Octavianworld

Fly-By-Wire Marketing, Part II: The Limits Of Real Time Personalization

A few months ago I posted on what I called "Fly-By-Wire Marketing", or the emergence of the automation of marketing decisions -- and sometimes the automation of the development of rules for guiding those decisions.

More recently Brian Stein introduced me to Hunch, the new recommendation service founded by Caterina Fake of Flickr fame.  (Here's their description of how it works.  Here's my profile, I'm just getting going.)  When you register, you answer questions to help the system get to know you.  When you ask for a recommendation on a topic, the system not only considers what others have recommended under different conditions, but also what you've told it about you, and how you compare with others who have sought advice on the subject.

It's an ambitious service, both in terms of its potential business value (as an affiliate on steroids), but also in terms of its technical approach to "real time personalization".  Via Sim Simeonov's blog, I read this GigaOm post by Tom Pinckney, a Hunch co-founder and their VP of Engineering.  Sim's comment sparked an interesting comment thread on Tom's post.  They're useful to read to get a feel for the balance between pre-computation and on-the-fly computation, as well as the advantages of and limits to large pre-existing data sets about user preferences and behavior, that go into these services today.

One thing neither post mentions is that there may be diminishing returns to increasingly powerful recommendation logic if the set of things from which a recommendation can ultimately be selected is limited at a generic level.  For example, take a look at Hunch's recommendations for housewarming gifts.  The results more or less break down into wine, plants, media, and housewares.  Beyond this level, I'm not sure the answer is improved by "the wisdom of Hunch's crowd" or "Hunch's wisdom about me", as much as my specific wisdom about the person for whom I'm getting the gift, or maybe by what's available at a good price. (Perhaps this particular Hunch "topic" could be further improved by crossing recommendations against the intended beneficiary's Amazon wish list?)

My point isn't that Hunch isn't an interesting or potentially useful service.  Rather, as I argued several months ago,

The [next] question you ask yourself is, "How far down this road does it makes sense for me to go, by when?"  Up until recently, I thought about this with the fairly simplistic idea that there are single curves that describe exponentially decreasing returns and exponentially increasing complexity.  The reality is that there are different relationships between complexity and returns at different points -- what my old boss George Bennett used to call "step-function" change.

For me, the practical question-within-a-question this raises is, for each of these "step-functions", is there an version of the algorithm that's only 20% as complex, that gets me 80% of the benefit?  My experience has been that the answer is usually "yes".  But even if that weren't the case, my approach in jumping into the uncharted territory of a "step-function" change in process, with new supporting technology and people roles, would be to start simple and see where that goes.

At minimum, given the "step-function" economics demonstrated by the Demand Medias of the world, I think senior marketing executives should be asking themselves, "What does the next 'step-function' look like?", and "What's the simplest version of it we should be exploring?" (Naturally, marketing efforts in different channels might proceed down this road at different paces, depending on a variety of factors, including the volume of business through that channel, the maturity of the technology involved, and the quality of the available data...)

Hunch is an interesting specific example of the increasingly broad RTP trend.  The NYT had an interesting article on real time bidding for display ads yesterday, for example.  The deeper issue in the trend I find interesting is the shift in power and profit toward specialized third parties who develop the capability to match the right cookie to the right ad unit (or, for humans, the right user to the right advertiser), and away from publishers with audiences.  In the case of Hunch, they're one and the same, but they're the exception.  How much of the increased value advertisers are willing to pay for better targeting goes to the specialized provider with the algorithm and the computing power, versus the publisher with the audience and the data about its members' behavior?  And for that matter, how can advertisers better optimize their investments across the continuum of targeting granularity?  Given the dollars now flooding into digital marketing, these questions aren't trivial.

Filtering The Collective Preconscious: Darwin Ecosystem

More and more, people agree that filtering the flood of information that's coming at us is supplanting publishing, finding, and connecting as the problem of the Information Age.  Today, the state of the art for doing this includes several approaches:

• Professional filters: we follow people whose jobs are to cover an area.  Tom Friedman covers international issues, Walt Mossberg covers personal technology.
• Technical filters: we use services like Google Alerts to tell us when there's something new on a topic we're interested in
• Social filters: we use services like Digg, Reddit, and Stumbleupon to point us to popular things
• Tribal filters: we use Facebook, Twitter, LinkedIn, and (Google hopes) Buzz to get pointed to things folks we know and trust think are important

In addition to what gets through, there's how it's presented.  RSS readers for example offer a huge productivity boost to anyone trying to keep up with more than a few sources of information.  However, once you get several hundreds items in your RSS reader, unsorted by anything other than "last in", it's back to information overload.  To solve this, innovative services like Newsmap provide multi-dimensional visual displays to try to push your information awareness productivity even further.  But so far, they've seen only modest adoption.

