Dan Goodman at Ogilvy — the interactive guru — made a statement today during a discussion about metrics and optimization that made me stop, think hard, and then ask him to back up.
“Don’t over-engineer things,” he said.
As the interactive marketing guy for the company that makes the best-engineered PCs, who is often the only non-engineer in most meetings; the guy who only passed algebra II because the teacher was his coach, who sat at a dinner last night in NYC and listened to the inimitable Uncle Fester from this blog’s comments tell my eldest son that his father “was a serious math retard …”
For me to hear an A-team interactive marketing guy tell me not to be over-weening in online advertising, behavioral targeting, multivariate testing, A|B analysis, demographic segmentation, continuous improvement`cycles, dashboards, NPV, E-to-R ….
Well, it made my day to hear someone throw a bucket of cold water and basically say, there’s so far you can take it.
The maddening thing about web marketing is it represents a collision of the logical precision of information technology with the creative chaos of media. Maddening because in theory one should be able to measure and improve with a high degree of precision. But in practice no one has enough, time, money or talent to get to the Valhalla vision that in theory, you know is possible.
As T.S. Eliot wrote in The Hollow Men:
Between the idea
And the reality
Between the motion
And the act
Falls the Shadow
That’s what keeps me grinding away in the online medium — the idea, the promise, the illusion that if you do it all just right, it will all start to swing along on its own. Integrating metrics with a content management system and an ad server, building a neat little closed system that sort of self-optimizes …..
The reality is fouled up insertion orders, weird metric tagging, site errors, and that big unknown … user behavior. Yet, I suspect every online operator — from the service providers, the agencies, the publishers, the bloggers, the vloggers, the podcasters — will, if caught in the right optimistic mood, express an idealistic hope that online media is the most perfectable medium ever known.
The “Paralysis of Analysis” as my pastor used to say, huh?
People get more satisfaction and happiness from the anticipation of a purchase than from taking ownership of the item itself.
Most of us drones and Luddites use the P.C. for emails, some online shopping, (many of my friends do online banking, taxes, and bill paying), reading news and blogs, posting opinions, and in my case, use a VPN connection to the mainframe where I do my work. A large segment of current users are in the 1990s where technology is concerned. The future lies with the young who can’t envision an unconnected life. For them consumerism must be interactive and visionary marketers sell to this generation. If you do not over-engineer for them, they will gravitate to the market forces who will.
David,
I’m going to go out on a limb so far as to probably be wrong….
It seems to me that a lot of metrics observe correlations, and people assume them to be cause and effect and try to manage to it. There are often more unseen and unmeasured variables that account for the behaviors than the ones that are being meseasured, no matter how many you put in. That’s not to say, don’t try, likely you may have more accuracy in predicting the macro movements than the micro ones.
In my opinion, too many companies take an end objective like revenue / profit and try to work backward to figure out what to do to get it. Instead, they should seek real cause and effect. Customers have a need – fill it, and do it well. The rest takes care of itself.
Statistical analysis only gets us so far, as I think all the other commenters have said. When we start looking for the meaning in things, we tend to find what we seek.
It’s like the prognosticators during a sporting event. “No left handed red headed son of a garbage man has ever hit a homer in the bottom of the 3rd inning of the 4th game of the series before.” The problem is, that the explanation of the numbers implies that it can’t ever happen. The truth is, that outcome has just not occurred yet.
But what do I know…I barely made it out of Algebra I.
The law of diminishing returns and ROI are linked.
Unfortunately, many consumer products are manufactured with the ROI conclusion that trying to improve quality out of the factory is more expensive than dealing with X% returns and DOAs.
Some things *should* be over engineered. Airplanes, buildings, etc.
One of the most famous cost benefit analysis cases gone awry is, of course, the Ford Pinto and that exploding gas tank that Ford knew about for 8 years.
“Don’t over engineer it” is a simplified concept that means nothing without context. If you can spend 40 labor-hours and tweak a metric that improves your bottom line, your knowledge, or your happiness in some meaningful way, then do it. Perfection is impossible but it is in the striving for it that we achieve greatness.
Ford’s cost benefit analysis on the Pinto is the shining example or there is my experience with Linksys routers (out of 50, maybe 10 were DOA). It may be cheaper for Linksys to deal with returns and ship a new box than improve QA but it hurts their image. Considering most manufacturing ROI doesn’t include the cost of marketing (not sales), the hit to the company’s image isn’t captured when doing the calculations. Six sigma, blah, blah
And yes, you’re a math frigtard and your eldest has heard me say it before. But you have other qualities that make up for it. Like your patience in dealing with trolls…
The meaning of a metric is a matter of its position and relation to the signified and the goal. Predictive stats has amazingly relevance to the online world. You just need talent who know how to do wield it. Not “math frigtards” or “little picture” people.
And it’s all in the tools we choose to use, the poison we drink. Page tags aren’t necessary for web analytics, AB, MVT, or any other testing (Unica, SiteSpect). Innovation happens elsewhere with open source and a global strategy.
Dan’s advice is a warning not a rule.