This is a development which I expect enterprise search vendors like Mercado and Endeca will watch with interest, as it both complements and competes with what they do. (Anybody know how Omniture is pricing SiteSearch?) This is part of a bigger trend already underway for a decade, though my sense is the vendors' features are perennially ahead of their customers' bandwidth (time and resources) to leverage them -- namely, the automation of the feedback loop between analytics and execution elements. See for example Listrak's nice explanation of its dynamic email campaign solution. Maybe a step toward an alternative future for Skynet as online merchant?

What does this mean for marketers relying on analytics to make decisions? Analytics are built on regressions on past activity and tests across present efforts. With recent developments, it's highly likely that many regression-based models that look backward to predict ahead will be obsolete, especially for highly seasonal businesses. So, this should mean a shift in analytic efforts toward testing in general, and consequently toward test-friendly media (e-mail, search, display, lightweight direct mail) and easier-to-execute components (copy, price) to help firms feel their way forward to a better understanding of how consumers will react under the new realities.

Unfortunately, in many cases, firms will be organizationally constrained from making this important adjustment. Skills and resources for regression-based analytics are different from those for test-based analytics. Big-bang model building efforts of the past year or two may create a drag on the ability to gather and re-direct the necessary budgets to test more extensively and aggressively. But marketers that took a pragmatic, efficient, and balanced approach to analytics will find the transition an easier one to make.

Related: Monster.com results for "web analytics"

Skepticism about the advertising value of social media by Pat LaPointe in the Online Metrics Insider newsletter.

Good practical advice about testing from Aaron Smith in the Email Insider Newsletter on email testing, consistent with much of the advice we also offered here.

From Steve Smith's article in the Mobile Insider, some useful metrics on mobile browsing and mobile search in particular.

A good public-sector example of excellent reporting on the right issue, relevant to all of us right now, is electoralmap.net. This service uses state-by-state contract prices from Intrade, the world's largest public prediction market, to predict the outcome of the Electoral College vote.

Reading Dailykos and Michelle Malkin have me convinced that regardless of how any of the candidates perform over the last month and a half of the campaign, 99.9% of the voters have already made up their minds. And right now, according to electoralmap.net the election appears to be a dead heat, with only Colorado's 9 electoral votes hanging in the balance. Looks like the DNC was fairly prescient in choosing Denver for its convention!

Should we believe it? This analysis suggests we can. Further, in prediction markets, market volume is a proxy for sample size. A closer look at the trading in Colorado (in the left-hand nav, go to "politics", then "US Election by State", then expand "Alabama-Florida" and look at the Colorado contracts, where it currently looks like Obama trades at $5.30 for a $10 payoff, and McCain trades at $4.70) indicates a total of about 2600 contracts in the market, for a total contract value traded of $26,000. That's not too much, but with a 3-point bid-asked price spread on the Obama and McCain Colorado contracts, it's enough I'd think to begin to attract trading by folks with inside local knowledge away from the main contracts ("2008 US Election" in the left hand nav), which collectively have $12 million in contract value traded, but where the bid-asked spreads are only 10-20% what they are in the Colorado contracts.

So, this is a fancy way of saying that, if you follow the money, "It's Colorado, stupid!" It will be interesting to see if the national election coverage in major online outlets begins to highlight places where things are really tight and selectively aggregate news and opinion from them.

And congratulations to the electoralmap.net guys for creating one of the most imaginative, useful, and usable mashups for following and filtering the election. But may I suggest a widget, or a FB app, to boost traffic? At only 6k uniques a day, you're really missing a big opportunity! Then maybe call Disney about a sponsorship to support its release of Swing Vote on DVD.

• What happened in November 2005?
• What's sustained the growth in interest since then?

My guess is that the huge spike in interest in Q4 2005 was related to Microsoft's PR push for its release of Yukon (SQL Server 2005 with Analysis Services).

Tom Davenport's HBR article "Competing On Analytics" appeared in January 2006, followed by the book of the same name in March 2007.  Ian Ayres' eminently-readable Supercrunchers appeared in August 2007.  The spike in press interest in the topic this summer appears to have coincided with yet another SQL Server release, highlighting the influence of the Microsoft marketing machine once again.

More broadly, analytics is being brought to the fore by the confluence of a bunch of different things:

• a critical mass of complementary data sets, in electronic formats as more behavior occurs through electronic channels (prediction: following the lead of others, Google will soon add "data sets" as a specialized category you can search on, as it has with so many others already
• bandwidth, storage, processing power, grouped as cloud computing utilities
• the software to go with them, not just from MSFT, but also from folks like Sun
• the maturation of standards for integration of different data sets, making the whole mashup trend possible

Today however, our reach still largely exceeds our grasp. The bottleneck to future growth looks to be fluency, in both the computer languages and tools required to assemble and manipulate data, as well as in the statistics to interpret them. It's particularly interesting in that light to note the geographic concentration in India for searches on the term, as a proxy for where future leadership on the topic might come from.

The first and most important lesson has been that there is no market (so far) for "marketing analytics services" per se. All of our assignments so far have been motivated not by our clients' needs to "bolster their analytics capabilities", but rather by something more closely related to the "story" of the business: improving performance, figuring out how to incorporate and expand the use of digital (and especially social) channels in the marketing mix, or a major event or change of some kind (like launching a new product, channel or business). 

