I’m building my first new product in over a decade. I started my career back in 1993 as a black box tester on Media 100. I joined the team just as the first engineering prototypes were coming in from manufacturing, well prior to shipping v1.0. After that I helped create Adobe ImageStyler 1.0 in 1998, and then Adobe LiveMotion 1.0 in 2000.
On the suggestion of a coworker I picked up “The Lean Startup” by Eric Reis which lead to a whole lot of book reading (see my other post here).
As my team and I began applying the tools such as Problem and Solution Interviews, mock web sites with A/B testing, and in product analytics, I had a growing sense of unease. A long time ago, in a place far, far, away, I studied chemistry, a fair amount of psychology, and statistics. I’ve also been deeply personally involved in the evidence based medicine movement and often find myself reading meta-analyses at the Cochrane Collaboration.
The tools from the LSM were generating a fair amount of structured data, some qualitative, and some quantitative. As my coworkers attempted to communicate their findings, they started committing data abuse. They would run a survey on an incredibly small population and attempt to assign meaning to differences that were beyond any reasonable threshold of noise. Or make a graph comparing data without appropriately normalizing it, rendering the results meaningless. Or score aspects of qualitative interview to allow quantitative comparison without controlling for the lack of rigor in how we conducted the interviews.
In “The Lean Startup”, Eric talks about Vanity Metrics. But he only looks at some small ways to bias your data so you can lie to yourself. As I continued to deepen my research through reading more books and talking to our researchers, I realized that there are many, many ways to lie to yourself via these methods.
I had to sit back and collect myself.
Vexed that these new tools I was so inspired by when I read “The Lean Startup” can be so easily abused.
I then realized that data driven decisions (aka The Quants) is a whole continuum of tradeoffs. On the quick and dirty side we have some of the tools I used 12 years ago: Informal customer chats, launch and pray, intuition. On the deep quant side you have extensive surveys, large scale A/B testing, structured and appropriately mediated customer interviews (using tools from the social sciences to avoid bias). For medical work you get all they way to double blind control studies as the gold standard.
But moving across this spectrum greatly changes the amount of time and cost associated with making a decision. In the business world, it can often be cheaper to fail (or make a wrong decision earlier) then to delay action until you have the quants figure it out.
That was the key part for me: Failure can be cheaper then learning through experiments.
Your job as a leader is to understand the risks and choose the level of rigor and science that is appropriate to the situation.
The Oakland A’s helped revolutionize baseball by adding in some quant tools to the player recruitment process. This gave them a huge first mover advantage. It didn’t take very long for the other teams to adopt the same tools, therefore rendering that lead negligible. But now just to be in the game you’d better be using those tools or you’ll be at a distinct disadvantage. (http://blogs.hbr.org/davenport/2011/09/six_things_your_company_has_in.html)
The Lean Startup Movement (LSM) tools introduce a little bit more quant (and therefore rigor, discipline, and skill) with the hope of reducing risk. But they are a tradeoff. They take a bunch of work. Work that could be spent elsewhere. You need to decide if you are in a game like baseball where all your peers are using these tools. If they are, you’d better dust off your copy of “The Lean Startup” or just go home.
But you also need to be realistic about what you’ll get from the tools. They are incredibly rough approximations of tools that have been used in the sciences for years. They are full of bias and inaccuracy, just (hopefully) slightly less then the techniques you were using before (and of course slightly better then your competitors). On the flip side, you may be making a decision that is so high risk that you want significantly more rigor then the LSM tools.
I strongly believe that managers who use Quants appropriately will kick ass in the long run. But it’s that ‘appropriately’ part that’s hard. As a community we are collecting, quantifying, and tabulating more data then ever.
Do you have the skills to choose the appropriate level of rigor for the decision at hand?
I didn’t used to. I’m getting better though. My next post will go into details as to how I buffed up my inner quant so I can select the appropriate tool for the task at hand.
Till next time,