Data Science

The promises of analytics are many: business insights, competitive advantage, increased profits, and more. In some cases it may even seem like the opportunities delivered by analytics are nearly endless. So how could one possibly fail with analytics?

Sadly, as Meta S. Brown points out by titling her recent piece in Forbes, “86% of Executives Can’t Make Analytics Pay. Here’s How You Can,” many initiatives never break even. Meta sees the difference between successful projects and unsuccessful project as coming down to planning and execution (read her full piece to understand why).

We’ve found there are five common pitfalls that can delay, kill, or otherwise cause analytics projects to fail. Every situation is unique, but understanding these traps can help you avoid them, thus, increasing the odds of your analytics project becoming a success.

Here are the five potential pitfalls of analytics projects:

1. Not understanding why you want analytics
2. Not planning for measurement and metrics before you begin
3. Not having a goal for your analytics project
4. Not having your analytics team work with your business users
5. Not understanding the critical need for speed to insight

We’ve created a whitepaper to address how to avoid these five potential pitfalls. Download it here.

A great way to explain the analytics maturity and value curve is to use the idea of descriptive, predictive, and prescriptive analytics. Analytics consultants and thought leaders love to frame the field in these terms, if for no other reason than to give a broad, important topic some structure. In a nutshell, let me define these concepts:

• Descriptive Analytics is backwards looking and focuses on telling users what happened. Examples of questions that descriptive analytics can answer are how many, when and where? Traditional Business Intelligence tools like Cognos BI excel in this area.
• Predictive Analytics is about predicting what comes next. By leveraging historical data and predictive algorithms, users can create models to predict the likelihood of future outcomes. This opens up huge new ways of operating for businesses. Now they know the answer to questions like: Who will buy this product? Who will default on a loan? What message would be most effective for this consumer? Tools like SPSS Modeler are the focus here.
• Prescriptive Analytics takes predictive one step further by suggesting the best action to take within a business context to give a business the best chance of achieving their goals. This takes the form of decisions: Which marketing campaign should I sent to a customer? How much inventory should I stock to prepare for expected demand

Plenty has been written about these degrees of maturity and the business value that they can create. Anyone reading this blog is familiar with descriptive/BI and predictive modeling. But how does one go from just predictive into the realm of prescriptive analytics? What’s the secret to success at that level? That’s where IBM Analytical Decision Management comes into play.

IBM Analytical Decision Management (ADM for short) is IBM’s answer to this transition. ADM is available through IBM SPSS Modeler Gold and it combines three core capabilities – the ability to manage business rules, predictive models and optimization. Let’s take a look at each of these and see how the magic happens.

Business Rules + Predictive Models

Business rules and predictive models are very different things – the former is a seemingly subjective, arbitrary condition that must be met. Rules can be driven by corporate fiat, industry tradition or regulatory mandate. The later – predictive models – are the height of objective reasoning, indicating a probability of a future outcome based on historical data.

The reality is that businesses need both – and they should work together but be kept separate. An insurance company can build a predictive model to get the probability of someone making a claim before the policy is issued. But that predictive model won’t (and shouldn’t) consider the fact that the company’s reinsurer has placed new risk thresholds on their portfolio which may preclude some policies from being profitable. By combining these two factors together, the insurance company can make a repeatable, justifiable and profitable decision.

Optimization

This is the general term for the prioritization or selection of actions in order to meet a mathematical outcome. Optimization is a cornerstone of management science and modern industry couldn’t function without it. Optimization can help a company determine the best way to reroute planes after a storm, how to pack merchandise on a truck and much, much more.

IBM Analytical Decision Management allows a user to use an optimization equation which essentially defines what outcome they are trying to achieve. This might be the steps needed to maximize revenue, , or to maximize a marketing campaign’s profitability. And importantly, ADM lets you define these terms using a mixture of predefined rules and predictions. This way your optimization can be done dynamically.

Climbing the Continuum

When combined with predictive analytics, optimization is ultimately what moves a company along the maturity continuum – from predictive into prescriptive. It is this mixture that moves a user from making a prediction towards taking action on a prediction.

ADM automates decisions throughout the organization by selecting an ideal action to be taken. This decision is then consumed at many levels throughout the organization – by retail sales employees, by insurance agents, call center staff, and more. All these groups can benefit by having their normal work supplemented and improved by an infusion of analytics.

Traditional BI and even predictive analytics produce an output which must be studied and understood for insight to be drawn and for the proper action to be taken. The complexity and domain knowledge required in this step ultimately limits the end “consumer” of insight. ADM democratizes analytics, making it easier for regular employees to consumer analytics and benefit from insight drawn from advanced analytics. And that’s the real power of decisions.

I recently presented the Predictive Customer Intelligence solution from IBM to a group of executives at the Memorial Golf Tournament in Dublin, Ohio. The audience spanned the retail, insurance and banking industries and all had one common challenge: how can we “do better” by knowing our customers and making smarter, more targeted offers? Some in the group have active analytics programs, while others recognize the need to get started.

My story – and that’s where the best predictive analytics discussions begin – was about how customer analytics can transform a customer experience for the better and create significant business value in the process. The executives in the room asked some great questions – the answers to which can certainly help others thinking about customer analytics. I’ve summarized the top five questions from the event below.

How do you build the case for next best offer?

Building the business case for a new system or investment can be challenging, but it all comes back to measurable results. Every company that is working with customers on a day in, day out basis has interactions that are being (or can be) measured. Anything that can be measured can be improved – so to build the case, look at where improved customer interactions can make the biggest impact. Where and when are you touching the most customers? Are these repeatable transactions? What does your business want to sell more of? Answer these questions and you’ll have the beginnings of a case to invest in next best offer.

What portions of the analytics implementation did you control directly? What was the span of control?

This question was from a VP of Marketing who was challenged with getting the idea of improved customer analytics to spread within his organization. He saw the opportunity, but needed buy in and support from others in the company. Champions of predictive analytics often sit within line of business, like marketing, within the overall organization so this is a common scenario. The answer is in the question above – have a defined business case for analytics and take it to others in the company to solicit support. Saying “we should do analytics” won’t get you the support you want, so its pays to be specific. Approach others with a specific change and result – for example, “We want to improve customer retention by empowering sales reps to make retention offers to customers statistically at risk and who are most likely to stay with an incentive. We can identify these customers through predictive modeling and manage the overall offer budget through an optimization process.”

Who comes up with the use cases?

Initially the internal analytics champion needs to steer the conversation. With help from partners like Revelwood and IBM, that champion must paint the vision of how business can be transformed using this technology – and the vision needs to be specific and detailed. As your analytics program progresses you’ll find that everyone in the organization from front line employees on up will begin thinking about how predictive, optimization and statistics can help guide customer interactions to a better outcome for all. In the most effective program I’ve seen, the marketing analytics group set up a bottoms-up feedback system to collect new ideas for vetting and development.

Do we need a data scientist?

This is actually the wrong question. A better question is when do I need a data scientist. The tools and technology around predictive analytics and customer intelligence have progressed to the point that a company can collect early wins and drive results by having an internal analytics champion and users who understand two things: their own business and their own data. As your program moves forward and tackles more complex use cases and looks to drive value from additional predictive accuracy, you should look into a statistical professional.

How much data do we need?

A common question. It’s rarely about how much data is needed to get started on an analytics project – rather its about having the right data. For example, if you need to predict purchase propensity of a given product, you don’t need years worth of data on all your customers. Instead, focus on a few months of purchase transactions combined with CRM and pricing data. Going wide before deep will steer you in the right direction most of the time – at least until you need a data scientist.