Consequently, our assignments have been more "executive" than "functional".  For example, we've been retained to design and run overall marketing programs, or to advise at this level.  In these contexts, "analytics" have come to the fore as means to an end, and "building the analytics capability" has been an ancillary deliverable.   So, our starting point has been to look for flags -- things that seem odd or missing, or quantitative performance that seems out of place to us, based on internal and external benchmarks, the "business story", and our own experience -- as a point of departure for where to focus "marketing analytics services".

For example, in a recent assignment with an overall charter to "drive revenue and build the brand", we noticed that one channel seemed to be performing below where we'd expect it to have, and its under-performance had become a bit of a self-fulfilling prophecy.  Looking more closely at it, we noticed two things.  First, the number of people we were targeting seemed low.  Second, looking back over a year, there were significant differences in the performance of offers with certain themes as compared to others, and the lower-performing themes had been over-emphasized.  A match-back analysis revealed information we had internally to allow us to target several times the number of customers, and that this disconnect was due to some broken technical plumbing that could be  fixed easily.  On the theme front, by mining past order data to organize customers into a few straightforward purchase-behavior categories and targeting accordingly, we identified opportunities to raise the performance of themes that hadn't previously been doing as well as we thought they should have.  Together, these improvements when implemented will be worth a lot of money, both directly in terms of the increased value from this channel alone, and indirectly in terms of the substitution  potential for segments deemed only marginally profitable if reached through other channels as part of the overall marketing mix.

Second, with the exception of the biggest firms (and even then rarely),  we haven't seen "analytics" get executed as monolithic efforts where complex models are built on huge data sets in separate labs.  Rather, "analytics " tend to happen as a series of ad hoc queries, regressions, and tests across disparate data sets and interfaces, prioritized by the opportunities of the business, rather than by "build it and they will come".  Aimed at "on-the-fly" tuning of the business tied to the "story" we described above, these efforts have been justified on a "pay-as-you-go" basis, with tight cycles for finding, validating, and realizing opportunities to bootstrap investments in further opportunities.

For example, we were recently asked to offer a proposal to help a firm better use online channels to drive offline sales.  The range of  issues we were asked to address was quite broad, but in the end the firm chose based on more focused expertise in its sector, and especially with tuning sector-specific channels, with a refined knowledge transfer objective: "We need to get smart first, and walk before we can run."

Third, for any given analytic insight, be it customer segmentation and associated offer/channel tuning or something else, there seem to be two versions: the "great" one that fully exploits the opportunity but is hard to grasp and often harder to implement, and the "good" one that is straightforward to grok and to go after.  Don't try to start the car in fifth gear -- do the simple version that gets 30-40% of the benefit first, and use that momentum to fund the cash and credibility, and develop the informed perspective, you need for shifting up.  More specifically, I suggested recently that analytics shouldn't get more than three months ahead of the ability to act on the answers.

Smart analytics platform vendors get this.  Recently I've spent time with the top teams at two such firms, VisualIQ and Proclivity Systems.  Each of them have sensible migration paths for ramping up practically to the full value of what they can do, through sequencing the data sets they combine and focusing the breadth and depth of the customers they target.

Fourth, while the "data deluge" of the digital world demands that you improve your infrastructure for taking advantage of it, I've come to believe that we need to think about building a "data warehouse" as a continuous process and not a single or episodic deliverable.   To this end, what's important is having

• a map that shows where all the data lives,
• a dictionary that explains what piece of data means,
• a directory that explains how to get to the data,
• a guide that describes how routine analytics get done and associated decisions get made, and
• a library (searchable at minimum, if not indexed) that stores all of the individual bits of research (again: queries, regressions, tests, benchmarks) and conclusions you've reached in the past. 

These artifacts need to be living documents, like Wikipedia, rather than static editions.  But like Wikipedia, they need to be "curated". Or if you prefer a different metaphor, think of them as parts of an "insight engine" for driving your business, which collectively need to be lubricated and maintained to run your business at peak performance.  Based on recent assignments, I'm convinced there's a 20-30% time-to-market advantage associated with having these things up to snuff.

(As for technical issues, and in particular what's required to support integration, "hit-and-run" manual integration ("dump flat files out of systems X and Y, import, scrub, integrate, and analyze them in Excel or something more exotic) is expedient but unsustainable for anything you plan to do more than once or twice a year, as for planning a seasonal catalog mailing.   At the same time, trying to get everything into a single system is likely to pay off only if you plan to crunch that data very frequently, like every day or at least a few times a week, like if you're targeting offers in real-time in an online store.  For channels and campaigns with "intermediate" timing, like email, or changing online ad copy or offers through affiliate programs, a "Service Oriented Architecture"  ("SOA") approach (to support "mashups" off a number of "data warehouselets" for analytics projects) may make more sense.

You also need to consider the skills and experience of your marketing team.  If every analytics project requires help from IT to assemble the necessary data, you may need some sort of intermediate interface and / or training to ease that bottleneck.  Conversely, trying to build the ultimate interface is a quixotic quest guaranteed to be crushed under its own weight at some point.  An emerging litmus test for me is the degree to which marketers in any given firm have moved away from using spreadsheets (Excel) to databases (Access, etc.) for ad hoc analytics.  Either extreme is problematic, though I haven't yet seen the marketing team that is 100% SQL-literate.  Great tutorial for non-programmers here, BTW.)

