick is finding people that can balance that curiosity with actionable and timely results.

Q: In the energy sector, what efforts are being taken to segment and predict customer behavior, specifically as it relates to future payment behavior, customer satisfaction and participation in utility services such as E-Bill and Direct Debit at the time of enrollment?

A: Here, the idea of big data is focused on data integration and test design. You are trying to build a model or create a set of business rules that will show a customer’s propensity to exhibit some sort of future behavior without the benefit of historical data. A good start would be to:

• With a well-planned design, make your customer data as rich as possible at sign-up through integration of as many relevant data sources as you can.
• Create a rigorous test design to ‘test into’ each of the behaviors you want to predict.

To explain further, you want to make your customer profile information as rich as possible. Create a well-designed, integrated, customer-specific record from day one. When the customer walks out of the door you should have captured all the operational data needed to service the customer. Also, consider having things like credit report information, a geo-demographic profile, microclimate telemetry, detailed home information and appended third-party data enrichment variables. Furthermore, build a critical question survey right into the sign-up process. The key is to avoid missing values. You want to have this data on every new customer you acquire. This may require infrastructure changes as well as customer service representative training, but it will be worth it.

As this is established, set up your design of experiment around each of the programs you want to predict and integrate it into your enrollment process. Create an offer group and a random control. It is possible to get fancy and have a multi-factorial design, but it isn’t necessary. The most important thing to remember is this will take time, and you need to be comfortable with that. With big data and data sciences, you aren’t trying to create something once and call it a day. You want to create something that will deliver results for years to come.

Q: Are there any plans to study the impact of smart meters; the process changes, efficiencies, etc. and the impact on customers and customer satisfaction?

A: Absolutely. Almost everyone is looking at smart meter data for the first time. Most of what I’ve seen revolves around how much electricity the customer is using and when. Some of the more interesting variants center on trying to catch customers that monkey around with the hardware to steal power. However, rumor has it that a savvy analyst will be able to use telemetry data to identify when a customer turns on a specific appliance in their home, and if that appliance appears to be as efficient as possible. As this is refined, it will give the provider the opportunity to create targeted programs such as free in-home audits, no- or low-cost upgrades as well as many other prospective programs that can help consumers save money and increase satisfaction.

Q: In order to view the customer experience from all sides, are there plans to match customer transaction research with on-going satisfaction studies to provide a 360-degree view?

A: We are already doing this for our clients, including a few from the energy sector. From this point forward I believe merging behavioral data with attitudinal research will become standard practice. It gives the analyst the ability to target surveys better, ask fewer questions and dramatically enhance the research gathered. I believe this will be one of the fundamental changes that will dramatically increase the insights we have in every aspect of our work. Not only will it enhance the research studies, but it will also be used to drive segmentation, program creation and internal tactical and strategic decisions.

A Brief Summary

All of the above can be summarized in these few points:

• Big data looks like more of what you’ve always had, but it’s not. If you treat it as if it is, at best, you will continue to get more of the same.
• Working with big data takes more planning and time than you may expect. However, the payoff can have a dramatic impact on the daily tactical and strategic decisions that will drive your company forward.
• Data integration and transformation is the key to creating long-lasting value, which will deliver results for the foreseeable future.

While the industry is still in the early stages of working with big data, I believe companies that successfully leverage this new landscape in the next 18 to 24 months will have a significant competitive advantage for years to come.

< Back to Library