One Marketing Analytics Myth? The Perfect KPI... published July 6, 2015
I lead Force Five Partners, a marketing analytics consulting firm (bio). I've been writing here about marketing, technology, e-business, and analytics since 2003 (blog name explained).
Email or follow me:
One Marketing Analytics Myth? The Perfect KPI... published July 6, 2015
Apropos of enlisting a crack team (Jane Repetti Chang, David Chang, and Caroline Repetti) to ride with me in this year's NLG Bike-a-thon (please support us, and thank you to those who have!), I thought it fair to warn them for what they're in for. This recap of the 2007 ride seemed just the thing:
Thank you all for being so generous on such short notice!
Fresh off a flight from London that arrived in Boston at midnight on Friday, I wheeled myself onto the starting line Saturday morning a few minutes after eight . Herewith, a few journal entries from the ride:
Mile 2: The peloton drops me like a stone. Dopeurs! Never mind; this breakaway is but le petit setback. Where are my domestiques to bring me back to the pack?
Mile 3: Reality intrudes. No domestiques. Facing 47 miles' worth of solo quality time, I plot my comeback...
Mile 10: 1st major climb, L'Alpe de Bolton (MA), a steep, nasty little "beyond classification" grade. I curse at the crowds pressing in. 'Allez! Allez!' they call, like wolves. A farmer in a Superman cape runs alongside.
Mile 10.25: Mirages disappear in the 95-degree heat. (First time I've seen the Superman dude, though. Moral of this story: lay off the British Airways dessert wines the night before a big ride.)
Mile 10.5: Descending L'Alpe de Bolton, feeling airborne at 35 MPH
Mile 10.50125: Realizing after hitting bump that I am, in fact, airborne. AAAAARRH!!!
Mile 14: I smell sweet victory in the morning air!
Mile 15: Realize the smell is actually the Bolton dump
Mile 27: Col d'Harvard (MA). Mis-shift on steep climb, drop chain off granny ring. Barely click out of pedal to avoid keeling over, disappointing two buzzards circling overhead.
Mile 33: Whip out Blackberry, Googling 'Michael Rasmussen soigneur' to see if I can score some surplus EPO
Mile 40: I see dead people
Mile 50: I am, ahem... outsprinted at the finish. Ride organizers generously grant me 'same time' when they realize no one noticed exactly when I got back."
"Analytics Is Too Important To Be Left To Analysts", published February 20, 2015
"How to select a programmatic advertising solution in 2015", published February 2, 2015
Philip Greenspun kindly invited me to present a talk on database models for evaluating digital advertising in his class, "6.S186: Special Laboratory Subject in Electrical Engineering and Computer Science -- Relational Database Management System and Internet Application Programming" on January 28, 2015
McGraw Hill Education VP R&D and Analytics Alfred Essa and I will present a hands on workshop for senior business leaders and the senior analytics executives who advise them. The session will blend case studies and lessons from practical experiences with a tour through getting, prepping, exploring, and modeling data for insights, using the Python programming language.
I'm moderating a panel at Analytics Week Boston on November 6 at 4pm at the Boston Courtyard Hotel. participants include: Jim Barrett, VP Marketing, Rue La La; Andy Dubickas, Senior Solutions Consultant, VisualIQ; Sara Radkiewicz, Director / Product Management, Localytics; Lawrence Schwartz, VP Marketing, Attunity.
Session description: "One of the biggest opportunities in marketing is the ability to recognize and target an individual user across all the devices he or she might use to proceed through an experience. But the foundational mechanisms for this -- cookies and the information we can get from mobile devices -- are increasingly limited in their utility and availability. Botnets are compromising the quality of the cookies advertisers target. Recent compromises of photo archives and credit card records, and other data by both official and criminal entities have driven mobile device suppliers to tighten up privacy and security controls and features. Where does this leave us today, and how (well) can we continue to realize seamless experiences in this newly-fraught environment?"
http://bit.ly/watbs-econtmag, published October 3
http://bit.ly/htmyap-cmo, published September 26
A year ago, while doing research for my new book Marketing and Sales Analytics I wrote a post titled "Organizing For #Analytics -- Seven Considerations". I argued that you should "think organization last, not first", and presented, well, seven considerations that would shape the structure you end up with.
