Selling is HardPosted: December 25, 2011
Sales and Science may both start with the same letter but that’s the only thing they share. I am a scientist by training and every time I would meet a salesman whatever is that he was selling (and it’s always a he) it would just reinforce me how much I absolutely despise selling. Selling often involves lying, having no understanding of what you are selling, being willing to screw all of your co-workers so that you can “finish the kill”, not comprehending what engineers and product managers are really doing and so on.
Here I just want to reflect on my personal experiences with selling and say this. Selling is hard. If salespeople are lying bastards, scientists and engineers are naive. We think because we understand equations and are more analytical, everything else will fall in its place. It rarely does. I have seen a number of sales pitches and presentations that involved data mining and analytical solutions that were full of esoteric details that made me wonder if the person presented had any understanding of who he was talking to. The marketing guy of an e-shop or of retail banking does not necessarily know or cares what support vector machines are and is not really aware what precision and recall are.
So armed with better understanding I vowed to avoid these common pitfalls. I would always have the target audience in mind and I will always try to focus on the value (== money) that these solutions bring. One of the products I spent at least 4 full months creating and also paid other people out of my personal savings to aid me, was an analysis of the Groupon-like deal sites. The analysis was based on a few thousand deals and each deal was annotated with a number of attributes such as the degree of discount offered, category of deal, location of deal and so on. I had more that 40 attributes that characterized the deal both in absolute terms as well as in relative terms to previous deals.
My product that I set out to sell was a set of easy-to-understand and concrete recommendations on how to increase sales for a deal site. I had a list of 25 recommendations that according to the underlying models, if all applied, would bring a 20% revenue increase for a deal site. The recommendations were in the form of “Increase your discount by 10% and your price by 15%” or “For restaurants never block Fridays”. I steered away from explaining what models I used, or how I measured success of the models since these were the mistakes that the practitioners of the field usually do. (I thought) I was focusing on value. I tried to price the study in a similar way. If there is a 20% increase in revenue then this would mean that I can price it at X. I sent emails and called the 15 biggest sites in Greece, and their response was basically the following
“I know better than any analysis.”
Why they knew better? Because they said so. It was a surprise to me that everyone I talked was basically trying to form a rule by looking at exceptions. They would look at an over-performing or under-performing deal and would try to make a rule out of this one case. If it happened that there was another deal with seemingly same characteristics then the rule was like one of the Ten Commandments. Until the next exception broke the rule and a new set of Ten Commandments was formed. This in statistics is called over-fitting and it is taught in Statistics 101, but I just couldn’t convince anyone that what they did was wrong.
In any case, I got a hard lesson from this. It is not easy to sell. Not at all. And yes people are unfair and naive and feel threatened by many things around them because if this new fancy thing called data mining can decide what deal to put next then all of a sudden he can’t be Very Important Person. But all these don’t matter. Because I couldn’t sell and because it was my mistake that believing that if I had a set of data-supported recommendations that people would just buy them. It is much more complicated than that but also much more simple than that. Because it has to do with people not with numbers. And that’s a world where models don’t apply.