One limitation of today's filtering and productivity tools is that they pick items up either too early, before it's clear they represent something meaningful, or too late, once the advantages of recognizing a trend have passed.

Yesterday, I visited the team behind a new service called Darwin Ecosystem  that takes a different and potentially more powerful and useful approach to helping you "filter the collective preconscious"  -- that is, to identify emergent signals in the vast noise of the Internet (or any other body of information you might point to -- say, for example, customer service call logs).  Co-founder and CEO Thierry Hubert is a veteran of the knowledge management world going back to senior technical roles at Lotus and IBM; his partner Frederic Deriot shares similar experiences; and, my friend Bill Ives -- formerly head of Accenture's KM client practice -- is also involved as VP Marketing.

Briefly, the service presents a tag cloud of topics that it thinks represent emergent themes to pay attention to in the "corpus" filled by sources you point it to (in the demo, sources run to hundreds of news sources and social media).  The bigger the font, the more important the theme.  Hover your mouse over a theme, and it highlights other related themes to put them all into a collective context.  The service also provides a dynamic view of what's hot / not with a stock-ticker-style ribbon running at the top of the page.  You can view the cloud of emergent themes either in an "unfiltered view", or more usefully, filtered with "attractor" keywords you can specify.

This interface, while interesting, will likely not be the eventual "user/use-case" packaging of the service.  I can see this as a built-in "front page" for an RSS reader, for example, or, minus the tag cloud, as the basis for a more conventional looking email alert service.

The service is based on the math behind Chaos Theory.  This is the math that helps us understand how the proverbial beating of a butterfly's wings in China might become a massive storm.  (Math nerds will appreciate the Lorenz-attractor-plot-as-butterfly-wings logo.)  The service uses this math to tell you not only what individual topics are gaining or losing momentum, but also to highlight relationships between and among different topics to put them into context  -- like why "underwear" and "bomber" might be related. 

Now in beta, with a few large organizations (including large media firms) as early adopters, the service has had some early wins that demonstrate its potential.  It told users, for example, that Lou Dobbs might be on his way out at CNN a week before his departure was reported in the mainstream press.  It also picked up news of UCLA's planned tuition hikes 48 hours in advance of this getting reported in popular mainstream or social media.

It strikes me that a service like Darwin is complementary to that of  Crimson Hexagon, a sentiment analysis firm based on Prof. Gary King's work at Harvard (here's the software that came out of that work), with a variety of marketing, media, and customer support applications.  Darwin helps tell you what to pay attention to -- suggests emergent themes and their context; Crimson Hexagon can then tell you how people feel about these issues in a nuanced way, beyond simple positive / negative buzz.

The current business model has Darwin pursuing enterprise licensing deals with major firms, but depending on partners that emerge, that may not be the last stop on the adoption / monetization express.  For example, it seems to me that a user's interaction with a tool like Darwin represents highly intentional behavior that would be useful data for ad / offer targeting, or personalization of content generally.  This potential use as a marketing analytics input makes it especially interesting to me.

Bottom line: if you are responsible for syndicating and helping users usefully navigate a highly dynamic information set collected through a multitude of sources -- say, a news organization, a university, a large consumer products or services firm -- and are evaluating monitoring technologies, Darwin is worth a look.

Beyond #A/B and #MVT #Testing: Optimizing Experiences

I caught up recently with Jeff Eckman and Geordie McClelland of BGC to learn more about their service for deploying, testing, and optimizing direct-response page flows.  The easy-to-use service (I saw a live demo), which is based on ion Interactive Pres & CTO Scott Brinker's LiveBall platform, combines a template-based authoring tool, CMS platform, and testing engine.  A marketing organization creates a subdomain on its web domain, and then works with BGC to develop different landing "experiences" (sequences of pages arrayed along a logical path) on it, instead of a single landing page or static, product-focused microsites.  Then the system tracks conversion rates along and through these experiences and, with full transparency to the marketers, shifts a greater share of requests to the winner.  Meanwhile, the marketers develop new page elements, pages, and paths based on what the more granular feedback about those suggests might make sense.

Contrary to popular wisdom about how fewer clicks drive increased conversion rates, BGC has found that an experience that asks for an "optimum" amount of information in an "optimum" series of screens delivers far higher results: on average 300% higher and as much as 3000+% higher in the 3+ years this practice has been in use.  I asked if their experience across multiple clients had yet suggested an overall rule of thumb, such as, "For products in this price range, three pages and three questions to each page work best", to help jump-start the creative process and save on initial creative costs, but Jeff and Geordie hadn't seen one emerge yet, and were somewhat skeptical that it could / would be worth it to pursue.  I struggled with that answer for a bit, but it's occurred to me that maybe I need to break out of my heuristic-seeking box and accept that in 2010 and beyond, the Process is the Answer (see my earlier post "Personalization Is A Process").  Jeff and Geordie are putting their money where their mouths are on this point:  The service can be offered on an easy-to-try-and-buy CPA basis (after a very modest setup fee).