But I made an important mistake in writing that post. Enquiring minds still want the boxes. So, based on the 15 case studies I presented in the book, as well as other experiences over the past decade with even more firms, this post offers a "reference structure" and suggests how it might be tweaked under different circumstances.
In constructing an analytic organization, you've got two objectives to serve that should guide you. One is answering questions. The other is answering them better the next time.
To answer questions, once again you've got two things to balance. One is "business intimacy" -- how close you need to be to the business you're serving to understand what's going on, so your analysis is grounded in that reality. By what's going on, I mean things like terminology, the particular patterns and quirks in the data the business generates, and the options for acting on insights that the operational infrastructure, people, and process of the business make possible.
Balanced against business intimacy is "analytic capability". This means having a critical mass of skills for supporting decisions appropriately. If a decision requires a sophisticated model using lots of data, for example, having a bunch of business-intimate MBAs who last ran a regression a few years ago in business school using Excel might not cut it. So you might need one or more specialists who can do that sort of thing. Or, if it requires specialized knowledge about, say, how digital analytics work, you might need folks who know about things like tag management systems, or inferred matching across devices for multi-screen tracking.
In implementing these two requirements, you then face -- wait for it -- another two choices. One is how much to distribute into the field, or into business units, or aligned to functions, versus how much to centralize. The second will be how much to separate analysis from reporting. With respect to the former, a lot depends on how different or similar the needs in each business / field location are (the more different, the closer you want support to them, and the less leverage there is from centralization anyway). As for the latter, My thinking and research has suggested that separate "report generation" groups are a bad idea -- they encourage conveying data rather than insight, and thus the development of a second-class citizenry in analytics groups; they disconnect report generation from report usage, resulting in the proverbial pallets of unused reports that consultants on cost-hunts feast on; and, they discourage automating reporting (because most folks don't like to work themselves out of a job).
To answer questions better the next time, you need some folks thinking about everything from whether there are better techniques and associated tools, to whether there's more useful data, and how to better organize and provision the data you've already got. There are going to be -- yes, of course -- two options for how to proceed here. One is a "rapid prototyping" approach, and the other is a more traditional "waterfall" method.
Wrapped around all this is "governance", only with a capital "G". In most cases, governance is applied to data, but not to the prioritizations of the questions and decisions that analysis will answer and support. Capital G governance also sets analytic priorities, and recruits, develops, and evaluates senior analytic leadership.
So, here's your org chart:
OK, now that you got what you came for, here's the price: You have to consider the design above not an immutable multi-year reality, but rather a delicate, dynamic compromise that balances the objectives and tensions mentioned above. Where a group should live, as the footnotes suggest, and how relatively large each group becomes, should be a function of a constantly evolving portfolio of opportunities and challenges your governance process should determine. Conceiving of this not as a static chart prevents a pattern too often repeated -- a guardrail to guardrail "centralize versus distribute", "locate in the business versus locate in IT (or finance)" swing, where each successive move trashes the logic of the last. But that's a topic for another post...
What's your chart look like? Why? How often has it / does it evolve?
Via the WSJ, Emarketer's latest forecast has mobile ad spending eclipsing newspapers and radio (individually, not together, yet):
Ad spending on smartphones and tablets will this year eclipse radio, magazines, and newspapers, according to research firm eMarketer. Spending on mobile, which includes smartphones and tablets, will jump 83% in 2014 to nearly $18 billion, eMarketer predicts. Newspapers and radio, on the other hand, will draw nearly $17 and $15.5 billion respectively.
