Last week I spoke for IIA at Predictive Analytics World. Whereas I often speak to executives who aren’t yet persuaded of the virtues of analytics, at this gathering—which also included attendees of the Marketing Optimization Summit and Text Analytics World—that wasn’t the problem. I was preaching to the converted, who already undertake a wide variety of analytical initiatives in their organization. The problem becomes not how to get going with analytics, but how to “industrialize” them.

As analytics grow more popular and central to business strategies, there is also a need to produce analytical miracles repeatedly, reliably, and quickly. The old days in which an analyst could take his or her time to produce custom analytical solutions are almost gone; instead, groups of analysts need to have “analytical service lines” in place. Some organizations might refer to them as “analytical solutions.”

These are necessary not only because the customers of analytical groups within organizations need quick and reliable delivery of analytics, but also they need to be familiar with the possibilities for analytical work. A “menu” of services that can be provided with speed and reliable outcomes can be very useful to the consumers of analytics. Incidentally, external analytical consultants should also provide a similar menu of analytical service lines.

One of the best examples I have found of such service lines or solutions is at HP Global Analytics—the analytical shared services group of the giant computer company. This group, largely based in India, has created a set of repeatable and scalable analytical capabilities serving many different functions across HP. For marketing, for example, they offer market intelligence, customer targeting, marketing spend allocation, and pricing analytics. In sales they offer sales force allocations, pursuit and conversion optimization, compensation optimization, and sales performance reporting. The group also has offerings in customer service, supply chain management, and HR—19 solutions in total.

So what comprises an analytical service line? It should have the following attributes:

• You’ve done it before, ideally several times—and it’s achieved a good outcome;
• You have at least some sense of a process that is followed to produce the desired result;
• You either have data readily available for the analysis, or know where to find it;
• You know what the likely decision outcomes are for the analysis;
• You know who the likely customers are for this service within your organization or client;
• You have created some degree of marketing materials to describe this service and its benefits.

As with all analytical capabilities, you can’t assume that your customer will understand analytical jargon on the service line card. Just as a restaurant markets the items on its menu by creating appealing descriptions and training service personnel to describe them, the analytical offerings on the menu have to be marketed and sold too.

Creating a set of analytical service lines, and executing on them effectively, will go a long way toward scaling up analytics in your organization and delivering them efficiently. Some analysts may yearn for the one-off, ad hoc analytical approach, but most will probably appreciate the increased influence they are having on the business.

Next best offers (NBOs) are targeted offers to customers of products or services that they are likely to be particularly interested in and to buy. I was interested in the NBO topic because it is a tangible project using analytics to benefit customers. The world of customer analytics is a very complex and broad topic, and most organizations need to pick something tangible to focus on and accomplish. NBOs are a worthy goal for many organizations because they require a decent knowledge of customers and their wants, product or service attributes, and the purchase context. If you can gather some information about all three of these and combine them in more-or-less real time, you can make a big difference in your marketing.

What’s difficult about NBOs is that there are so many variables that companies can employ to improve them. What have you bought in the past? What are your demographics, and how do they combine into a life stage and lifestyle segment? What product attributes do you find appealing, and which of our products have them? Did you drop in on us via the web, a call center, email, or a physical store? Do you happen to be walking by our store? And are your friends discussing or “liking” our products in social media?

Most organizations struggle with demographics and purchase history information about their customers. Forget about detailed product attributes or SoMoLo (social, mobile, localization) variables. How can you pull together a well-targeted offer when there are so many types of information to master—most of which you don’t have today?

As we said on the webcast, there are two key strategies to make this work. One is to think carefully about your strategy and product/service offerings. What information is likely to influence the purchase of an offer? What channels are you likely to
exploit? Do social media matter in your business? Do you really need to know whether your customer has children at home? Some of these questions can only be answered through empirical data, but logic-based shortcuts can help.

The other key strategy is to start somewhere, and try to make progress. You may not be ready to gather and exploit SoMoLo data, but maybe you can start with purchase history, gender, and age. Most offers made by companies today don’t even take those rudimentary factors into account (and it shows). So start with the limited information that you have available and slowly add to it over time. Make sure that your data warehouse or mart can grow as you add information, channels, and data types to it.

No organization on this planet today has “mastered” NBOs. We’re all finding our way. If you read something impressive about a particular company’s use of social media in offers, for example, they are probably not using something basic like marital status information. As they say in the UK, “Keep calm and carry on” making your offers better.

Here’s an example. A few weeks ago, at a conference for financial executives interested in analytics, I met a quite senior analyst for a company in the home furnishings industry (I’m not providing the company name for reasons you will understand if you keep reading).  My speech at the conference was partly about the need to use analytics to understand what nonfinancial factors might be driving your business, so that companies can begin to address how to change or at least react to them.