We discussed technical directions that could broaden the span of the experiences they can engineer and test.  For example, you might imagine that an experience might start within a Facebook application for a couple of screens, and then make the jump to your site for the rest of the process and, hopefully, a conversion.  While they haven't done one of these yet, since a url is a url and its contents can be rendered ecumenically, this should be straightforward.

I haven't seen all the direct and indirect competition yet, but BGC's solution seems fairly unique and easy to use.  In theory you could cobble a GA-based solution together that combines their funnel analysis with their Web Optimizer A/B testing tool, but that would be clunky and likely brittle, perhaps a false economy.  And, not too long ago I wrote about Sitespect's url tunneling capability, but it strikes me that while somewhat similar, the two are currently focused on different parts of the "attract-engage-convert" process.

Bottom line: have a look, especially if you use form-based landing pages to qualify and convert leads for higher-price-point products and services.

Ecommerce On The Edge In 2010 #MITX

Yesterday morning I attended MITX's "What's Next For E-Commerce" Panel at Microsoft in Cambridge.  Flybridge Capital's Jeff Bussgang moderated a panel that included Shoebuy.com CEO Scott Savitz, CSN CEO Niraj Shah, Mall Networks CEO Tom Beecher,and Avenue 100 Media Solutions CEO Brian Eberman.

The session was well-attended and the panelists didn't disappoint. Across the board they provided a consistent cross-section of the sophistication and energy that characterizes life 2 SDs the right on the ecommerce success curve.

My notes and observations follow. But first, courtesy of Jeff, a quiz (answers at the end of the post):

1. Name the person, company, and city that originated the web-based shopping cart and secure payment process?

2. Name the person, company, and city that originated affiliate marketing on the web?

3. Name the largest email marketing firm in the world, and the city where it's headquartered?

Jeff opened by asking each of the panelists to talk about how they drive traffic, and how they try to distinguish themselves in doing so.

Brian described (my version) what his firm does as "performance marketing in the long tail", historically for education-sector customers (for- and non-profit) but now beyond that category. What that means is that they manage bidding and creative for 2 million less-popular keywords across all the major search engines for their customers. Their business is entirely automated and uses sophisticated models to predict when a customer should be willing to pay price X and use creative Y for keyword Z to reel in a likely-profitable order. The idea is that the boom in SEM demand has driven prices way up for popular keywords, but that there are still efficient marketing deals to be mined in the "long tail" of keyword popularity (e.g.,structured collaboration").

Niraj noted that there's an increasing returns dynamic in the SEM channel that raises entry barriers for upstarts and helps firms like CSN preserve and expand their position.  Namely, as firms like his get more sophisticated about conversion through scale and experience, they can afford to pay higher prices for a given keyword than smaller competitors can, and can reinvest in extending their SEM capabilities.  CSN now has a 10-person search marketing team within its total staff of 500. Since SEM is, to some degree, a jump-starter for firms that don't yet have a web presence sufficient to drive traffic organically, this edge is a powerful competitive weapon.  CSN is up to $200 million in annual revenues, and now manages the online furniture stores for folks like Walmart.

Scott sounded a different note, with similar results.  Shoebuy has focused more on cultivating its relationship with its existing customers and on Lifetime Value -- including referrals.  This focus has had a salutary effect on SEO, allowing them to rely less on SEM as it gets pricier.  Last year Shoebuy experienced double-digit top line growth and hit 8M uniques for December's shopping season, while realizing its lowest marketing expense as a percentage of sales since 2002.  They've continued to plow the savings into a better overall customer experience.  One way Shoebuy guides this reinvestment is through extensive use of Net Promoter-based surveys.  They keep the surveys brutally simple:  1)"Were you satisfied?" 2)"Whould you shop with us again?" 3)"Would you recommend us?".  Then they calculate the resulting NP scores to different things they try in their marketing mix, to give them a more nuanced insight than the binary outcome of an order can provide.

Tom described how while Mall Networks' traffic is "free" -- it all comes from their loyalty program partners' sites (e.g. Delta Skymiles website awards redemption page) -- they still have to jockey for Mall Networks' placement on those pages. (Though Tom was too polite to say so, the processes for deciding who goes where on popular pages is often a blood sport and ripe in most organizations for a more structured, rational approach.)