And yet, you will not be able to get full value from your investment, because the Revolution will not be Mobilized... until we can track and properly attribute the impact of advertising in this channel.
By various estimates, mobile now accounts for 30% of web traffic, but only 10% of ad revenues. Some, like Mary Meeker, make the case that this means mobile advertising spending still has lots of room to go. Here's my critique of her argument a couple of months ago. In that post I described demand- and supply-side factors that inhibit the "equivalency thesis" she outlines. One of these is understanding the properly attributed influence of mobile in a customer experience.
A couple of weeks ago I listened to Visual IQ's Solutions Consulting lead Andy Dubickas crisply describe three approaches for tracking a user across the desktop-mobile divide, for purposes of attribution analysis:
If you're not ready for this, you can at minimum set up your mobile ad buy with some variation in it -- waves, spikes, whatever -- that allow you to see, either with your naked eye or through modeling, the incremental direct and indirect effects of your investment. In fact, there's no reason the tighter-tracked approaches described above can't peacefully co-exist with this simple approach, even if it's just a cross-check.
(Apologies to Gil Scott-Heron)
Warning: due to author sloth, this post is dated. Most of you have likely moved on with your lives. Me, my Heart Is Full for los albicelestes, despite a disappointing outcome, but I still have a lingering case of futbolytics to get over.
Fivethirtyeight.com contributor Benjamin Morris had a fascinating article a couple of weeks ago that examined Argentine footballer Lionel Messi's play in recent years. Until this year's World Cup of course, Messi had come under some criticism for, against the backdrop of his ethereal play for Barcelona, under-delivering for Argentina in recent years. The article -- warning, 4600+ words long, with charts and videos and hyperlinks and footnotes, as in "Ask not, Bill Simmons, for whom the bell tolls..." -- explains persuasively that Messi is an outlier among outliers, even in his performances for the Argentine national team.
The analysis of his shooting is fascinating enough. But what really caught my attention, was the analysis of his passing and other influences on the game. In particular, here's a fascinating chart from Morris' article that makes the point neatly.
My friends at the multi-channel marketing attribution analytics firm Visual IQ are fond of a soccer metaphor to explain what they do. "Giving all the conversion credit to the last-touched marketing channel is like giving Mario Götze all the credit for his goal in the 113th minute of the World Cup final," they say (with a zesty cruelty that borders on the sociopathic).
In the sales world, it's a holy grail to get to this kind of dynamic, or even an understanding of where on Morris' chart members of your team would be. It can be a thorny path to get there though, because unlike in Messi's world, a helping hand in sales can be harder to observe, and even if you try to measure it, often it can be (will be) gamed mercilessly and unhelpfully.
One way you can track this sort of thing is through online and offline knowledge sharing by members of your team. Winning proposals, presentations, good answers to FAQs, and then views of these by others can all be tracked in relatively painless, game-free ways. A number of years ago when I worked at ArsDigita, we worked with Siemens to build ShareNet, a global sales and marketing knowledge management system that for many years was a poster child for applications of its kind (see here for the HBR case study). The secret behind Siemens' success with ShareNet was the flexibility with which it could adapt what was captured, and how, to make it easy for people to contribute and consume. Today, fortunately, the tools and costs for building capabilities like this are far more accessible. And now as attribution analysis moves closer to the center of the marketing analytics agenda, we have the opportunity to put the resulting data to work in a way that moves the dominant motivation for this kind of behavior beyond altruism to proper credit.
So if you'd like to improve your organization's gol-orientation, perhaps it's time to compile and publish your own assist chart?
