I Was the Black Box
A black box is a complex piece of equipment with contents that are mysterious to the user. Inside it could be whirring gears, complex databases, or a giraffe – the user doesn’t know, but don’t assume he doesn’t care.
I know that problem well, since during the past five years I have spent an incredible amount of energy trying to force a marketing solution from a “big data” platform. Let me tell you what I learned.
Faced with intractable marketing problems like churn, next best action/offer, and personalized client communications, marketers were hungry to exploit vast amounts of customer data. They invested heavily in big data appliances, teams of people, and a variety of other expensive tools.
But they mistook tools for solutions. My clients were many times left with things they couldn’t use, didn’t trust, or were too inflexible to meet their deadlines. To find solutions in the tools, I was brought in to work the tool bench; to fiddle, to grind, and to make pieces fit. It became an exercise in the esoteric and the solutions were not easily repeatable. I now realize that I had become the Black Box, and the clients themselves could not peer inside, let alone climb in.
What my solutions lacked were three simple things that marketers needed:
- Simplicity and Independence
- Raw Speed
- Built-In Measurement
For a marketing tool to become a solution, it has to solve these three things.
Simplicity and Independence
Q: Does the absence of simplicity define the black box?
A: I had a pattern in my work. First, I met with the business and gained an understanding of their use case and requirements. Next, I loaded data into a platform from multiple customer sources. I then iterated over and over again until the right model was formed. A few weeks later, I packaged up my work and presented it to the business. My clients were generally appreciative of my efforts but they always had questions. They were often skeptical of the results or weren’t sure of what to do next. If you look at the underlined words in proceeding sentences, I WAS THE BLACK BOX. The client depended on me to understand their business use case, wrangle the data, solve the problem, and help them understand it. Why? The tools I used were complicated to use and disconnected from their business universe. The solutions I built were overly technical and technology speaks one language and marketing speaks another. No matter how hard I tried to simplify, or help them understand, the business was skeptical of my work.
My interface for the past four years was ANSI SQL and the command line. It was a pleasure for me, a ‘Windowed’ interface slows down a developer! But the business user ran away in horror. I cringed every time the customer asked me to show them how the sausage was made. A well constructed graphical user interface enables user productivity and provides lesser cognitive load. The reason I am with ActionIQ is because we understand the value and importance of an easy to use interface. Our first guiding principle: User interface is a constant in simplicity.
Also, don’t forget how important the ‘data integration interface.’ The three most expensive letters in Data are: E-T-L (Extract/Transform/Load). If it is difficult to get my data from my customer sources, make use of it, and then integrate my answers into other systems then my solution is siloed! This is another guiding principle of ActionIQ. The graphical interface must support easy data integration in both directions: inbound and outbound.
Q: When time is the essence, am I at the mercy of the Black Box?
A: We all remember as kids the story of the tortoise and the hare. The moral of the story was ‘slow and steady wins the race.’ I agree being steady wins the race but the game has changed. Your competitors want you to be the tortoise while they become the smarter, governed, and disciplined hare. The hare was able to learn from its mistakes and is now steady too.
Timing is a very critical component of messaging to clients. Take this real world case for example: It is six business days before a major shopping holiday and your competitor just launched an offer that could create customer defection. You have to react, and you have to do it as soon as possible. It’s all hands on deck and the first people you call are IT and data scientists. The faster you respond the better, but data science takes time and you are at the mercy of your traditional technology investments. It gets done but a few days later than you would have liked. Seems a bit dramatic but this does happen frequently to many subscription based businesses.
At rest database applications are getting faster but queries still take a long time. The guiding principle of AIQ is: Answers must come back in seconds not hours or days. This is compounded by the need to iterate and test in an environment shared by many users. What really matters is the speed at which I can determine two things: how quickly can I recognize that I have failed and second, how quickly can I change direction and try again. The goal is to shrink this window of time so that the user becomes the bottleneck and not the platform.
Q: Can I accurately measure the Black Box?
A: All successful projects have a First, Middle, and Last mile of its lifecycle. With analytic projects the First and Last miles are often overlooked or not given attention. Most data scientists focus on the “Middle” Mile or what I call the Analytic Mile. This is where the analytics are built and tested and finally render a score. In the First Mile we determine the use case, build the team, set goals, define how we are going implement, and measure success. In the Last Mile we actually put the analytics to work given the goals and measurements defined in the First Mile. This Last Mile efforts are where the measurement is done to see if our solution has accomplished the goals of the effort.
A bank in the North-East United States gave me a bunch of data about their account holders. Standard things like demographic attributes, account transactions, and some behavioral data from its web sites and call centers. I quickly turned around a list of 8,000 people who will churn out or defect in the next 90 days. I also showed them the reasons why it was going to happen. They said thank you but that they wanted to prove if I was right and see what the next 90 days showed. They operationalized my analytic by just watching the list to see what would happen. My phone rang about 60 days later and it was from the bank saying that I was right 80% of that list you gave us has defected. I asked them why they didn’t campaign and reach out to those customers? We didn’t believe your prediction would come true and all we had was a manual process to track success. We couldn’t close the loop.
There are two guiding principles: One: Analytic projects that are deployed inside of a business process are highly valuable. Two if you cannot measure the value of a deployed analytic, you shouldn’t do it.
Here’s the Takeaway
For the first time in history we have a glut of customer data from many different touchpoints. This isn’t a ‘Big Data’ problem it is a HUGE Data problem. Data Science is an essential need for all organizations wanting to survive the new ‘Customer Intelligence’ era. However, we have to free up these people so they can constantly tackle higher level needs and continuous innovation. Data Science, coupled with Big Data, will fuel the inventions that will drive growth and organization sustainability. However data science and the tools, are very complex and require people who are highly skilled in mathematics and programming. Data Science and Big Data are the Black Box. What is needed are solutions that allow people to develop solutions to lighten up that box. Tomorrow’s ‘Smart Data’ solutions need to be easy to use, nimble and flexible, and provide closed loop measurement in order to calculate ROI.