All this has led to a deeper reflection on the meaning of the current "analytics" craze.  How much is hype, and how much is real?   The first thing to consider is that we're all using "analytics" already, to differing degrees, to answer "how can I usefully define target customers, and what combination of offers, communications, and channels gets them to respond?"  So prioritizing "analytics" is really more a question of degree and direction rather than a discrete decision.  In fact, ultimately, it's about business governance.  How much decision-making -- and for that matter, learning -- will I centralize, through models and tests developed and interpreted by analytic "experts", versus delegate, through selection, training and development, to managers and employees?

To make "analytics" a higher priority for your business going forward, you have to believe:

• more sophisticated tests and models would provide insights and potential returns that would justify prioritizing them
• these enhancements lie within what's practically and affordably possible

These tests move the prioritization decision away from philosophical evangelism to practical assessment, based on how competitively sophisticated you are in parts of your marketing program where it matters to be so.  They also move the decision away from one to generally surf a trend, toward one to selectively focus building your capability.

So, we're tuning the business to address these lessons...

Based on direct client experience, here are three themes I've worked out so far about how to approach building a firm's "marketing analytics" capability:

• Analytics  should answer high-priority questions about how to grow the business.  In one client engagement, we've built a simple financial sensitivity model to help us figure out where the biggest leverage from analytic "nuance" would be, and made sure up front that the potential payoff is worth the research effort.
• Analytics shouldn't be pursued without ready access to, understanding of, and trust in the underlying data.  In another situation, we're untangling several years' worth of Webtrends report cruft.  It's not only unclear what the reports are saying but also what data they are based on.  More than likely, we'll start from scratch, and rebuild a very limited set of reports which we completely understand.  We'll make better, faster decisions as a result.
• Analytics shouldn't get more than three months ahead of the ability to act on the answers.   Each segmentation or test "cell" you define is pointless unless you can actually execute against it.  Beware diminishing marginal returns and exploding operational complexity of finer and finer slices of the pie, and remember that "modularity" only goes so far before you spend so much time building towers of abstraction that you never actually get to market.

So, if your analytics project isn't clearly tied to answering a question a senior operating executive cares about, or you can't "see the data" in the analysis, or there's nobody around to do something with the insights, you're "overhead", at risk, and rightly so.  A good alternative in these circumstances is to partner with an academic researcher who's willing to do the work and give you an advance peek at the insights in exchange for access to your data set.

I'm not much on slogans or buzzwords (though after listening to this, maybe I should be).   But my head kept turning over the words "pragmatism" and "analytics".  Finally, thunder struck:

pragmalytics

One concern: services like this lose a bigger picture of "I'll give you this, you give me that" so important to reaching consensus and driving progress from an evolving middle.  To be fair, OpenCongress has tried to address this somewhat by showing how lawmakers are aligned in their voting.  However, you have to work hard to figure out possible horse-trading, since it's difficult to grade any individual piece of legislation on a single dimension. 

OpenCongress has written a really good "about" page that describes its present state and future direction.  It would be interesting, perhaps, to crowd-source an assessment of how economically and socially liberal or conservative a bill is.  Or, to associate new bills with past legislation.  Certainly, to allow users to tag legislation (as they've suggested).  All together: "Senator X and Senator Y frequently vote together on health mildly progressive health care legislation but part ways on radical defense policy".  This would allow us to see what's got a chance and what doesn't -- ultimately the crucial question if we want action, not just ideological posturing and gridlock.  Also, it will be interesting to see if in the future, services like these allow users to track -- or project -- legislators' stances on bills (based on how they do or don't announce support or opposition), and what effect that has on the process.

Postscript: Marshall Kirkpatrick's review on ReadWriteWeb

Recently I've been reading Ian Ayres' "Super Crunchers" and Tom Davenport's "Competing On Analytics", both aimed at general audiences.  I'll publish reviews on each book here shortly.  For now, just this observation, based on our experience so far: there's certainly lots of potential for "analytics", but realizing it depends on asking the right questions, and on being able to assemble data and systems to analyze, act on, and track this potential.   The biggest leverage comes from blending these skills.  As a result, we're more confident than ever that we're onto the right problem at the right time.

Super Crunchers: Why Thinking-by-Numbers Is the New Way to Be Smart

Competing on Analytics: The New Science of Winning

Today, we have real-time video streaming (now -- though not always practically -- in 3G) via folks with Nokia N95's and Qik.  Qik lets people know you are streaming via Twitter, and you can filter these "tweets" with Summize (which I wrote about yesterday)  You can also get your Qik streams onto YouTube automatically.  Spinvision, a brother to Twittervision and Flickrvision, lets you see videos as they are uploaded to YouTube -- superimposed on a map of the Earth.

Now let's roll ahead 12-18 months.  N95's won't be the only devices with high quality camera/ video capture and GPS capabilities -- so, many more people will have this capability.  3G will be more widely available and adopted.  Twitter and Summize will be features of much larger players' services, so they too will move from the fringe to the mainstream as more people inevitable discover the utility of microblogging for different purposes, and the utility of filtering all that microblogging (and microvlogging).   Presumably, you'll be able to stream simultaneously on Qik and YouTube.   Google's just announced the availability of Google Earth running in a browser (though strangely, they didn't keep in sync with the release of Firefox 3.0), so we'll be able to  make our mashups even more dynamic and accessible.  Throw in a little facial recognition to boot, while you're at it.

