Analytics over lunch

Analytics conjures images of mathematicians crunching data with computers. It’s not an altogether inaccurate stereotype, but in reality good analytics is a far more nuanced discipline. In fact, the power of analytics can at times be realized without performing any analytics at all.

This point was yet again driven home to me during a recent lunch meeting with potential clients. Arriving early I noticed another man in the lobby who looked as if he might be there for the meeting. Sure enough, when I introduced myself, he was. His name, for the purposes of this column, was John. John was cordial, but not nearly as friendly as the individual whom I’d spoken with on the phone when setting up the meeting. I would discover why later, but for the moment I simply picked up that the meeting wasn’t his idea and that he’d rather have been elsewhere. Fair enough.

The four others who joined us displayed the congeniality more typical of a business meeting. When everyone arrived, we were seated at a table, chatted, ordered and then set to work. I had very little information about the actual project, so I listened.

The company, which had buildings located in a geographic area covering a few square miles, operated a small fleet of mail delivery trucks. Letters and parcels arrive at a main location where they are sorted, assigned to trucks and routes and delivered to various mail stops. It sounded like a relatively straightforward vehicle routing problem, and my queries only reinforced my initial impression. Deliveries were completed early in the day, suggesting considerable excess time and therefore room for improvement. On top of everything else, the problem was small, involving fewer than a dozen trucks, and a fixed weekly schedule was considered adequate – no need to worry about dynamic routing. So when asked if analytics could be used to develop more efficient schedules, schedules that required less time or fuel or trucks, my answer was yes. But I was interested in hearing more of the complicating details – details that could only be brought to light by someone who actually dealt with the mail on a daily basis. That person was John.

Not surprisingly by that point, John took a defensive tone. When asked why drivers finished early and why certain trucks were rarely used, anecdotes of instances where people needed deliveries “right away” came pouring forth. John asked how a mathematical model could possibly deal with these special cases. I could have replied that such requests can be modeled with probabilities, but that wouldn’t have satisfied John, and it really wasn’t the right answer. As it happened, I didn’t need to answer, since others at the table jumped on the more conspicuous response: what was so important about these deliveries that they had to be immediately attended to? John’s answer wasn’t clear.

As the conversation continued I remained on the sidelines. The more senior individuals were extremely polite as they pressed John, but the dynamics were such that he couldn’t help but feel ganged up on. During the course of the discussion I was asked to recommend potential courses of action. I mentioned off-the-shelf software and modeling by a consultant, but suggested that the time and expenditure probably weren’t worth it for such a relatively small problem.

As the conversation wound down it wasn’t clear what the next step would be. Frequently such meetings conclude with people going away to think about things, and either a decision is made to bring in a consultant or the problem quietly slips into the background. But this time that wasn’t the case. “John,” said the most senior individual at the table, “can’t you figure out a way to reduce the fleet by a couple of trucks?” After a brief pause, John responded, “Yeah, I think I can do that.” I couldn’t have been happier.

Analytics is about the use of analytical tools and data to solve problems, but that doesn’t necessarily require sophisticated mathematics and computers. When faced with analytics, John realized he could, and really should, do a more efficient job. And in this particular instance, I suspect John could have solved the problem at least as well as an optimization model – an optimization model that incorporated the many constraints John would have imposed on it. The more important issue was to get John to face his self-imposed constraints. Do I have to drop everything and make an immediate delivery just because I’m asked to? Do I need extra trucks and drivers sitting around for such occasions? Can I deal with letting people go that I really don’t need? Analytics made him face these difficult questions.

As an analytics professional, I’m drawn to the technical aspects of the discipline. But it’s always important to keep the final goal in mind, and that goal isn’t data and mathematics. The goal is making better, fact-based decisions. If that can be accomplished over lunch without so much as a pencil and paper, why not?