After my speech, this home furnishings analyst came up to me and said that he had already done this at his company. He said he hypothesized that his company’s sales were highly dependent on housing starts, so he ran some regression equations with various degrees of lag on housing starts—about one year was optimal—to see how much of the variation in company sales it explained. I won’t say the exact R2 number, but it was high—well over 50%. So in essence, most of the company’s sales this year depend on how many new houses were built last year.

There’s nothing better than knowing what factors drive your business performance, right? Well, not necessarily. He told me that his executives seemed largely uninterested in this finding, and didn’t use it for financial and production planning purposes. They didn’t dispute it—they simply ignored it.

Being the naïve fellow I am, I asked why. He cited several reasons:

• They wanted to portray to the investor community that their growth would be higher than what the dismal numbers on last year’s housing starts would suggest;
• They believed that their management talents were sufficient to overcome a weak housing market;
• If their business was simply to fulfill the demand from the housing market, what value were they adding as executives?

In retrospect, of course, it seems obvious why they didn’t want to use the analytics. They were in full contradiction to their self-perceptions of their skills as managers. They also made it appear that no matter what the executives did in a poor housing market, they weren’t going to be successful in growing revenues.  The executives proceeded without the analysis, and of course overshot their actual revenues in terms of their Wall Street forecasts and production plans. The company is profitable, but not very much so.

The lesson is that before you go to a lot of trouble to create an analytical model, make sure that someone actually wants the result. Individual initiative is great, but there is only so much persuading one can do when the very idea of an analysis contradicts an executive’s view of the world. Maintaining a close relationship with a decision-maker will help to prevent this type of problem.

Last year, my co-author Jeanne Harris and her colleagues performed a survey of quantitative analysts within large organizations. This was the first of its kind, to my knowledge. The survey sheds some interesting light on which organizational models for analysts are the best. Of course, organizational structure can be a complex topic. There are strengths and weaknesses of almost every model. However, the survey data suggest that the more centralized approaches have some strong advantages.

The data suggest that if you care about having your analysts being engaged with their jobs, and likely to remain in your employ, the two most successful organizational models in that regard are the center of excellence (29% engaged, 41% likely to stay) and centralized (35% engaged, 33% likely to stay) models (see previous post for descriptions of these). Although the percentages for the more decentralized models aren’t horrible, they are clearly worse on both measures. The decentralized model had only 18% of analysts engaged, and 27% likely to stay.

The data don’t tell us why the central and CoE models are more successful on engagement and intention to stick around. But I have some hypotheses. People in central analytical groups are, according to my research, more likely to work for organizations that really care about analytics. They’re more likely to work closely with other analytical colleagues. They can have more varied work assignments because they’re part of a large pool. All of those explanations seem plausible, right?

One interesting question is why the “internal consulting” model ranks so low, even though it is relatively centralized. Its analysts were 23% engaged, and 24% likely to stay. The burden of billability, perhaps?

One somewhat scary finding is that even the best models are somewhat low in engagement and intent to stay. These analysts are incredibly valuable to any company pursuing a data and analysis-based strategy. Companies need to find ways to make their analyst jobs happier and stickier. This will become even more important as the general labor market picks up, which I believe is happening. There is already a lot of demand for good quantitative analysts.

How about you? What’s your organization’s model, and how does it relate to your personal levels of engagement and employer stickiness?

One issue that organizations are often interested in is how to organize analytics. In this post I’ll describe five different organizational models; surely one of them will fit your organization.

First there is the totally centralized model, in which a group of analysts in a central function serve the entire organization. The strengths of this model—which, as I pointed out in the last post, is growing—include the ability to work on cross-functional projects, the ability to share ideas, and the ability to assign analysts efficiently out of a central pool. The downside is a potential unresponsiveness to the needs of the business, so other mechanisms (see the last post) will have to be adopted to address that issue. Companies primarily employing this centralized model include Procter & Gamble, UPS, Expedia, and United Airlines. The analytical function may report to IT, strategy, or a corporate services function.

A variation on the centralized model is the consulting model, in which analysts are centralized, but are expected to recover their costs by charging for their time. The chargeback process does ensure that someone in the business finds the analysts’ services to be of value. However, it may also prevent analysts from working on the organization’s most strategic analytical problems. Just because someone can pay doesn’t mean their problem is important. Schneider National and Disney have this model in place.

A somewhat less centralized approach—but still with some enterprise-level coordination—is the center of excellence model. In this structure, analysts are based in business units, but their activities are coordinated by a small central group. The CoEs are  typically responsible for issues such as training, adoption of analytical tools, and facilitating communication among analysts. Bank of America, Citigroup, and Kimberly-Clark have this model, though the latter is primarily focused on business intelligence more broadly. This is an analytical version of the Gartner- promoted “business intelligence competency center.”

