Potential: Revealed
Strategic Thinking, Innovative Ideas, Growth Marketing, and Revealing of PotentialArchive for experimentation
Predictive Analytics: How it Works (#2)
In the first post about predictive analtyics we learned about the essential building block of predictive analytics: the predictor. This is a value calculated for each entity (say, a customer) who’s actions or behaviors are to be predicted – for instance the recency, in months, since a customer’s last purchase.
Prediction power is enhanced if you use more than one predictor at a time. In doing so you are creating a model. Models are the heart of predictive analytics. In this post I’ll discuss how you can find the “best” predictive model. I put “best” in quotes because from a practical standpoint, unless you assume unlimited time and resources you may be best off finding a model that improves your results (e.g., reduction in customer churn) over previous experience. Today there is available very powerful modeling software and well-trained and talented statisticians, but the number of variables to consider in any predictive model (across demographics, transactions, behaviors) can be extremely large making determination of the “best” model cost prohibitive.
Fortunately, taking an incremental, continuous improvement approach can yield solid results for most any business and the promise that results will improve over time. A common tool is to develop a yield curve. For example, plotting the results of a predictive model for churn with amount of churn on the Y axis and percentage of customers contacted in a retention campaign on the X axis will show a curve the decreases to a point — i.e., up to a certain percentage of a universe of customers contacted, attrition rates will fall — but will bottom out and then move upward. Meaning that not all customers will respond to a retention campaign and you are best off contacting only those predicted to respond well. After that point, you are best leaving the balance of the universe of customers alone – either because they are not likely to churn anyway or because the predictive models say campaigns to retain them will be unsuccessful (and possibly other methods are needed – along with models that might predict how these approaches can be equally tuned to expend effort on just those predicted to be successful).
Now, although the model does not work perfectly, the socring and ranking of customers according to their likelihood to be retained provides clear guidance on how to invest in retention programs to yield the best results. It will prevent campaigns to retain customers that are too aggressive (trying to retain those that are not likely to respond positively, or wasting effort on those that are likely to stay).
There is a great deal more to predictive analytics than I’ve covered in the past two posts. But I hope one message is clear: you can gain practical improvements in marketing results or other customer touch points through the use of analytics that don’t need to be complex (at least to start) nor perfect. Commitment, willingness to experiment and continuous improvement are what’s really required.
Thanks for reading and I’ll look forward to comments.
Predictive Analytics: How it Works (1)
Sorry for the delay between posts. For past month or so we’ve been working on a very interesting project dealing with product ideas based on financial transaction data and powered by predictive analytics. While we are working to develop some early prototypes we have also been talking about challenges that need to be addressed when taking such products to market.
One issue over and over has been risk of market launch failure due to lack understanding of how analytics work (often lacking even rudimentary let alone deep understanding). A majority of key stakeholders – potential customers and internal business unit and functional area team – have heard of and are relatively convinced of the potential for analytics to optimize decision making. Whether that be to improve marketing effectiveness or precision of sales forecasts. Yet the basis for belief is often what they’ve read about or been led to believe by others. Analytics are not perfect and an important approach to achieving long term benefits from analytics is experimentation, challenging current results, and continual tuning of analytical models. We can foresee a gap forming where confidence in what is being developed and sold to clients falters due to lack of basic understanding of predictive analytics.
So, I thought I’d put together a brief series of posts (sort of like I did on “Practical Strategy” a little while ago) to explain predictive analytics.
The essential building block of predictive analytics is the predictor. It is a value calculated for each entity to be predicted – for instance the recency, in months, since a customer’s last purchase. Typically, the higher the calculated recency the more recent was the last purchase. As you’d expect, a good predictor is usually a reliable variable that consistently improves accuracy of some decision or action. Such as “customers with a high recency value typically have a higher response rate to marketing programs.”