Former Molecular founder and CEO Ralph Folz asked about display -- is that making a comeback?  Brian indicated the lack of performance and the lack of placement control through ad networks made that a highly negative experience.  He did note that they are now experimenting with participation in real-time-bidding through ad exchanges for inventory that ad networks make available, sometimes for time windows only a hundred milliseconds long.  Jeff reinforced the emergence of "RTB" and mentioned MIT Prof. Ed Crawley's Cambridge-based DataXu (which Flybridge has invested in) as a leader in the field.

Affiliate marketing came up next.  Tom explained the basics (in response to a question): each of the 600 stores in Mall Networks stable pays Mall Networks, say for example, a 10% commission on orders that come through Mall Networks.  Mall Networks gives a chunk to the members of various loyalty programs that shop through it -- say 3-5% of the value of the order; some goes to the loyalty programs themselves, as partial inducements for sending traffic to Mall Networks, and the rest goes to Mall Networks to cover costs and yield profits.

All the other panelists include affiliates in their marketing mix, and all appeared satisfied to have them play a healthy role.  Niraj specifically mentioned the ShareASale and Google Affiliate networks.  Jeff asked about everyone's frenemy Amazon; the answers were uniformly respectful: "they're a tough competitor, but they build general confidence and familiarity with the ecommerce channel, and that's good for everyone."  Niraj noted the 800 lb. gorilla nature of their category dominance: "They're at $20m and NewEgg is the next biggest pure play at $2B.  They're a fact of life. We just have to be better at what we focus on."

Someone in the audience raised email.  All of the panelists use it, with lists ranging from millions to hundreds of millions of recipients in size.  They noted that this traditional pillar of online marketing has now gotten very sophisticated.  In their world, they look well beyond top line metrics like open- and clickthrough rates to root-cause analysis of segment-based performance.  Re-targeting came up, and Niraj noted that for them, email and re-targeting weren't substitutes (as some have seen them) but in fact played complementary roles in their mix.  (Jeff explained re-targeting for the audience: using an ad network to cookie visitors to your site, and then serving them "please come back!" ads on other sites in the network they go to after they've abandoned a shopping cart or otherwise left your site.  A twist: serving ads inviting them to *your* site after they've abandoned one of your competitors' sites.  Hey, all's fair in love, war, and ecommerce...).  A common theme:  unlike most of the rest of the world, email teams at these leading firms are tightly integrated with other channels' operators to better integrate the overall experience, even to the point of shared metrics.

What about social?  Scott: "Building community is key for us.  We run contests -- "What are you hoping will be under your tree this Christmas?" -- to stimulate input from our customers.  And, while we have social media coordinators, many people here participate in channels like Twitter in support of our efforts."  Niraj: "Our PR team came up with a 'Living Room Rescue' contest which we did in partnership with [a popular] HGTV host [whose name escaped me -- C.B.].  We got six thousand entries; we used a panel of professional decorators to narrow the list to a hundred, and then used social voting to choose a winner.  We publicized the contest, and it took on a life of its own, as local papers tried to drum up support for their local [slobs -- my word, not Niraj's].  While we couldn't / didn't measure conversion directly from this campaign, our indirect assessment was that it had a great ROI."  Jeff observed that social's potential seems greater when the object of the buzz is newsworthy.

It was a short leap from this to a question about attribution analysis, the simultaneous-dream-and-nightmare-du-jour for web analytics geeks out there.  Brian was surprisingly dismissive.  In his experience (if I understood correctly), he's seeing only up to 20%, and usually only 5-10% of order-placing customers touch two or more properties they source clicks from, across the broad landscape they cover, across a time frame ranging from a day to a month long.  "In the end, only a couple of dollars would shift from one channel to another if we did attribution analysis, so in general it's not worth it."  We chatted briefly after the panel about this; there are large ticket, high-margin exceptions to this rule (cars).  I need to learn about this one some more, it surprised me.

Mobile!  Is it finally here?  Scott reports that 6-9 months ago *customers* finally began asking for it (as opposed to having it pushed by vendors), so now they have a Shoebuy.com iPhone app.  Jeff noted that customers are rolling their own mobile strategies -- some folks are now going into (say) Best Buy, having a look at products in the flesh, then checking Amazon for the items and buying them through their iPhone if the price is right.  So, your store is now Amazon's showroom.  If you can't find something, or didn't even know you wanted it, but happen to stray near a store carrying it, location-based services will push offers at you -- and the offers may come from competitors.  (Gratuitous told-you-so here.)  Niraj:  "Say you're in Home Depot.  You want a mailbox.  Their selection is 'limited' [his description was more colorful]. We have 300 to choose from.  Wouldn't you want to know that?" Jeff:  Soon we'll also see the death of the checkout line: you'll take a picture of the barcode on the object of your desire, your smartphone will tell the store's POS system about it, and the POS system will send back a digital receipt you can show someone (or in the future, something) on your way out of the store. 