What does all this add up to? A crowd-sourced, global/hyper-local, digital video, roll-your-own-channel, keep-your-friends-close-and-your-enemies-closer news network. 

What does that make you?

Postscript:

Imagine if rather than turning over a videotape to the authorities, she had streamed this.  Or if Zimbabwe, Darfur, Afghanistan, Iraq, or New Orleans for that matter, were live and unedited, 24/7, from a thousand sources each.   How will that change us?

My read:

• For social networking to be really valuable, it's got to be really relevant to and really usable by a large number of people.
• Today, these two conditions are true only for 15-25 year olds and for tech early adopters and evangelists (of all ages).
• The way forward for relevance lies not in thinking "social-networking-out", but in "high-value-use-case-back".
• The way forward for usability lies in "opt-out" vs. "opt-in" interface engineering.

This "second phase" of social network service development will be harder than the first, since "build-it-so-they-will-really-use-it" is much harder than "build-it-and-they-will-come".   I know this from first-hand experience:  I'm on Facebook, LinkedIn, Twitter, Typepad and Second Life, and a dozen more I've forgotten about, plus a number of group and enterprise versions of these services, and I've aggregated and syndicated my content and others' across these services.  In fact, in a number of these settings, I'm the organizer/"evangelist".  (Here's my FriendFeed page.)  I'd say 20% of the people I know (and my sample is biased toward early adopters) are active users of these services -- they regularly update  "What are you doing right now", update each other's Fun Walls, compare tastes in books and movies, try to enroll each other in causes, etc.  The other 80% sign up and are passive, if they pay attention at all.

What will it take to make them active users, so that we can "monetize them"?  We need to consider three things:

• they are *not* early adopters -- they are much more utilitarian in their thinking about tools to use;
• introducing something *new* to them means replacing something *else*;
• "Replacing" is a process that must be thought through on multiple levels: rational, financial, emotional, social, and practical.

Here's one example:

Last week I participated on a panel called "What Blogging Brings To Business" at the Enterprise 2.0 conference in Boston (covered by moderator Jessica Lipnack and panel-mates Patti Anklam, Doug Cornelius, and Bill Ives, among others.)  In an SRO room with nearly 200 people, we bumped up hard against the knee in the adoption curve.  The dominant question was "should people in businesses blog?".   We had asked the audience to participate actively in the discussion, and debate raged (well, politely), Mars-Venus-like, on this point.

I suggested it was the wrong question.  Thinking use-case-back about the "what" and opt-out about the "how", we should be asking, "What should we write about, whom should we publish it for, and how can we make both the contribution and consumption of the resulting content as easy as possible?"  I offered trip reports or meeting summaries (sales calls would also work, perhaps even better) as a case in point.  No one would argue they are good business practice -- even more worthwhile if published and archived where people can read them and search them.  Where did you go?  What did you learn? 

Using MadLibs for insipration (or alternatives -- the "conference attendance summary" version of this one is just begging to be written), perhaps the public and enterprise blog platforms should provide templates, or allow users to define templates, for such reports.  After all, writing is intimidating and unpleasant for many people, especially when they know they're going to get wide distribution.  And, maybe the calendaring systems in which we blocked out our attendance at such conferences should send us a batched email at the end of each day with a links to our various blog platforms, and ask/ remind us to document what happened at each meeting?  Maybe these same platforms can be configured to automatically schedule meetings for 45 minutes and leave 15 minutes at the end for group or private summary? In enterprise settings, maybe alerts or reports get sent to managers when people have/ haven't documented things?

Here's another example:

I recently went to Argentina to celebrate my grandmother's 100th birthday.  I took lots of pictures with my BlackBerry Curve.  I've set up a private blog to share these photos with my family (which I prefer to Flickr for this purpose).  I've been tediously posting them to this blog from my BlackBerry via the email address Typepad's given me for this.  I had to RTFM to figure out how, and I still haven't finished uploading the photos.

A better way would have been for RIM to expose an API to its BlackBerry Curve camera app that SixApart could have built on, to allow me to download a Typepad  add-on that would have let me  pre-associate pictures taken from Argentina (the same way Google Maps for mobile knows where I am) with the appropriate blog.  Then, as I took each photo, it would have automatically made its way to this blog, after giving me a picture by picture opt-out to cancel the send or to post it to another blog.

What does this mean for the social network services themselves?

The biggest innovations have of course been the social graphs they have established, and their re-conceptions as application development platforms with APIs for third parties that leverage this information.  The smart platforms will now extend the power of their core-asset social graphs by adding "Who's collaborating with whom via what application?" to the simplistic "Who's connected to whom?", "Who's part of the same group?", or combinations thereof.  By knowing what applications we're collaborating through (along with the volumes, directions, timing, etc.), and the use cases these applications support, the social network services can better combine reach with behavioral targeting for the advertisers who foot the bill. 

Facebook's Beacon program was an admittedly clumsy foray in this direction.  Facebook was too focused on the viral marketing aspect of this thinking, and suffered a backlash as a result.    Today, when I look at the Facebook profile of a friend I know  professionally in the context of publishing and social media, I get served an ad for a stock trading service.  When I look at the Facebook profile of another friend I know in the context of a shared sports interest, I get an ad from a professional recruiter.  I'm guessing these imperfections won't last.  Out of necessity, folks with an interest in this universe are adopting the perspectives I've described.  And, as the right context gets considered in these services and ad response rates rise, the money will start flowing faster toward social media once more.

