Analytics and Decision-Making: The Limits of Analysis

I've recently had several customer support experiences that got me thinking about the role of analytics in decision-making – particularly with respect to the proper limits of any analytics or decision-making technique.

I was recently involved in a minor auto accident and the resulting experiences with my insurance provider – Geico – struck me as remarkably emblematic of a certain trend in customer service as provided by some large companies.

Every person I talked to was remarkably polite. They had obviously been well and carefully trained in maintaining a good attitude, being respectful, sounding helpful, etc. But in every case, and in almost every way, the policies and actions they were pawning off on me were awful, abusive, annoying and clearly calculated to maximize Geico’s interest regardless of my needs, interests or desires.

“Certainly, Mr. Angel. Let’s get that car towed. Let’s see, your accident occurred in Marin County. I’m very sorry. But it seems like we don’t have an approved facility there. Yes, yes, it is quite large. How about here in Berkeley. That’s across a bridge and an hour away? Oh my. Well give me another zip code. San Francisco? No – I’m afraid we don’t have an approved facility there either. Yes, you’re right, that’s very surprising. How about here in Berkeley? And by the way Mr. Angel, is there anything on this call that wouldn’t cause you to rate me 10 in customer satisfaction.”

And so on. The Geico experience just got worse and worse – like a nightmarish encounter with an extraordinarily polite rapist.

“Sorry while we hold a gun to your head Mr. Angel. This may hurt a little. Apologies. Now Mr. Angel, was there anything in this rape would cause you not to give Geico a 10?”

In the end, I’m not sure if Geico’s politeness or the policies ended up being more annoying – for the politeness adds an extra veneer of dishonesty to the process. I’d rather be held up by a gunman than cheated by a con artist – even though the loss be the same.

We all know how such policies get put in place. We’ve all had experience in how a veneer of politeness can – at least for a time – hide customer support policies that are designed not to help the customer but to fleece them.

I have always believed that such policies are mostly bad business in the long run. But it would take an almost religious conviction to believe that this always true and I have little doubt that in the short run such “customer service” is likely to be highly profitable.

I recall a situation with a very large internet company that gave away free trials but made it extraordinarily difficult to cancel them. They always had studies that “proved” the existing policy was more profitable than when they offered cancellation or even simply allowed it. Perhaps it was, though I have little doubt that it also contributed to the gradual demise of the service. I’m equally sure that somewhere in Geico an analyst has proven how much more profitable it is to abuse the customer than serve them.

So while part of what I’d like to say about this goes back to a point I’ve made repeatedly about optimization – if you don’t take account every aspect of your company’s goals then optimization can go disastrously wrong (like having people use your company as an example of rotten service in a blog) – I don’t believe that the issue can be entirely encompassed within optimization.

Just as it is wrong to suggest that crime does not pay (it may pay quite well), so too it would be wrong to believe that calculated bad service must always be bad business in the strict sense of profitability.

So what does this say about the role of analytics – what should an analyst say when numbers suggest that a policy clearly intended to be harmful to customer interests is potentially more profitable?

It’s not an easy question.

In the end, I think it is simply not the case that any policy which maximizes shareholder return is “on the table” and must be considered analytically. There are many prudential reasons why this should be so (brand, risk, customer loyalty, etc.) but we need not rely on these. Policies that are obviously predatory should never, in my opinion, be considered in the first place.

The role of analytics is to help decision-makers understand and choose between a range of options. It is, as well, our job to help set the table with those options. But it is not our job to consider or admit any and every legal policy - nor do I think we our bound to do so. The broader culture that we live in and, where a company is well-ordered, the culture of the company itself should both set many bounds on the type and nature of policies that may be acceptable.

Free markets are by no means a “state-of-nature” tending to making life “nasty, brutish and short.” They are a shared cultural enterprise that demands a high degree of discipline, shared commitment, and institutional maturity. The goal of business is to create wealth by providing value – not by suctioning it off like some rapacious vampire. Like any other human endeavor, business cannot be done without ethics. And ethics comes before analysis.

It takes courage for any business person, analyst or otherwise, to speak against practices based on the exploitation of the customer rather than the provision of value. It takes courage especially when we cannot rely on the comfort of our numbers. It is part and parcel of the analyst’s role to make sure a company understands the full range of impacts that operational decisions and policies may have. But our role – and the role of any business decision-maker or influencer – should not end there.

I wish I believed that every decision harmful to the interests of customers could always be shown to be bad. I wish I believed that every evil-doer will be commensurately punished and good intentions will always be rewarded. These things are true often enough to get by. But they are by no means universal. We cannot count on such certainty and we should not rely on it. There is more to life than making money – and there is more decision-making than optimization.