With all these channels in use, I asked how often they make decisions to reallocate investments across (as opposed to within) them -- say from search to email, as opposed to from keyword to keyword.  Brian: "Every day, each morning.  Some things -- like affiliate relationships -- may take 3-4 days to unwind.  But the optimization is basically non-stop."  Later we talked about the parallels with Wall Street trading floors.  For him, the analogy is apt.  Effectively he's a market-maker, only the securities are clicks, not stocks.  It's now reflected in their recruiting: many recent hires are former Wall Street quants.

A final note: The cultures in these shops are intensely customer-focused, flat, and data-driven.  Scott reads *every one* of the hundreds of thousands (yes you read right) of customer survey responses Shoebuy gets each year.  He also described the enthusiasm with which their customer service team embraced having all company communications to customers end with an invitation to email senior management with any concerns.  Niraj described CSN's floor plan:  500 people, no offices.  Everyone in the company takes a regular turn in customer service.  Everyone has access to the firm's data warehouse.  Brian told us about a digital display they have up in their offices showing hour-by-hour, source-by-source performance.  They also recently ran a "Query Day" in which everyone in the company -- including sales, finance, HR -- got training in how to use their databases to answer business questions.  Tom described that they “watch the cash register every minute, hour, day during the Christmas shopping season.”

This was a terrific session, and I've only captured half of it here.  Further comments / corrections / observations very welcome.

Quiz Answers:

1. MIT Prof. David K. Gifford, Open Market, Cambridge

2. Tom Gerace, BeFree, Cambridge

3. Constant Contact, Waltham

What's NYT.com Worth To You, Part II

OK, with the response curve for my survey tailing off, I'm calling it.  Here, dear readers, is what you said (click on the image to enlarge it):

(First, stats: with ~40 responses -- there are fewer points because of some duplicate answers -- you can be 95% sure that answers from the rest of the ~20M people that read the NYT online would be +/- 16% from what's here.)

90% of respondents would pay at least $1/month, and several would pay as much as $10/month. And, folks are ready to start paying after only ~2 articles a day.  Pretty interesting!  More latent value than I would have guessed.  At the same time, it's also interesting to note that no one went as high as the $14 / month Amazon wants to deliver the Times on the Kindle. (I wonder how many Kindle NYT subs are also paper subs getting the Kindle as a freebie tossed in?)

Only a very few online publishers aiming at "the general public" will be able to charge for content on the web as we have known it, or through other newer channels.  Aside from highly-focused publishers whose readers can charge subscriptions to expense accounts, the rest of the world will scrape by on pennies from AdSense et al. 

But, you say, what about the Apple Tablet (announcement tomorrow! details yesterday), and certain publishers' plans for it?  I see several issues:

• First, there's the wrestling match to be had over who controls the customer relationship in Tabletmediaworld. 
• Second, I expect the rich, chocolatey content (see also this description of what's going in R&D at the Times) planned for this platform and others like it to be more expensive to produce than what we see on the web today, both because a) a greater proportion of it will be interactive (must be, to be worth paying for), but also because b) producing for multiple proprietary platforms will also drive costs up (see for example today's good article in Ad Age by Josh Bernoff on the "Splinternet"). 
• Third, driving content behind pay walls lowers traffic, and advertising dollars with it, raising the break-even point for subscription-based business models. 
• Fourth, last time I checked, the economy isn't so great. 

The most creative argument I've seen "for" so far is that pushing today's print readers/ subscribers to tablets will save so much in printing costs that it's almost worth giving readers tablets (well, Kindles anyway) for free -- yet another edition of the razor-and-blade strategy, in "green" wrapping perhaps.

The future of paid content is in filtering information and increasing its utility.  Media firms that deliver superior filtering and utility at fair prices will survive and thrive.  Among its innovations in visual displays of information (which though creative, I'd guess have a limited monetization impact) is evidence that the Times agrees with this, at least in part (from the article on Times R&D linked to above):

When Bilton swipes his Times key card, the screen pulls up a personalized version of the paper, his interests highlighted. He clicks a button, opens the kiosk door, and inside I see an ordinary office printer, which releases a physical printout with just the articles he wants. As it prints, a second copy is sent to his phone.

The futuristic kiosk may be a plaything, but it captures the essence of R&D’s vision, in which the New York Times is less a newspaper and more an informative virus—hopping from host to host, personalizing itself to any environment.

