With online retail increasingly becoming an integral part of a brick-and-mortar retailer’s strategy, there has been a lot of focus on developing an integrated supply chain, integrated marketing plans, an integrated assortment strategy, etc. Multi-channel strategy in one sweep covers all of these. Said in a nutshell, the process of creating and leveraging synergies between online and the brick-and-mortar channel is multi-channel strategy.

There is no retailer who isn’t focusing on Test and Learn or Multi-channel strategy. The million dollar question is “Do retailers see these strategies as interconnected?” Here’s some food for thought – Can online (which is a cheaper and easier channel for testing) be leveraged to do testing before we do a larger roll-out in the stores?

The naysayers will argue that online customers are different from in-store customers and online assortment is different from that in stores and hence any test done online is not valid for stores. The believers, like Mu Sigma, will tell you that the time is ripe to use online tests to guide retail strategy. Which side are you on?

A lot of the recent increases in data volumes can be attributed to what is being written in the social media, what is being captured by store video cameras and user comments about products on product pages and blogs. Making sense of this kind of data doesn’t just need more memory and more computing power. It needs advances in our understanding of unstructured data. Most of this data tend to be subjective and context dependent. A comment like “Unbelievable!!” can have very different meanings based on the context in which it was said. A video of a bunch of people walking around in a store by itself may mean nothing but when combined with the fact that there was a promotion being run in the bakery which is on the north east side of the store may lead to something insightful.

This kind of contextualization of data requires developing sophisticated algorithms and logic and this side of Big Data seems to be not getting enough attention. It requires a combination of business, IT, math and behavioral sciences to define and systematically capture context. If your Big Data strategy only covers the size of your database and the number of nodes in your network cluster and their computing power, think again!!

1. Companies are finally beginning to recognize that people (i.e. their customers) are not the utility-maximizing, rational agents but are usually driven by a complex set of intrinsic and extrinsic motivators. Any business decision that does not factor this fundamental fact about the customer is at best, sub-optimal.



2. While behavioral economics has been around for some time, we are reaching at an inflection point in our ability to apply the concepts – thanks to a combination of the explosion of data that can be used to better understand the customer (e.g. online behavior; data from video feeds) and technology innovations to work with unstructured data.

Here are some of the key concepts that are going to be very important for Decision Sciences practitioners in the years to come.

1. Loss Aversion: This stems from the ‘Prospect Theory’ (developed by Daniel Kahnemann and Amos Tversky) – the paper (and the subsequent body of work) that can lay claim to have kick-started behavioral economics as a discipline. At the core of this theory is the concept of loss aversion – that is, people prefer to avoid losses as compared to acquiring gains. In other words, people tend to be risk averse when evaluating a possible gain and conversely, prefer risks that might mitigate a loss (called risk seeking behavior). In sporting terms, people would rather play defense. This has implications in different areas: one example being subscription-based services. People who sign up for a service at a heavily discounted rate show a higher propensity to opt-out if there is a steep increase in the rates at the end of the period, which is an example of risk seeking behavior. In a competitive scenario, the impact of a price decrease by one provider may not have the same impact on the defection rates as the increase in price by the competitor.



2. Choice Architecture: It has long been known that choices are never made in a vacuum. Behavioral economics has finally put the ‘rational agent’ who makes choices based on perfect information back in the realm of fiction. It is now clear that the conditions under which choices are made have a significant bearing on the way they are made. That opens up significant possibilities in terms of creating the conditions that could present the choices so as to influence the outcomes. This in turn has powerful implications on how businesses can offer options to their customers to influence the outcomes. Armed with a combination of experimentation techniques and the ability to rapidly process data from experiments, companies are increasingly adopting a culture of simulating multiple choice alternatives before narrowing down on a product or service offering.



3. Anchoring effect: This is probably the best known of behavioral concepts. Simply put, this recognizes the fact that individuals rely on a specific piece of information to drive their thought process. The classical assumption of an individual looking at all the available data with an unbiased lens before making a decision was never really true – and the anchoring effect is what challenges this view. A focusing illusion is one such cognitive bias where people tend to skew their emphasis on one specific event or data point, which ends up influencing their overall decision. This effect has been used by marketers for quite some time now to influence consumer decisions.



4. Framing: When people are expected to choose between multiple options, their decisions can be influenced simply by how the options are expressed. Simple as it may sound, this has profound implications on how decisions can be influenced. For instance, the response to a disease prevention strategy (e.g. polio shots) can be strongly influenced by how the options are framed in a positive or negative manner.



5. Ambiguity Aversion: People show a remarkable tendency to prefer a known risk to an unknown risk, even if the impact of the known risk is high. This tendency has been shown to kick in when the choice allows for a comparison between multiple options – where the least vague option has a higher probability of selection. Once again, this is a clear departure from the classical utility theory. Choice models would do well to recognize this fact.



6. Decoy effect: This is a phenomenon whereby consumers tend to change their preference between two options when a third option which complicates the decision making is presented. This is a clever trick that is routinely applied by companies to nudge their customers towards specific products. This is especially useful when product options are evaluated on multiple dimensions, often making it difficult for a customer to identify the best-fit product. In such cases, the introduction of a new ‘decoy’ option can be effectively used to influence the decision towards a particular choice.