Om speculates that the way forward in terms of monetization is more niche-y  social networks.  He writes:

The way I see it, the market has shifted its focus onto niche social networks, such as those dedicated to sports, music, automobiles and pets. You know, sites like Dogster! They have focused, engaged communities, which means they can attract a higher amount of advertising dollars. (Liz came up with a taxonomy of social networks back in February 2007 that offers up an easy way to understand the nuances of the social networking landscape.)

Not only do they have a purpose, but they don’t depend on hit-or-miss behavioral targeting-based ad systems that many hope will one day turn social networks into a gold mine. After all, if you sell dog food, then everyone on Dogster is a potential customer...

The problem with this assumption is that being niche-ier doesn't necessarily mean you move beyond the usual "1% often contribute - 9% sometimes contribute - 90% never contribute" online community profile.  I believe Om's leap here -- that narrow-interest social networks have a purpose -- is flawed.  They may be more focused, and perhaps purposeful in that sense, but it doesn't mean that the applications they provide have been tuned to motivate usage.

Om's colleague Liz Gannes was thinking a year ago down the path I've suggested here, with the taxonomy of social networks she developed.  I think her classification scheme is a good one, but again I don't think we can make the assumption that just because LinkedIn is for business people that it's a use-case-based tool.  In fact, it's arguably the absence of such thinking and tools from LinkedIn that has been, to date, its biggest missed opportunity, and perhaps in the future the way it can regain the march that Facebook stole from  it.

For an upcoming post: top ten ideas for such apps.

Postscript: Josh Bernoff agrees.

Here's the embed code:

<iframe 
 src="http://78cb960a3fad4111ab601e6daa1f615c.c24eff9
67d6b41828058657c028c9946.zembly.com/things/78cb960a3
fad4111ab601e6daa1f615c;iframe"></iframe>

Cool service; easy to use, you can build versions of this to run on FB and other social networks. Took me less than an hour, most of that time spent getting the widget reduced to its essential elements.  Figuring out how to use Zembly took 5 minutes, and I didn't even need to RTFM.

(Sadly, I haven't yet been able to figure out how to track traffic to and from the widget with my Google Analytics account (since it's not accepting the Zembly url for the widget I'm offering), advice appreciated; I might try this.)

Broader point:  application widgets like this one (along with mini-games) are the future of the display ad business (in particular a good chunk of the rich media unit segment of that business), which has been criticized for skewed CTRs ("Natural-Born Clickers"), and which is trying to work out appropriate engagement metrics with the advertisers and media companies on either side of it.  Services like Zembly that make it easy to author such units will become this generation's Frontpage.

It occurred to me to think about this because today, everyone is a software and media firm in a very real way.  Web services, SAS, and Cloud Computing have changed some things considerably, like initial access to markets, but some things haven't changed at all -- today "syndication" is a fancy Web 2.0 way to say what we called "channel program" in 2002.  So while the names of the players, technologies, and deal particulars may have changed, the principles seem timeless to me, waiting for a new generation to discover them. 

So please have a look at the old article, and I hope to hear from you about where the parallels are solid and where we are indeed on new ground.

The opportunity:

Media firm publishes content.  Traffic spikes beyond the ad inventory it has sold and has available to run on that content.  Missed monetization opportunity. 

Or, media firm swaps in alternative ads, placed if/when excess page views get dumped into an ad network it may have cut a deal with.  Marginal revenue is good, but what it has *paid* to drive that excess traffic (via SEM, deal with a portal, allocation of precious space from its own home page, or otherwise) may be higher, making marginal profit on the page view negative.

So, media firm would like to optimize how it drives traffic.  Specifically, it would like some mechanism, as automated as practical, that would direct its traffic-driving mechanisms to the incremental page views with the highest marginal *profit*.

Getting there:

1. This requires having a way to calculate the marginal profitability of a page. (Let's ignore for the moment that what we should be saying is not page, but rather "unique piece of content with which an ad can be uniquely associated", for example a video unit.)  So let's unpack that a bit:

a. On the one hand, media firm has "orders" for ads, or "demand": advertiser X will pay for Y impressions of a given ad unit (or clicks, or actions, again let's ignore for the moment) at $Z CPM. (I don't know of course how they have represented these orders across their systems -- how broken up they are, how dis-integrated they are).

(1) these orders are subject to certain rules, like on what pages and when the ad should run.  (These rules are important of course, as they associate/ limit orders to relevant page inventory.)   
(2) media firm may have multiple competing orders -- orders generated via direct sales, vs. orders coming in through intermediary ad networks -- and rules for allocating/ prioritizing them. Media firm can decide how its page inventory and associated views may or may not be filled to different orders (e.g., don't want to set precedent by dumping premium pages into networks)

b. On the other hand, media firm has pages and page views, or "supply".

