- July 25, 2013
Thomas Bayes, Save Me From Statistics!
The only class I’ve ever failed was Statistics. I mean I flat-out failed that class.
Since then, I’ve been statistically impaired. I’ve never argued with the value of crunching collected data, but I’ve always struggled to see the value in statistically-predictive analysis. In my role in sales, however, I’m very interested in understanding the behavior and the alchemy behind our sales process. Many of our leads come in through our online planner, but discerning how they arrived at the gates of our form has always been a mystery.
As I gained experience, I noticed trends in Happy Cog’s sales. As I met and got to know other design shops and their owners, I realized those trends were universal, broader, bigger than just Happy Cog-related, and more meaningful. I thought it was remarkable that since January of last year, the number of our inbound leads tripled, but lo and behold this happened for at least four other shops in our network, over the same exact time period. Insights that I thought were unique to Happy Cog’s business were eerily applicable to our peers. Statistical analysis seemed more like the bitter pill I’d have to swallow to get to the bottom of this data and find out what it all meant. Ugh.
The good news is, Nate Silver rode in on a winged charger of baseball stats and electoral college domination with his book, The Signal and the Noise, to help me understand these trends. The bad news is, we just don’t have a sample size large enough to pull off some super sweet Freakanomics-style statistical analysis or prediction. Even if I pooled our sales data with our competitors’ (which I wouldn’t), we still wouldn’t offer the numeric bounty that Major League Baseball, or the U.S. Census, or the National Weather Service offer. We fielded over 400 project inquiries last year, and we’re on pace to shatter that number this year, but this is still a raindrop of data, not the monsoon necessary.
With a lack of data and a tendency to mistake coincidence for pattern, I was afraid I was in big trouble. I was trying to force patterns, rather than letting the data surface insights. Silver explains that we’re hardwired, from our primordial past, to seek out patterns in order to survive. We transpose these broad observations on our surroundings instinctually, in order to realize that that peripheral movement behind a tree is a stalking sabertooth. That’s all noise, though. The trick is finding the signal (if you can).
Enter Thomas Bayes. Silver is an adherent to a Bayesian probabilistic approach, which values raw data, but doesn’t submit to the numbers entirely. Think of it as number crunching—only improved and informed by expertise. The point is not to helplessly forfeit yourself to big data. The point is to impose structure on information, merging the quantitative insights of your data with the qualitative insight only human experience can provide.
For example, the raw numbers would lead me to believe that a majority of our clients reach out to us in the first half of the year, with well more than 50% of our planners coming in before the end of June. If I’m debating a week off in the summer, I might look to the end of the summer when I’ll theoretically be less busy. The truth of the matter is that while the majority of our planners arrived before June 30th last year, the projects we were really excited about, the ones we wound up actually pursuing, landed on our doorstep in July and August, after the new fiscal year.
It takes human insight to distinguish the promising, prospective projects from the less desirable options and to impose this additional layer on our data. Not all leads are created equal. If they were, we’d be working with that Florida local news anchor who wanted to build his version of “Facebook meets Craigslist!” during our weekends and offer us equity rather than actual legal tender.
My predictive instincts, honed over three years on the front lines of Happy Cog’s online project planner, are as valuable a tool as the data we might squeeze out of Highrise, or Capsule, or some other 30%-inadequate, overly-data-visualized CRM tool. Like our work, our industry is ever-evolving. What was a reliable pattern for 18 months is no longer the case a year later. An economic downturn like that in 2008 shot many patterns right in the boohine, and the ripples last for a long time. There is no Holy Grail; there is no roadmap. Your suspicions will never be 100% born out in data analysis. But, that doesn’t make them invalid.
Trust your expertise and instincts. Watch the numbers, and see if they tell you anything, then filter their message through your experience. Bringing in additional opinions is extremely valuable, too. Again and again, I’m surprised by how another pair of eyeballs looking at the same data I’ve reviewed sees something very different. No one understands your business the way the people in your business do. Partnered with some smart data collection, all that knowledge becomes a very powerful prediction engine. If you tip too far in one direction, you’re in real danger of following false leads, but if you can find the right balance of data analysis and creative reasoning, well, you’ve just made your sales approach a hell of a lot smarter.