7. Hyperbolic discounting: Humans have always shown a bias for instant gratification – i.e. given two similar rewards, people almost always show a preference for one that would materialize sooner than later. For instance, given a choice between $50 now or $100 a year from now, most people would prefer the $50 option. However, when asked for a choice between $50 in 10 years and $100 in 11 years, most people would switch to the $100 option, even though the choice is essentially the same. Such biased decisions are regularly made in the world of insurance – and insurance companies are taking note of this phenomenon.



8. The Allais Paradox: This is a subtle, yet powerful phenomenon that was first reported by Marice Allais (Nobel Prize 1988). It implies that in a situation of uncertainty vs uncertainty (i.e. different degrees of risk), people tend to maximize expected value, whereas in a situation of certainty vs uncertainty the same set of people tend to prefer certainty with an attempt to maximize expected utility (i.e. satisfaction) rather than value. In other words, depending on how the choices are made available, there is inconsistency in the decision making process.

Obviously, this list is neither complete nor comprehensive. Looking forward to inputs and in addition, any interesting example of applications from the real world would also be very valuable.

To read more about this, click here.

Really?

The team has fallen victim to one of the most common mistakes people make in the name of statistical analysis: confusing Correlation with Causation. Today’s “Information Management” newsletter includes an article we wrote on the topic.

To learn the remaining four mistakes, check out the article, and let us know what you think!

There’s no doubt that institutionalizing analytics leads to better decision making – but there are three different options for doing it: creating a Centralized analytics function, creating a Decentralized analytics function, or a Federated approach.

Which is best for you? The answer, I’m afraid, is “It depends.” Each has its pros and cons. There is no “one right answer”.

Some factors to consider include:

• Company size: Smaller companies may have an easier time with a centralized model, primarily because good analytics talent is so difficult to find. Centralizing the group will enable you to use resources more efficiently.
• Degree of diversion within the organization: Let’s face it – any organization with more than one person has politics. If politics are rampant at your company, you may be better off with a decentralized or a federated approach.
• General pace of the business: If decisions need to be made quickly, avoid a federated approach, which tends to be the most time consuming.
• Availability of analytics talent: It’s well known that there is a serious shortage of analytics talent, at least in the U.S. If you have trouble finding and keeping analytics talent, consider a centralized approach. Mu Sigma also provides an extended arm to ensure bandwidth, industry expertise and scalability on the fly for big projects.

Need help determining the best approach for your organization? Send me an email.

Well, last week, EMC Corporation announced the results of its own study – which found the same results:

“EMC Corporation (NYSE: EMC) today unveiled the findings of the largest-ever global survey of the data science community. Spanning the United States, the United Kingdom, France, Germany, India and China, the EMC Data Science Study reveals and quantifies a rampant scarcity across the globe for the prerequisite skills necessary for a company to capitalize on the opportunities found at the intersection of Big Data and data analytics. Only one-third of companies are able to effectively use new data to assist their business decision-making, gain competitive advantage, drive productivity growth, yield innovation and reveal customer insights.”

This doesn’t surprise us, because clients tell us on a daily basis how hard it is for them to find, recruit and retain analytics talent. The very few people they do find tend to take jobs in the financial sector (read: hedge funds), where they feel the rewards are richer. One of our clients – a senior executive for a national retail chain – told me that at one point his company had ten openings in its analytics group, and that in one year of serious effort he was able to fill only one position.

This trend has been a key driver for Mu Sigma’s swift growth – many companies are finding that it’s faster to simply outsource advanced analytics to Mu Sigma than to spend years trying to build up an in-house team.

Has your company found it challenging to find good analytics talent? What strategies have you used to attract these people to your firm?

In our experience, the vast majority of companies today are still at the Analytics Laggard stage. They still make gut-based decisions, primarily because they don’t have the centralized data infrastructure required to get started with analytics. In fact, a key indicator of an organization’s progression to the second level is when it begins to manage data more effectively.

The goal of course is to get to Analytics Master, where your organization is using analytics as a strategic differentiator.

This is a multi-year journey – Mu Sigma has long-time clients who are still working toward that designation. But there are significant benefits associated with each promotion through the model. The payoff begins almost immediately and accelerates as you go.

What level is your company at, and what are your plans for progressing through to Analytics Master?

Each researcher typically has 1-2 Simmer Projects going on at any given time, and the best ones eventually get put on Microsoft’s roadmap. I’m told that some of Microsoft’s most well-known products have begun in this manner.

What a great example of a company taking a long-term view on innovation – instead of putting every available body to work in short-term goals, Microsoft has given its employees creative license to experiment in the kitchen.

What steps are your organization taking to foster innovation?

This focus on better creation and faster consumption of analytics within organization has led companies to form the Analytics & Insight Council (AIC). AIC will become an integral function of all organizations in the next decade. The mandate of this council will be to go beyond setting strategic direction for the creation of analytics across business groups. It will include, amongst other things, enabling the consumption of analytics and cross pollination of ideas across the organization.

What do you think should be the role/structure of the AIC? Share your opinions with us.