(1) these pages have a fixed cost (cost of producing and serving them). We'll ignore these costs for now: the former since it's fully sunk, and the latter since it makes things too complicated for now, even though it may be relevant in the case of rich content with high processing and bandwidth costs. 
(2) these pages also have a variable cost associated with driving traffic, sometimes fully sunk but "allocate-able" (link from portal or site home page), sometimes real cash (SEM).

c. So when a reader requests a page from media firm,  it needs:

(1) a message reporting page revenue, presumably from the ad servers populating the page with the available ads with the highest effective net CPMs, and the *projected revenue* from the *next request* for this page
(2) a message from media firm's log analyzer, so we know where that page request came from, which we can then pass through a "traffic cost table", culled from a variety of different sources, to calculate associated charges for that page and for the *next request* for that page through that *and other* sources

d. To figure out what's feasible here, we need to dig into *what's available from which system* on *what reporting cycle*. 

(1) for example, on the ad server side, the stream of information about whose, which, and how many ad units have been served isn't necessarily integrated with the information about the (net) price of those ad units. 
(2) more specifically, for example, of DART's bunches of tables and "modules",  I'm not sure which this media firm runs or how it may have integrated them (or not).
(3) on the traffic-driving side, we'd need to look at how media firm has set things up in AdWords, for example -- how big are its incremental buys, how varied are its keyword options for a given page.

2. Assuming it can get this "marginal value of a page" pulled together, media firm then needs to act on it.  This implies a traffic direction process, supported as much as operationally and financially practical by an automated system.  This process would:

a. report the value of the most recent page served

b. project the potential values of the next page to be served, by traffic source

c. suggest sensible (profitable) "traffic-driving" options across a range of pages

d. allow media firm to pre-set automated responses to those suggestions, and integrate these responses with relevant systems (e.g., shut off incremental SEM spend, re-sequence options for articles to be linked from its home pages, etc.)

e. report a "pre-" and "post-optimized" value, for a given page and for a range of page inventory, so media firm can see whether the time and investment it's making is useful or not

3.  This assumes, for simplicity's sake, that we're wrapping all of this around whatever demographic or behavioral targeting media firm may have going on to dynamically serve different ads to different reader profiles.  In other words, if media firm has that running, the incremental value of a page associated with reader "Joe Chef" may not be the same as the value of that page associated with "Jane Chef". 

a. So, media firm might want to "throttle" your traffic driving efforts against those different segments accordingly to the extent possible.

b. This means tagging traffic sources appropriately, and checking the accuracy of those "segment tags" periodically.

Pretty interesting to think about the potential value.  (Numbers disguised.) Let's assume $2 / "OPM" -- optimization per thousand pages; you heard the term here first ;-)  Let's assume 10% of media firm's 75M monthly page views are candidates for this.  That's 7,500 "optimization opportunities". That's $15k/month in incremental marginal profit, or $180k per year.  Playing with these assumptions we can begin to shape an R for the ROI calculation on this effort.

Who's doing this?

Prof. Chandy on the "Why?" for the conference:

• "Many options, few certainties"
• "Lots of data, little information"

Here are some notes on the talks (comments very welcome!):

"Long-Term Impact of Marketing Spending" -- Professor Dominique M. ("Mike") Hanssens, UCLA Anderson School of Management

• 93-94% of single marketing actions have no long-term lift
• Permanent impacts occur *almost exclusively* in emerging, evolving markets -- like HDTV's, social networks
• Cumulative long-term advertising effect = 2x short term effect.  Customers learn quickly, forget slowly; economic effects duration = 2 months
• Sales call resp = 5x ads, LT effect = 2x
• Sales promotions have no LT effects
• Sales promo impacts (Sunil Gupta 1988)-- 84% on brand choice, 2% on purchase quantity, 14% on category incidence; LT effects are 11%/ 23%/ 66%.  Meaning retailer is biggest beneficiary of LT effects of sales promos
• Can you get permanent lift from temporary action (I learned a new word: hysteresis)? Yes, when people are learning -- when the cement is still wet
• E-Curve vs. perceived utility -- former is steady decline, latter spikes up during initial E-Curve decline, then drops steeply v E-Curve, then recovery to E-Curve
• Example MSFT vs. Sony game, MSFT had one year availability lead, THAT was the time they should have marketed aggressively to build long-term brand value
• Cesar's takeaway -- really hard to build brand value with marketing spend in relatively stable, fixed markets.  SO marketing spend should focus on ST promotion.  Prof. Hanssen agrees with this conclusion in follow up Q&A
• The courageous exception: P&G shifted $260M from promo to advertising in 24 categories, 118 brands, lost 16% share but increased profits by $1B.  They can do it because they have tremendous distribution leverage, and reinvest some of the profits into product innovation that justifies higher premium (from fewer price promotions)
• Prof Hanssen's takeaways include conclusion that single marketing actions must pay for themselves -- can't count on LT effects, since whether or not these happen depend on timing AND a broader sustained and consistent application of marketing strategy and policy that individual initiatives can't typically count on.  Consumers are "cats, not cows" -- they require continuous prodding and they talk to each other, so it's hard to isolate price promo efforts.