The functional model puts analysts primarily within the function that dominates analytical activity within an organization. If almost all the analytical work supports marketing, for example, then why not make most or all analysts part of that function? The bad news with this model is that it may limit the ability to expand analytical work to other functions that could benefit from it. Fidelity Investments (in customer relationships) and Harrah’s (in marketing) are two organizations that
have adopted this model as the primary home for analysts.

Finally, there is the fully dispersed model, in which analysts are spread throughout an organization with no vehicle for collaboration or coordination. While all the other models have some rational justification, I’m not sure this one does. It usually means that the senior management of the organization doesn’t recognize the importance of analytics. Of course, it does ensure that the different functions and units get what they need, but it’s suboptimal for enterprise applications of analytics. Since I’ve
criticized it, I won’t name any adopters of this model—but there are plenty of them.

Next time I’ll present some empirical evidence that some of these models are better than others.

In the last several months I’ve been subjecting much of the world to my spoken thoughts on analytics. I’ve been to Europe (twice), Brazil (twice), Mexico, Argentina, Hong Kong, Chile, Canada, and various places in the US. One thing that has struck me as I travel around and talk to different companies is the incredible geographic variation in analytical approaches within the same company. You find some impressive analytical work going on at corporate headquarters, and then you discover that it’s not present outside the home country—or sometimes it’s vice-versa.

I’ll give you a couple of examples. Take Tesco, for example. Based in the UK, it’s the world’s third largest retailer and has operations now in 13 other countries. With the help of consultants dunnhumby (in whom Tesco eventually bought a majority ownership share) the company pioneered the use of its loyalty card (ClubCard) data to target promotions to members. It’s been a fantastically successful program, and is responsible in large part for Tesco doubling its UK market share since Clubcard was introduced in 1995.

However, there seems to be substantial variation across countries in whether ClubCard—or an equivalent program—is offered, and the extent to which data from it is used to target promotions. It’s definitely not offered in the company’s US Fresh & Easy chain, where loyalty programs are even somewhat disparaged on the website. However, Kroger is an aggressive user of the approach (and a part-owner of dunnhumby USA). I heard recently that Korea—where Tesco has a joint venture with Samsung in the Homeplus brand—is a big user of a loyalty card (Familycard) and the resulting data. I got blank looks when I mentioned Clubcard to a Tesco employee in the Czech Republic, although I suppose it could be a language problem.

Another example of dramatic variation is Banco Santander, the Spain-based bank that is now the world’s eighth largest in terms of assets. Last year when I visited Spain, I was told by some consultants that Santander was pretty analytical there, albeit not the market leader in that regard. Brazil people told me they were rapidly catching up to Spain. Chile’s not doing much. I heard in Mexico that Santander is quite aggressive on credit card analytics, basically emulating the very successful (in the US, at least) approaches of Capital One. I read recently that Santander is doing great work in Germany on credit scoring and automated loan decision models. I have yet to hear much going on with analytics at all at Sovereign Bank, the US bank that Santander owns. As far as I can tell, the only global approach to analytics at Santander involves risk management, a consistent approach to which is somewhat mandated by Basel II regulations.

Is this geographical variation good or bad? You could argue that it’s somewhat necessary, since regulations and available information vary across the world. In Brazil, for example, there is no such thing as a credit score, which limits the ability to make loans on the basis of it.

But I’m guessing there is an opportunity to do more to standardize and collaborate around analytical approaches. In Europe (at Berlin’s great Neues Museum) I ran into Giles Pavey, a dunnhumby executive who is “Head of Analysis.” He told me he’s spending lots of time spreading the analytical gospel around the world, particularly in the Far East. Maybe what we need more of is roles like his: “Analytical Ambassador.”

In the August IIA Office Hours we heard about another example of global coordination. David Fogarty, our speaker, is the head of customer analytics (they call it “Customer Value Management”) for CIGNA in Asia. David is based in Hong Kong, where he has a very small staff. But he operates a Center of Excellence (CoE) that has identified a set of analytical competencies around Asia in various “data labs.” As we said in the Office Hours report: CIGNA’s data lab in Taiwan concentrates on testing and learning, in China the data lab is focused on product optimization, and in Spain, which is the company’s first market for a new private medical insurance (PMI) product, the data lab focuses on analyzing PMI. The learnings from each data lab are shared with the CoE in Hong Kong, which distills the findings and packages the key insights with the labs around the globe.

This is a neat model that I believe more organizations should emulate. I am sure that we will see many examples in the future of global management and coordination of analytics, but the time to start is now. There are great analytical capabilities out there to be spread and leveraged!

There are many businesses and industries around the world that are virgin territory for analytics. Yes, many industries have considerable analytical activity underway, and more and more companies are embracing the competitive potential of analytics.  But there are far more that haven’t even begun to compete analytically.