There are other predictors that might work better with certain actions or decisions. For example, if you have an online subscription-based service, customers who spend less time logged on are less likely to renew annually. Tuning attrition or churn reduction campaigns by targeting customers who have low usage predictor values can boost effectiveness.
To make prediction even more precise you can use more than one predictor at a time. In doing so you are creating a model. Models are the heart of predictive analytics. Some simple models that might predict likelihood of a customer to renew their subscription:
- Linear – adding predictors together. For example: Recency + Household Income.
- Behavioral Rules – joining two or more behaviors with rules defining predictions of another behavior. For example: Usage (high or low) and Responded to Offer in Past 3 Months.
The best predictors will be predictive models that combine multiple aspects of a customer (e.g., demographics) and their behavior. A predictive model characteristically must be deeper and more complex than the above examples – uniting sometimes dozens of predictors. More on determining the best predictive model and harnessing rich sources of data to create powerfully predictive analytics in the next post. Thanks for reading and let me know if you have comments or can share your own experiences.
Imagination
Albert Einstein once wrote that in science “imagination is more important than knowledge”. That’s a powerful thought. I suppose you might expect nothing less from an intellectual giant such as Einstein.
What resonated with me, as someone who often not only wants to understand but who finds fully understanding something (i.e., “knowledge”) to be particularly satisfying, is the caution it offered about seeing knowledge as the only worthy end (to some research you’ve conducted, a project you’ve managed, a business problem or opportunity you’ve worked hard on).
Further, in reading more about the context of Einstein’s writing this line, he is saying bluntly that science like many pursuits in life is really just a journey, full of unknowns and unfolding unendingly. At any given point in time there are many truths or facts that are well-accepted and proven but an infinite number more truths and facts that are quite unknown and sometimes seemingly unknowable. Particularly in science there are many areas of study that deal with phenomena that are not readily or directly observable.
Einstein,and other great scientists, made many of their most astounding breakthroughs using their imagination rather than getting stuck trying to understand the seemingly unknowable. They would imagine some alternative reality to what was known at the time, think through how this alternative world might look and how it might operate if it were discovered to be true, then go about experimenting, searching and testing as if the alternate view were indeed true. This gave them great freedom to work creatively rather than be confined by the “known”. As a non-scientist, for me at least, this was very revealing and refreshing – creativity and science go together! I think I thought before this that they were mutually exclusive.
I began to relate this to my work with business clients where we might be talking about a new product or concept, or a new approach to promotional marketing and other challenges where some facts are well known and many others are for practical purposes unknowable. In such a situation how do you proceed? Einstein would say, if I may be so bold as to speak for him, to first beware of investing all your time into trying to know everything. This is similar to the common advice to avoid “analysis paralysis”. He adds to this common wisdom a more unique point of advice: use your imagination and then be bold enough to just try it out! Experiment. Try. Fail. Try again with another approach.
This is of course no guarantee of success. Your imagination might fail you. But when faced with a big challenge, using your imagination can be a powerful tool to spur action and overcome inaction. At the very least, doing so will give you a taste of how Albert Einstein thought and that alone will be fun!
Experimenting leads to Expanding
Recently I read an interesting research article on “The Contradictions That Drive Toyota’s Success” that I may blog a couple times on since it was full of, well, contradictions to conventional wisdom of what makes businesses successful.
In summary the authors describe three “forces of expansion” (defined as those that lead the company to instigate change and improvement) and three “forces of integration” (defined as those that stabilize the company’s expansion and transformation. The countervailing nature of these forces allow Toyota to be widely and sometimes wildly innovative, creative, and constantly renewing itself, without undo chaos or losing its very clear and constant cultural identity. First I’ll focus on the Expansion forces.