"Evaluating the Business Impact of Brand Perceptions" -- James Henney, Wells Fargo

• "Jeff Bezos -- 'A brand is what people say about you when you are not in the room.'"
• Branding goal = "shift demand".  How well have we been able to identify this impact?
• Effects on customer acquisition?  Deepening existing relationships?
• Research
  • Key variables: Brand Characteristics ("Brand Performance Index")
    
    • Awareness -- have you heard of
    • Consideration -- would consider?
    • Preference -- prefer?
    • Distinctiveness
    • Relevance
    • Overall Impression
  • Execution
    
    • 6600 customers, apr-dec 06, 55 minute survey
  • trying to predict ownership and bal in checking and saving, measured jan 06 - feb 07
  • control for # HH's, # stores, competitive intensity
  • unit of analysis = zip code, 2k units covered
  • dependent variable = 3 month aggregate cust behavior (new HH, new check acct, new sav acct) before and after brand measurement
  • WLS TS regression
  • Findings
    • R^2 = .86 on # new HH, .91 on # new check, .92 on # new sav
  • stat significant
  • but 2/3 of credit can be attributed to "lagged vars", 10-25% to store distribution, 15-20% attributed to BRAND SCORES
  • So, how much increase in (new, deepening) business would be attributable to 10% increase in brand scores?  About 3-5% in first and second , flattens year three. Don't know how much it would *cost* to do this.
  • Enhancements to come -- *which* of the brand factors matter most?

"Stronger Relationships, Better Results: New Metrics of Loyalty and Engagement" -- Linda Vytlacil, Carlson Marketing

• Revenue is by far the most important driver of marketing investment decisions, followed by ROI and Profitability, Brand Equity, and cost
  • Much less consideration of competitive response, customer lifetime value, etc.
• Partnership between Peppers & Rodgers Group and Carlson Marketing to assess the impact of relationship strength, and manage it practically
• Proposing Relationship Strength metric -- "RSx" -- as an intermediate aggregate of a number of "antecedents"
• Interesting analysis of different effects, industry by industry (fiserv, retail, auto, airline, (telecom, high tech), of antecedents on Relationship Strength "RSx" metric
  • auto industry scores strongest overall and for mutuality and alignment (emotional factors cited), fiserv for trust, commitment (given higher switching costs)
• Strongest RSX to outcome linkage is on recommendation (validates Net Promoter Score framework)

"Marketing Effectiveness: Beyond ROI" -- Gordon Wyner, Millward Brown

Statistics from his firm's experience I hadn't seen before:

• Optimizing marketing mix yields 10-20% improvements
• Experimental testing increases yields by 2x

"Leveraging Loyalty Card Data to Measure ROI" -- Adrian Sosa, CVS Caremark

Adrian (a fellow Tuck grad) is part of the team that runs ExtraCare, the largest loyalty program in the US -- one out of every six people in this country belong to it.  The program has three components:  circular-based ExtraBucks, clip-free-couponing, and purchase-based targeted offers (based on personas and customer life-time value, driving 140 variants).  Adrian described the RFM-based approach behind the program (very little data augmentation / appending in their world), and the home-grown tools they have built to run it.  Salient feature -- the sheer scale of the program (10 billion database rows updated daily) allows for statistically rigorous tests of offers vs. a variety of controls. 

The program works:  for example, Adrian presented data showing a 0-4% revenue uplift associated with a 0.2-1.0% redemption rate (on de-seasonalized sales) for their ExtraBucks incentive, with the majority of data points concentrated at 1-2% uplift for 0.2-0.4% redemption.  Adrian says, "ExtraCare is *the* biggest differentiatior for CVS...", a perspective echoed by his boss the CMO, Rob Price. (Shameless plug: according to Adrian, Rob's a Monitor Group alum.)

More specifically, Adrian noted also that email offers, no matter how well-targeted and personalized, don't perform as well as POS/kiosk offers (my take is that this is likely characteristic of the category, generally composed of lower-consideration products).

(Related nugget Adrian shared that I didn't know: 75% of CVS sales are pharmacy, only 25% are "front of store", though profits are more like 50-50; the opportunity to grab more of the profit associated with all of these prescription drug sales would explain why it was so attractive for CVS to acquire PBM Caremark.)

(Also related, I had a chance to chat with Adrian about a "physical hyperlink" strategy for promotions, using SMS or possibly (in the future) QR codes; Adrian mentioned they've been experimenting with it and see it as promising but that to date samples have been small and results more sporadic.  My own thinking has been that since a mobile-device-based approach to tracking and couponing is less expensive than rolling out kiosks and more a more granular and accurate method than PRISM-style store traffic tracking, it's the inevitable way forward in retail, but I guess today's mobile-savvy teens aren't yet at an age to constitute the vanguard of incentive-motivated shoppers.)

"Customer Value in a Networked Society" -- Prof. Sunil Gupta, Harvard Business School

Prof. Gupta has taken a long-term interest in customer value and applied it to trying to understand better the relationship between customer value and enterprise value in social media.  In his research, Prof. Gupta has developed a model that better explains (vs. customer value alone) the relationship between the balance of "buyers and sellers" in markets like eBay to the value of such firms.  In short, if you multiply the discounted customer profitability (customer lifetime value, or CLV) of firms like Capital One, and E*TRADE by their respective number of customers, you get a pretty close approximation of their market caps.  But this doesn't work for properties like eBay, Monster, and Amazon, which have created user-to-user " markets as most or all of their businesses. 

This is really interesting and relevant research, because a "revenue per unique user" is a common valuation metric for today's Web 2.0 property M&A market.  In the Q&A following his talk, we explored the potential for large media firms to use this construct to properly balance advertisers against readers and subscribers for maximum enterprise value.  This should be of interest not only for web pure plays but also for large media conglomerates looking to reinvent themselves as  "online community" franchises.