Aside from my curiosity about the answers to the survey questions themselves, I had another reason for doing this survey.  All the articles I saw on the Times' announcement that it would start charging had the usual free-text commenting going.  Sprinkled through the comments were occasional suggestions from readers about what they might pay, but it was virtually impossible to take any sort of quantified pulse on this issue in this format.  Following "structured collaboration" principles, I took five minutes to throw up the survey to make it easy to contribute and consume answers.  Hopefully I've made it easier for readers to filter / process the Times' announcement, and made the analysis useful as well -- for example, feel free to stick the chart in your business plan for a subscription-based online content business ;-)  If anyone can point me to other, larger, more rigorous surveys on the topic, I'd be much obliged.

The broader utility of structuring the data capture this way is perhaps greatest to media firms themselves:  indirectly for ad and content targeting value, and perhaps because once you have lots of simple databases like this, it becomes possible to weave more complex queries across them, and out of these queries, some interesting, original editorial possibilities.

Briefly considered, then rejected for its avarice and stupidity: personalized pricing offers to subscribe to the NYT online based on how you respond to the survey :-)

Postscript: via my friend Thomas Macauley, NY (Long Island) Newsday is up to 35 paid online subs.

What's NYT.com Worth To You?

Via Chris Schroeder's  (@cmsschroed) RT of Henry Blodget (@hblodget), the news of the NYT's decision to start charging (again) for content.

Blodget's prior analysis suggested this might be worth ~$100 million per year  (my deduction based on his math) to NYT Co.  If a tenth of its 130M monthly unique visitors end up being "heavy users" that pay, 4 bucks a month gets them ~$600 million annually (13m * $4 * 12 months = $624 million).  (Seems high; better data anyone?)

What's it worth to you?  See what some folks had to say in the chart below.  Please take this survey to add your perspective, and let your friends know about it:

Note: I removed one response of "1 million articles for free, willing to pay $0 thereafter" because it messed up the display, but am mentioning it here for full disclosure. And to the respondent, thank you for participating!

Postscript: conclusions and analysis

Looking for a Web Marketing Analyst

Force Five Partners is looking for a web marketing analyst.  See http://www.forcefivepartners.com/opportunities.html for more details.  Referrals appreciated!

@Chartbeat: Biofeedback For Your Web Presence

Via an introduction by my friend Perry Hewitt, I had a chance yesterday to learn more about Chartbeart, the real-time web analytics product, from its GM Tony Haile.

Chartbeat provides a tag-based tracking mechanism, dashboard, and API for understanding your site's users in real time.  So, you say, GA and others are only slightly lagged in their reporting.  What makes Chartbeat differentially useful?

I recently wrote a post titled "Fly-By-Wire Marketing" that reacted to an article in Wired on Demand Media's business model, and suggested a roadmap for firms interested in using analytics to automate web publishing processes. 

After listening to Tony (partly with "Fly-By-Wire Marketing" notions in mind), it occurred to me that perhaps the most interesting possibilities lay in tying a tool like Chartbeat into a web site's CMS, or more ambitiously into a firm's marketing automation / CRM platform, to adjust on the fly what's published / sent to users.

Have a look at their live dashboard demo, which tracks user interactions with Fred Wilson's blog, avc.com.  Here's a question: if you were Fred -- and Fred's readers -- how would avc.com evolve during the day if you (as Fred or one of Fred's readers) could see this information live on the site, perhaps via a widget that allowed you to toggle through different views?  Here are some ideas:

1. If I saw a disproportionate share of visitors coming through from a particular location, I might push stories tagged with that location to a "featured stories" section / widget, on the theory that local friends tell local friends, who might then visit direct to the home page url.

2. If I saw that a particular story was proving unusually popular, I might (as above) feature "related content", both on a home page and on the story page itself.

3. If I saw that traffic was being driven disproportionately by a particular keyword, I might try to wire a threshold / trigger into my AdWords account (or SEM generally) to boost spending on that keyword, and I might ask relevant friends for some link-love (though this obviously is slowed by how frequently search engines re-index you). 

(Note: pushing this further, as we discussed with Tony, we'd subscribe to a service that would give us a sense for how much of the total traffic being driven to Chartbeat users by that keyword is coming our way, and use that as a metric for optimizing our traffic-driving efforts in real time.  Of course such a service would have to anonymize competitor information, be further aggregated to protect privacy, and be offered on an opt-in basis, but could be valuable even at low opt-in rates, since what we're after is relative improvement indications, and not absolute shares.)

4. If you saw lots of traffic from a particular place, or keyword, or on a particular product, you might connect this information to your email marketing system and have it influence what goes out that day.  Or, you might adjust prices, or promotions, dynamically based on some of this information.