In the June 23 edition of The New Yorker, in an article titled "The Disruption Machine", Harvard history professor Jill Lepore critiques the theories of Harvard Business School professor Clayton Christensen. Her article is wonderfully written of course, but curiously sharp in tone (or is it me?). Re-reading it, I have the impression she's sick of the Manifest Destiny - like over-simplification and over-application of Christensen's theories:
Most big ideas have loud critics. Not disruption. Disruptive innovation as the explanation for how change happens has been subject to little serious criticism, partly because it’s headlong, while critical inquiry is unhurried; partly because disrupters ridicule doubters by charging them with fogyism, as if to criticize a theory of change were identical to decrying change; and partly because, in its modern usage, innovation is the idea of progress jammed into a criticism-proof jack-in-the-box.
But unfortunately the baby appears to go out with the bathwater. Lepore suggests -- accuses -- Christensen of cherry-picking his data, and describes a dynamic in business scholarship (as distinguished from practices in other fields) that over-promotes findings. In taking aim, fairly or not, at Christensen (who beyond his reputation as a scholar is even more respected for his integrity), Lepore goes for the jugular of the academic-industrial complex.
Professor Christensen replied at length to Lepore's piece in an interview he gave last Friday to Bloomberg Businessweek reporter Drake Bennett, taking significant issues with her points -- maybe too personally?
Well, in the first two or three pages, it seems that her motivation is to try to rein in this almost random use of the word “disruption.” The word is used to justify whatever anybody—an entrepreneur or a college student—wants to do. And as I read that, I was delighted that somebody with her standing would join me in trying to bring discipline and understanding around a very useful theory. I’ve been trying to do it for 20 years.
And then in a stunning reversal, she starts instead to try to discredit Clay Christensen, in a really mean way. And mean is fine, but in order to discredit me, Jill had to break all of the rules of scholarship that she accused me of breaking—in just egregious ways, truly egregious ways. In fact, every one—every one—of those points that she attempted to make [about The Innovator’s Dilemma] has been addressed in a subsequent book or article. Every one! And if she was truly a scholar as she pretends, she would have read [those]. I hope you can understand why I am mad that a woman of her stature could perform such a criminal act of dishonesty—at Harvard, of all places.
Why am I interested, and why should you care?
Christensen's research into innovation, and his books like The Innovator's Dilemma and its successors have pretty much dominated the landscape on the subject for nearly the last twenty years. If you're a senior executive at a big established firm, he's made you paranoid. If you're an entrepreneur at a small, ambitious firm, you're telling your investors, customers, and employees you're one of the "disruptive innovations" making those executives paranoid. There's even a slim chance you might be right! Lepore is a Pulitzer Prize and National Book Award winner (as well as a department chair at Harvard). Scholars of her prominence usually don't take on other scholars, particularly outside their field, without good reason or care. Christensen's reply suggests she had neither, and prior critiques of her work suggest this might be the latest in a pattern of occasional trip-ups. In any event, the critique's attracted some attention. If you follow either of these folks for professional reasons (or simply are drawn to MMA-style clashes between academic titans) you might be interested to track this story too.
I'm interested because the controversy highlights a pattern I often see, and as I describe in my new book Marketing and Sales Analytics, I use a variety of techniques to compensate for.
The pattern is to treat ideas and supporting evidence like Christensen's as the last word on a subject. After all, he teaches at Harvard Business School, right? But even Christensen says (in his Bloomberg Businessweek interview) that models and supporting stories like his are at best just partly predictive points of departure for any specific "truth" I need to be tracking, and hedging against.
The psychologist Daniel Kahneman, who wrote the best-seller Thinking, Fast and Slow about his research into cognitive bias, talks about the power of stories, and in particular about our susceptibility to them and to other biases when we're tired, frazzled, juggling a thousand things. I think Christensen's theory of disruptive innovations invites criticisms like Lepore's because it mirrors, purposefully or not, an especially powerful archetypal story, in this case of the hero who battles long odds but ultimately triumphs grandly.
Recent history and scholarship has reinforced the power of this particular theme. The Internet revolution has created a number of poster children for it (I'm reminded of Paul Simon's line "Every generation thows a hero up the pop charts", and in our generation we've had several), and created conditions for accelerating the pace of disruption. Nassim Taleb's scholarship, in particular his own best-selling book The Black Swan, has sensitized us to the existence of outlier possibilities our models often miss (and even explicitly discard to improve fit).