The Expansion forces are noted as Set Impossible Goals, Local Customization, and Experimentation. Each are interesting but the Experimentation force was of particular interest. First, it is an important tool to facilitate the achievement of Impossible Goals. The culture of Toyota is one of pushing the employees to move freely outside their comfort zone and into uncharted territories through regular experiementation — and learning from both successes and failures. There is an interesting illustration from the development lifecycle of the Prius hybrid vehicle. In 1993 (yes, 1993!) they began development and first came out with a car that had 50% improvement in fuel efficiency. This was summarily rejected by Toyota executives in favor of a goal of 100% improvement. This made them look beyond conventional technologies and experiment their way through a string of failures: engines that would not start reliably, ones that could only travel a few hundred yards, battery packs that would not operate in the heat — or the cold.
Two simple concepts that Toyota employs when in experimental mode leapt out at me:
- think deeply but take small steps
- never give up
These sound trite on the surface — too simple to be truly useful. But in thinking about them further, they go together beautifully (and powerfully).
On the first concept, my experience is that many companies get caught up in what I call “mistaking action for progress”. The steps they take may be indeed small but they are not small on purpose. And regularly they admonish their employees to take steps, any steps, so that they can report on “progress” (typically upwards to those above putting the pressure on). Rather than thinking deeply (which takes time but can look like lack of progress) and purposefully breaking a goal down into small, purposeful steps, the action appears to be guided by ready-aim-fire in reverse.
The second concept also sounds too pat but again my experience is that contemporary short term business thinking precludes applying a “never give up” attitude. It is not that companies want their employees to give up at the first sign of duress but without the advantage of using a small-step approach, which carries with it the corresponding advantage of low costs for any failures, costs can mount and patience for success wanes.
Experimentation is one of the most useful and powerful tools an organization can employ. The growing availability of data on markets and customers, the open foundation of the Internet, the near instantaneous pace of all communications, and many other aspects of the current business environment make experimentation both possible — and vital.
Do you agree? Are there other ingredients to successful experimentation?
Good D
In a previous post I mentioned that D for Data was a critical component for revealing insights about your customers. In today’s age, that is as close to “goes without saying” as you might be able to get. So this post is intended to define a simple approach to improving your D … or in sports term, “playing good D”.
First, a “customer advocacy” mechanism is required. For some established companies customer-centric thinking (and ordering of data and systems to deliver insights) is second nature and a distinct “advocacy” role, department, program or such would be unnecessarily bureaucracy. For most companies though, such a formally designated entity is a key building block. A customer advocate can help both develop and drive the systematic approach for discovering and leveraging customer insights, and be a peer at the management table when priorities and on-going investments must be clearly proposed or vigorously defended.
Second, in the infancy stages of a customer insights journey, a world-class, competitively differentiating customer insights data set and deep analytics capabilities are too grand and far off to achieve in the near term. The risk and complexity in achieving this higher stage of customer enlightenment can overwhelm and obscure what are most likely low hanging fruit — or what I like to call “pumpkins” (pumpkins are actually a fruit, and they are not just low hanging but they can be found on the ground ready to pick up). An iterative-experimental approach can give both a good chance to take advantage of the pumpkins along the way and build momentum toward achieving the higher stages of enlightenment within a reasonable time frame. How to do this?
I’ve used a three-part, interconnected process that includes Research, Analytics and Experimentation. These ideally are set up to work together, under one leader. There’s no one best way to set this up and the choice depends upon the organization’s culture and management style.
Research: this is the “compass” function within the overall process. It helps to set the direction for where the keenest insights might be found. It develops an on-going portfolio of customer data and sources through qualitative and quantitative research.
Analytics*: this is the “targeting” function. Using analytics techniques and sound experimental design approaches, it generates testable hypotheses about the next best place to go, within an achievable distance, from the current understanding of customers.
Experimentation: manages the experiment through to implementation and measurement of results, working in partnership with other parts of the organization, external partners, vendors, etc. to get it done.
The key is setting up such a process and team to iterate quickly, gaining the benefit of pumpkin insights and the benefit of honing the skills of implementation of customer insights in the most rapid way possible.
* not to be confused with the general, pervasively useful capability of “analytics” (see Kyle McNamara’s blog for a clear view of analytics).