"Linking Consumer Data to Business Metrics That Matter" -- David Krajicek, GfK

David's talk focused on the benefits of a more integrated approach to marketing data analysis, and on the mechanics of making that a reality.  More specifically, his model seeks to tie the impact of individual marketing investments through intermediate marketing-related operational metrics, to financially-based return metrics.  Of particular interest to me was a case study about MarCom strategy for a wireless carrier, where the model was used to optimize the mix of "reliability", "tech leadership", "value", and "non-theme-based awareness-building" imagery investments.   The growing experience of firms like GfK, combined with the increasingly powerful multi-channel analytic capabilities of firms like VisualIQ (whom I've gotten to know recently) together are accelerating the half-life of the "wasted half" of ad spending.

"Metrics of Pricing" -- Prof. Mark Bergen, Carlson School/ UM

Pricing as a marketing lever is often thought to be "costless".  It turns out to be anything but, for a variety of reasons, ranging from the sheer logistical complexity of changing listed prices, to the need to coordinate adjustments with other parts of the firm and with legal requirements, to the challenges of communicating the logic of price changes to employees and through them to customers.  Implementing pricing changes artfully, in harmony with the rest of a firm's marketing mix, turns out to be as much an organizational challenge as a logical puzzle that if mastered can be a real strategic advantage. 

"Budgeting and Allocation Across Corporate and Brand Advertising" -- Prof. Prasad Naik, UC Davis Graduate School of Management

Using Ford as an example, Prof. Naik demonstrated how by understanding the "brand sales elasticity of corporate advertising", and by managing brand advertising investments as a portfolio under a corporate umbrella, it's possible to make 1+1 >2.  In the Q&A, Jim Schroer (CEO of Carlson Marketing and former CMO of Ford) noted that he'd wished he'd had this analysis when he was at Ford, but noted that a lot also depends on the corporate campaign messaging.  For example, he observed that "Quality is Job 1" was a very complementary theme, a circumstance that might not always exist.  Others noted that much depends on having an established corporate brand that can help, rather than one that must first be built.

"Multi-Channel Marketing Optimization" -- yours truly

Here are the results of an audience quiz I did:

• about half of the relevant participants (i.e., those from big firms) report that they consistently and systematically optimize mix across channels; roughly a quarter say they do this somewhat, campaign by campaign; the balance indicated they do so rarely (much higher than I expected)
• about 2/3 indicated they have multi-channel marketing dashboards and data warehouses encompassing ~80% of available channels (again, much higher than I'd thought)
• but, only roughly one in five indicated they had active cross-channel teams, with related metrics and compensation

So, as with other things, our technological reach seems to exceed our organizational grasp.

"Measuring Tasks and Compensating Channel Partners: Guidelines and Consequences" -- Prof. George John, Carlson School/ UM

I never knew this, but apparently when IBM engaged Microsoft to write MS/DOS for the IBM PC, they paid them according to KLOC, or per thousand lines of code.  (MSFT skeptics will of course say that this explains so much.)  Here's a YouTube video Prof. John showed (excerpted from the PBS series "The Triumph of the Nerds" -- start watching at 4 minutes in) that really brought this issue to life.  A better way might have been to compensate Microsoft based on a base of time and materials, plus a grid-based  incentive, where the coordinates would be a function of functionality and performance.

My take-away from Prof. John's talk:  When conditions are "noisy" -- that is to say, a low correlation exists correlation between outcomes as measured by channel metrics, and the effort that a partner puts in -- you need a more sophisticated, multi-dimensional compensation scheme to properly align interests.

Panel: "Metrics of Growth" -- Greg Michaels, Kraft; Jeff Hunter, General Mills; Tom Sullivan, United Health; (moderated by Prof. Chandy)

Greg Michaels presented an in-depth look at how Kraft measures and reports Total Brand Value, as inspired by Jim Kilts' tenure there (and further described in his book Doing What Matters).  He also described how Kraft has worked with Canback-Dangel Predictive Analytic Integrators to blend a variety of inputs and models to predict demand for new products in emerging markets where pre-existing data is very limited.  For example, we looked at how Kraft assesses likely demand for "cheese for China"; turns out disposable income drives 50-100% more "elasticity of cheese demand" than any other factor.  Who knew?

(Unfortunately I wasn't able to stay for the discussions on Friday after lunch.  So, much appreciated if any readers can link to accounts of these sessions in comments!)

I really liked this conference:

• right content, right time -- with apologies to LOTR fans, "The time of digital evangelists is over, and the time of business people is at hand.  But we must have metrics to order all things as we will..."
• balanced mix of perspectives-- top marketing research/ analytics executives from some of the world's leading brands (Kraft, Unilever, CVS, Wells Fargo, American Airlines, Target, Astra Zenica, Seagate, 3M, and many others), combined with leading academics in the field, leavened with some trusted advisors with deep roots in the business and the research and data sets to show for them.
• intimate -- about 115 people
• run flawlessly -- even with last minute additions by folks like me.  (Thanks to Rebecca Monro and Letta Wren Christianson!)
• great location -- late-May Minneapolis was glorious; I enjoyed a run on the banks of the Mississippi, passing some interesting sights; and, if you haven't been downtown there lately (as I hadn't) you'd be surprised at how vibrant the scene is.  Prof. Chandy hosted two wonderful dinners for the speakers, including one at the tres hip Chambers Hotel, where we pondered art like this :-)

I look forward to tracking what's coming out of the Institute, and to visiting again soon!