Residential real estate is one of those relatively non-analytical domains. I spoke recently with Russ Baris, a real estate analytics maven who headed for many years a very impressive group of IT/quantitative analysts at Pfizer called Business
Technology and Information Science. He has a strong quantitative background, and he did great work on analytics for pharma for more than two decades.

Now, however, he’s started a company (called Elumindata) to develop analytical solutions for various industries where
they are underleveraged. One is the residential real estate industry. Several of Russ’ family members are in that business, so he knows it well. He also knows there is plenty of opportunity to make it more analytical.

Think about it—you want to sell your house (who doesn’t these days?). You call a Realtor (lawyers—please forgive absence of trademark—at least I capitalized it).  Do you know the average days on market for the houses being sold by that realtor? Do you know how frequently sellers re-signed with that realtor after the original contract period ran out? Do you know what percentage of appraised value the realtor tends to get for the houses he or she sells?

No, you don’t. Nobody does, except Russ and a few clients. Those analyses simply aren’t available to consumers. If it’s any consolation, realty office owners and managers don’t have them either. It’s an industry that has run in the past on charm and charisma and connections, but certainly not on correlations.

Russ is trying to fix this problem. He can take Multiple Listing Service (MLS) data for a particular town or region and quickly produce some very interesting reports about which companies and agents are doing well, and which are languishing. If
you’re a realtor, you should seek out Russ and implement his solutions. Maybe someday the information will even be available to those who want to sell their houses.

I have also spoken recently with several executives from Zillow, the Seattle-based company that publishes estimates (OK, “Zestimates”) of the value of houses across the U.S. Zillow is a very analytical company, both in terms of
the basis for its Zestimates (though mine is too low!) and how it manages itself. I suspect that it will be a major player in the further penetration of analytics into the residential real estate biz.

You’re probably not a realtor, and you may not even own a house. But in your industry—whatever it is—there is likely to be some area of the business that could be much more analytical than it is today. If you’re smart and enterprising, you’ll find some way to transform that operation with the help of data and analytics.

Leading payers, providers & life sciences organizations from the health care industry gathered recently for the kickoff of IIA’s Health Care and Life Sciences Analytics Research Council. The goal of the project is to uncover the leading analytical techniques being used within health care.

There were some fairly esoteric issues discussed, such as the role of patient registries, the possibility of post-market drug surveillance in hospitals, and doing analytics on genomic and proteomic data. On the call for providers (hospitals, medical practices, and home health organizations), however, some very basic and important questions were raised.

One was simply, “What should we use analytics to predict?” I refer to this issue as the targets for analytical activity, and it’s important for every organization—but it’s particularly important for health care providers. Many of them are just now implementing electronic medical record (EMR) systems, and they’re trying to lay the foundation for analytics. For what should they be planning? There are many options for “what to predict,” including prediction of patients:

… who are likely to acquire particular diseases;
… who are likely to require emergency care soon;
… who are likely to be rehospitalized;
… who are likely not to pay their bills;
….who are likely to purchase lucrative optional services.

And those aren’t the only choices; prediction can also be applied to clinicians, treatment protocols, bed occupancy levels, and utility bills. Which predictions should a provider endeavor to implement?

Of course, I know the answer…but there’s no room in a blog post to present it. Actually, of course, there is no one answer. But one of the key questions we’ll try to answer for provider organizations is what sorts of predictions make sense under
particular strategic, organizational, and technical circumstances. I predict that it will be useful information for those who provide health care.

This is a long-term trend, but I was motivated to blog about it after a conversation with a friend who is developing an automated analysis tool for financial data. He was previously at a large brokerage and investment firm. While employed there a couple of years ago, he had occasion to negotiate licenses with some large providers of corporate financial data. The annual license fee was $250,000 for up-to-date performance data on a variety of US firms.

Now, as part of getting his new firm going, he’s gone back to some of the same companies to negotiate licenses for more-or-less the same data. The cost is now less than half of what it was a couple of years ago, even for a customer that’s much smaller with considerably less leverage than his old firm.

My friend is hoping to make the data more useful with automated analysis. He points out that there is way too much data available for the number of equity analysts in the world. But it’s fair to say that other industries have a similar problem—too much data, not enough people to analyze it without the help of some automated or semi-automated tools.

Perhaps you can begin thinking of ways to make data less commoditized in your organization. But if you need help, we’re working on the issue at IIA. Jeanne Harris of Accenture, Russ Walker of Northwestern’s Kellogg School and I are working on a research project we call  “Leveraging Proprietary Data for Competitive Advantage.” I’ve just written an overview of the research project, and a research brief on how financial organizations leverage (or not) payments data. Both are available to IIA members in the proprietary research library in the Members Area.

If you have any proprietary data that you are starting to leverage and de-commoditize, please let us know—we’d love to talk with you. I promise not to blog (or otherwise write) about you without your permission!