Some of you will wonder how these ideas relate to personalization, which is already a big if imperfectly implemented piece of many web publishers' and e-retailers' capabilities.  I say personalization is great for recognizing and adjusting to each of you, but not to all of you.  For example, pushing this further, I wonder about the potential for "analytics as content".  NYT's "most-emailed" list is a good example of this, albeit in a graphically unexciting form.  What if you had a widget that plotted visitors on a map (which exists today of course) but also color-coded them according to their source, momentarily flashing the site or keyword that referred them?  At minimum it would be entertaining, but it would also hold a mirror up to the site's users showing them who they are (their locations and interests), in a way that would reinforce the sense of community that the site may be trying to foster otherwise. 

Reminds me a bit of Spinvision, and by proxy of this old post. 

Fly-By-Wire Marketing

One of the big innovations used in the F-16 fighter jet was the "fly-by-wire" flight control system.  Instead of directly connecting the pilot's movements of the control stick and the rudder pedals to the aircraft's control surfaces through cables (for WWI-era biplanes) or hydraulics, the pilot's commands were now communicated electronically to an intermediate computer, which then interpreted those inputs and made appropriate adjustments. 

This saved a lot of weight, and channeling some of those weight savings into redundant control circuits made planes safer.  Taken to its extreme in planes like the B2 bomber, "fly-by-wire" made it possible for pilots to "fly" inherently unstable airplanes by leaving microsecond-by-microsecond adjustments to the intermediate computer, while the pilot (or autopilot) provided broader guidance about climbs, turns, and descents.

Now we have "fly-by-wire marketing".

A couple of days ago I read Daniel Roth's October 19 article on Wired.com titled "The Answer Factory: Fast, Disposable, and Profitable as Hell", describing Demand Media's algorithmic approach to deciding what content to commission and publish.  The article is a real eye-opener.  While we watch traditional publishers talk about turning "print dollars into digital dimes", Demand has built a $200 million annual revenue business with a $1 billion valuation.  How?  As Roth puts it, "Instead of trying to raise the market value of online content to match the cost of producing it — perhaps an impossible proposition — the secret is to cut costs until they match the market value."  More specifically,

Before Reese came up with his formula, Demand Media operated in the traditional way. Contributors suggested articles or videos they wanted to create. Editors, trained in the ways of search engine optimization, would approve or deny each while also coming up with their own ideas. The process worked fine. But once it was automated, every algorithm-generated piece of content produced 4.9 times the revenue of the human-created ideas. So Rosenblatt got rid of the editors. Suddenly, profit on each piece was 20 to 25 times what it had been. It turned out that gut instinct and experience were less effective at predicting what readers and viewers wanted — and worse for the company — than a formula.

I'm currently in situations where either the day-to-day optimization of the marketing process is too complex to manage fully through direct human intervention, or some of the optimizations to be performed are still sufficiently vague that we can only anticipate them at a broader, categorical level, from which a subsequent process -- perhaps an automated one -- will be necessary to fully realize them.  I also recently went to and blogged about a very provocative MITX panel on personalization, where a key insight (thanks to Scott Brinker, Co-founder and CTO of ion Interactive) was how the process to support personalization needs to change as you cut to finer and finer-grained targeting.  So it was with these contexts in mind that I read Roth's article, and the question it prompted for me was, "In a future dominated by digital channels, is there a generic roadmap for appropriate algorithmic abstractions of marketing optimization efforts that I can then adapt for (client-) specific situations?" 

That may sound a little out there, but Demand Media is further proof that "The future's already here, it's just not evenly distributed yet."  And, I'm not original in pointing out that we've had automated trading on Wall Street for a while; with the market for our attention becoming as digital as the markets for financial securities, this analogy is increasingly apt.

So here are some bare bones of what such a roadmap might look like.

Starting with end in mind, an ultimate destination might be that we could vary as many elements of the marketing mix as needed, as quickly as needed, for each customer (You laugh, but the holodeck isn't that far away...), where the end result of that effort would generate some positive marginal profit contribution. 

At the other end of the road, where we stand today, in most companies these optimization efforts are done mostly by hand.  We design and set campaigns into motion by hand, we use our eyes to read the results, and we make manual adjustments.

One step forward, we have mechanistic approaches.  We set up rules that say, "Read the incoming data; if you see this pattern, then make this adjustment."  More concretely, "When a site visitor with these cookies set in her browser arrives, serve her this content." This works fine as long as the patterns to be recognized, and the adjustments to be made, are few and relatively simple.  It's a lot of work to define the patterns to look for.  And, it can be lots of work to design, implement, and maintain a campaign, especially if it has lots of variants for different target segments and offers (even if you take a "modular" approach to building campaign elements).  Further, at this level, while what the customer experiences is automated, the adjustments to the approach are manual, based on human observation and interpretation of the results.