So the combination of a story model we're especially tuned for, and recent examples that magnify its power, position it uniquely in our consciousness. The implication for executives is to be mindful of the bias this creates, and to have appropriate mechanisms in place for managing it.
In my book I describe a portfolio-driven approach to managing analytic efforts that can help with this. In brief, the idea is to manage analytic projects not just as a list of questions to be checked off as you answer them, but as a venture capitalist's portfolio of investments that need to generate target returns appropriate to their size and riskiness. In the governance of such portfolios we set up in our work with clients, we manage to a rule we call "3-2-1". In any given quarter, we aim to have the collection of analytic initiatives we're pursuing yield (notionally) three "news you can use" insights, two experiments based on those insights, and one outcome we're putting into production at scale to help pay the freight for the overall investment in the portfolio (including us). The portfolios are constructed to reflect priorities across a grid that itself is a collection of different business opportunities ("Need X for Customer Y") we're targeting and different purchase funnel or customer journey stages for these opportunities. This grid helps you judge how concentrated you are on existing versus new opportunities, and on whether your investments are appropriately focused on bottlenecks in the relevant funnels or journeys.
The main point of using a grid like this, as opposed to just a list of individual projects, is that it forces you to think backwards in a data-driven way from the customer-defined strategies, goals, and objective performance of the business, and about whether you've got sufficient attention on the hot spots thereof. In the parallel world of innovation, I believe the opportunity for subscribers to Christensen's ideas is to neither accept nor reject "disruption dogma", but to ask themselves whether they've got sufficient "3-2-1" style attention and results on the hot spots in their business. "What customers are important to us?" "What needs are important to them?" "How well are we serving those needs?" "What's / who's out there trying to serve these same folks and their needs?" "What kind of progress are they making, and what's contributing or impeding that progress?" "What research, analysis, or testing should we be doing to stay close to potentially meaningful threats?" "Are the hedges these efforts represent probabilistically - in likely magnitude and yield - a good match for the risk the potential competition out there represents to our business?"
A "Clash Of The Titans" around business theories, while good sport for some, can cause a lot of anxiety for many others trying to build and run their businesses. Framing your approach specifically and objectively using the techniques for managing analytic efforts can help you get past these concerns in a practical, tailored - and maybe even heroic - way.
Mary Meeker's annual Internet Trends report is out. It's a very helpful survey and synthesis of what's going on, as ever, all 164 pages of it. But for the past few years it's contained a bit of analysis that's bugged me.
Page 15 of the report (embedded below) is titled "Remain Optimistic About Mobile Ad Spend Growth... Print Remains Way Over-Indexed." The main chart on the page compares the percentage of time people spend in different media with the percentage of advertising budgets that are spent in those media. The assumption is that percentage of time and percentage of budget should roughly be equal for each medium. Thus Meeker concludes that if -- as is the case for mobile -- the percentage of user time spent is greater than budget going there, then more ad dollars (as a percent of total) will flow to that medium, and vice versa (hence her point about print).
I can think of demand-side, supply-side, and market maturity reasons that this equivalency thesis would break down, which also suggest directions for improving the analysis.
On the demand side, different media may have different mixes of people, with different demographic characteristics. For financial services advertisers, print users skew older -- and thus have more money, on average, making the potential value to advertisers of each minute of time spent by the average user there more valuable. Different media may also have different advertising engagement power. For example, in mobile, in either highly task-focused use cases or in distracted, skimming/ snacking ones, ads may be either invisible or intrusive, diminishing their relative impact (either in terms of direct interaction or view-through stimulation). By contrast, deeper lean-back-style engagement with TV, with more room for an ad to maneuver, might if the ad is good make a bigger impression. I wonder if there's also a reach premium at work. Advertisers like to find the most efficient medium, but they also need to reach a large enough number of folks to execute campaigns effectively. TV and print are more reach-oriented media, in general.
On the supply side, different media have different power distributions of the content they can offer, and different barriers to entry that can affect pricing. On TV and in print, prime ad spots are more limited, so simple supply and demand dynamics drive up prices for the best spots beyond what the equivalency idea might suggest.
In favor of Meeker's thesis, though representing another short term brake on it, is yet another factor she doesn't speak to directly. This is the relative maturity of the markets and buying processes for different media, and the experience of the participants in those markets. A more mature, well-trafficked market, with well-understood dynamics, and lots of liquidity (think the ability for agencies and media brokers to resell time in TV's spot markets, for example), will, at the margin, attract and retain dollars, in particular while the true value of different media remain elusive. (This of course is one reason why attribution analysis is so hot, as evidenced by Google's and AOL Platform's recent acquisitions in this space.) I say in favor, because as mobile ad markets mature over time, this disadvantage will erode.
So for advertisers, agency and media execs, entrepreneurs, and investors looking to play the arbitrage game at the edges of Meeker's observation, the question is, what adjustment factors for demand, supply, and market maturity would you apply this year and next? It's not an idle question: tons of advertisers' media plans and publishers' business plans ride on these assumptions about how much money is going to come to or go away from them, and Meeker's report is an influential input into these plans in many cases.
A tactical limitation of Meeker's analysis is that while she suggests the overall potential shift in relative allocation of ad dollars (her slide suggests a "~$30B+" digital advertising growth opportunity in the USA alone - up from $20B last year*), she doesn't suggest a timescale and trendline for the pace with which we'll get there. One way to come at this is to look at the last 3-4 annual presentations she's made, and see how the relationships she's observed have changed over time. Interestingly, in her 2013 report using 2012 data, on page 5, 12% of time is spent on mobile devices, and 3% of ad dollars are going there, for a 4x difference in percentages. In the 2014 report using 2013 data, 20% of time is spent on mobile, and 5% of media dollars are going there -- again, a 4x relationship.
So, if the equivalency zeitgeist is at work, for the moment it may be stuck in a phone booth. But in the end I'm reminded of the futurist Roy Amara's saying: "We tend to overestimate the effect of a technology in the short term and underestimate its effect in the long term." Plus let's not forget new technologies (Glass, Occulus Rift, both portable and large/immersive) that will further jumble relevant media categories in years to come.
The book marks a watershed moment in the Big Data zeitgeist. Much of the literature on the topic to this point has been more evangelical, telling us how analytics will make us all taller, smarter, and more handsome. But the general sense for me has been of stories that are "way out there" for most organizations. This latest book is much more about how to realize these visions with tactical, practical prescriptions across a range of issues.
Perhaps the most important of these dimensions is having a clear idea of the challenges or opportunities for which Big Data might be a part of the solution. In Chapter Two, Davenport presents a very helpful series of use cases for using Big Data in several industry applications, including business travel, energy management, retail, and home education. He pushes further to examine the relative readiness of a number of different industries and business functions, including marketing and sales (which are the particular focus of my own upcoming book, Marketing and Sales Analytics). In Chapter Three he builds on these examples and sector assessments to offer a framework for shaping business strategies that leverage Big Data. He suggests cost reduction, time reduction, new offerings, and decision support as broad objectives for focusing Big Data initiatives, and then further suggests a useful distinction between discovery-oriented application of Big Data (say, for sorting out emergent patterns of behavior to address) and production-oriented usage (applying Big Data to personalize experiences based on which emergent patterns might be worth the effort).
This "ends" focused approach to applying Big Data, in contrast to an "If I build it (my giant Hadoop Cluster) they will come" is an extremely valuable perspective to have introduced at this point in the evolution of this trend, and Davenport has wrapped it in a clean, well-organized package of specific advice executives interested in this space can profit from.