Two steps down the road, we have self-optimizing approaches where the results are fed back into the rule set automatically.  The Big Machine says,  "When we saw these patterns and executed these marketing activities, we saw these results; crunching a big statistical model / linear program suggests we should modify our marketing responses for these patterns in the following ways..."  At this level, the human intervention is about how to optimize -- not what factors to consider, but which tools to use to consider them.

I'm not clear yet about what's beyond that.  Maybe Skynet.  Or, maybe I get a Kurzweil-brand math co-processor implant, so I can keep up with the machines.

The next question you ask yourself is, "How far down this road does it makes sense for me to go, by when?"  Up until recently, I thought about this with the fairly simplistic idea that there are single curves that describe exponentially decreasing returns and exponentially increasing complexity.  The reality is that there are different relationships between complexity and returns at different points -- what my old boss George Bennett used to call "step-function" change.

For me, the practical question-within-a-question this raises is, for each of these "step-functions", is there an version of the algorithm that's only 20% as complex, that gets me 80% of the benefit?  My experience has been that the answer is usually "yes".  But even if that weren't the case, my approach in jumping into the uncharted territory of a "step-function" change in process, with new supporting technology and people roles, would be to start simple and see where that goes.

At minimum, given the "step-function" economics demonstrated by the Demand Medias of the world, I think senior marketing executives should be asking themselves, "What does the next 'step-function' look like?", and "What's the simplest version of it we should be exploring?" (Naturally, marketing efforts in different channels might proceed down this road at different paces, depending on a variety of factors, including the volume of business through that channel, the maturity of the technology involved, and the quality of the available data.  I've pushed the roadmap idea further to help organizations make decisions based on this richer set of considerations.)

So, what are your plans for Fly-By-Wire Marketing?

Postscript: Check out "The Value Of The New Machine", by Steve Smith in Mediapost's "Behavioral Insider" e-newsletter today.  Clearly things are well down the road -- or should be -- at most firms doing online display and search buys and campaigns.  Email's probably a good candidate for some algorithmic abstraction.

Personalization Is A Process #MITXMT

Last week I went to a MITX panel in Cambridge titled, "Get Relevant:  The Next Generation of Website Personalization".  I asked a question: "In between broadcast-to-the-masses and one-to-one, where in your practical experience is the crossover point (in number of customer segments you define and design for) where returns from finer-grained personalization are exceeded by the complexity of supporting an expanding number of them?" 

I got three good answers:

• Brett Zucker (CTO at Bridgeline Software) and Joe Henriques (Regional Director, Sitecore) manned up ;-) and each gave me a number for where they see many organizations today -- Brett said 6-10, and Joe put it at 10-20.  Gross average, but crudely useful nonetheless!
• Andrew Hally (VP Product Marketing and Strategy at Unica) offered that it really depends on the industry, and on the mass-customizability of the product and other elements of the marketing mix. In the credit card biz, for example, you see firms executing against thousands of segments continuously redefined through real-time testing, because it's really easy to change terms and assess response.
• Scott Brinker (Co-founder/CTO at ion Interactive) noted that there are several answers to the question, because the process changes with each order of magnitude of the number of segments you set up to support. Scott described how in his experience, the hardest jump to make is from a handful of segments, supported by a highly manual process for creating offers, to the next order of magnitude, which at minimum requires modularization of marketing mix elements so they can be re-combined easily.  Scott further noted that traditional, primary-research-based segmentation approaches are being replaced by emergent (my word, blame me) segmentations based on testing. 

(Gents, apologies if I misquoted you -- please let me know.)

There's another question embedded in all this, which is Personalization (segmentation) with respect to what?  In other words, you can usefully define different segments for different elements of the marketing mix.  Upscale leisure traveler and a business traveler may get the same luxury hotel room.  But, they may behave differently when it comes to price.  Price-sensitive travelers may have different preferences for using marketing channels to find the best price.  So, in theory, you could have a single segment for product (luxury buyer, for example), two for pricing ("service-focused vs. price-focused"), and yet another two for channel preferences (e.g., "online-dominant" vs. "offline-dominant").  This can make segmentation much more operationally relevant, but it then also puts a premium on coordinating the outputs of these segmentation efforts (for example, if someone's really price-sensitive, you may want to steer them toward lower-cost digital channels for conversion).

Finally, another take-away from all this is to match the technology to the degree and kind (e.g., implicit vs. explicit) of personalization you intend to design a process for.  There's no point in buying a "Hemi-size" personalization engine if you've got a "Yugo-size" gas tank for marketing mix execution.  On the other hand, when the technology for testing an order-of-magnitude jump in segments becomes affordable, maybe it's time to optimize for that capability and flex / redesign the execution capability?