Friday, December 12, 2008

The metrics system

Greg left a comment on CEO pay - Part 2, my Thursday post, and I will include it here in full:
Androcass: I agree that CEO pay is often outrageous, especially for struggling companies. Nevertheless, care needs to be taken when linking CEO pay with the stock price. Otherwise, you have a situation where the CEO has a huge incentive to manipulate the stock price. This is what led to the disasters at Enron, Global Crossing, MCI/Worldcom et al. I have little confidence that Sarbanes-Oxley will help, either.

Furthermore, how many times does stock price really reflect the market and not the individual company? A CEO shouldn't be rewarded just for being CEO in a bull market.

Other metrics would be better - profitability, market share, customer satisfaction. Unfortunately, many of those metrics are much harder to track, especially for the board of directors.
Ah, yes, metrics. By the way, I agree with everything Greg says here about manipulation and stock price and other metrics being harder to track, but let me ramble on about this topic a little more.

When I began working a real job for a bank (as opposed to my fake job as a substitute teacher for a year), I was a business guy in a programming environment. A short while after starting I was asked to look at measures of programmer productivity - I'm not sure exactly what the objective was, but, after moving into programming a few years later, I'm guessing it was something that no one would have liked very much. Back in April, I wrote this:
There have been studies on productivity over the years, but, strangely enough, the gold standard for programming productivity experiments was done 30 years ago, in IBM's Santa Teresa Laboratory. Gerald McCue's paper remains a primary reference. Without getting into the specific results, so quaint do they seem now (private offices, sufficient storage room, etc.), what is so striking is that a major corporation actually created a whole lab to study this issue. There was a sense, back in 1978, that making programmers as productive as possible was a worthy objective.
One of the other things I came across was that lines of code written was considered the only real metric one could measure for a programmer, and that it was widely considered a hopelessly flawed one. The reason, of course, is that programmers/developers are not hired to write code, they're hired to implement functionality.

There have been attempts to define different measures for software, though not expressly for the purpose of measuring productivity. The most well-known is function points, but this is hard to come up with in even close to a precise way, and I have never seen a real project use it.

The problem is that something simplistic like lines of code is too easy to manipulate; most programming languages offer several ways to solve almost every problem, and some solutions are a lot longer than others. Being verbose may give the developer a higher performance rating, but, in almost every other way, including subsequent maintenance, it is something of a disaster for the project.

Any time you measure someone by something that can be manipulated, you expose yourself to the risk that the manipulation will occur, and that the person will spend more time figuring out how to game the system than in actually being productive. (One is reminded of the famous Dilbert strip where a programmer is "gonna write me a new minivan this afternoon.") This is not just something that happens in programming, it's universal.

The slogan promulgated by consultants is, "You can't improve what you don't measure." This has a fair equivalence to, "Only do things that are measured." In either formulation, it is utter nonsense, and no way to run any enterprise. I recently had an animal control complaint, and the police were not particularly interested - want to bet that their attention would have been more sharply focused if they had some end-of-year metric that covered such issues?

[Note that I'm not saying that collecting metrics is useless, just that they need to be well-thought out and planned before implementation. Otherwise, perverse incentives are created as people, in effect, "teach to the test." I've worked in software development a long time, and I don't know what could measure productivity in a real world situation; it could well be that managers will simply have to manage, and evaluation is part of that job function.]

Which brings us to CEOs, and Greg's point that share price is a pretty poor metric, largely because of skewed incentives. And this is true, but his replacements aren't a whole lot better. They are harder to measure, they also suffer from context effects, and they can be manipulated almost as easily.

Ask yourself what you want from a CEO of a company in which you've invested, or for which you work. Assuming you're the average investor, or even a passive one through a mutual fund, you're probably not so much interested in next quarter's profitability as you are in long-term profitability. How do we generate that? Through prudent investments that create value over the long haul. In a technology company, you might like the CEO to build R & D capability, support the acquisition of patents, and so forth.

But the median CEO tenure is only 5.5 years (I can't recall where I read that, but I believe it's about right), so why would he or she commit to an incentive that might not pay off for 10 or 20 years? That's in the best interest of the company, but means nothing to the short-timer.

[I thought of putting a long riff here about why share price is seen as the most important metric, how modern theory contends that the price accurately reflects company value, so value added can only be seen through that price, but that's a big topic - needless to say, that flawed theory creates a lot of the problem.]

It's interesting that the explosion in CEO pay has coincided with variable compensation schemes, which were intended to give executives a greater incentive to improve the fortunes of the company. What has happened is that CEOs, aided by compliant boards, have gamed the system to give themselves riches despite their performance; I've said often that the most creative thing some CEOs accomplish is their employment contracts (and that's often the result of sharp work by compensation consultants).

Unless we find better ways to evaluate how a company is really doing, not just today or next week or next quarter, but 10 or 30 or 50 years from now, perhaps we should back off the assumption that variable compensation has positive effects - maybe it's time to go back to flat, well-publicized salaries.


Greg Glockner said...

I believe it was Peter Drucker who said that "you can't manage what you don't measure". I agree that many metrics can be manipulated - especially stock performance. But I don't share your pessimism - I do believe there are some metrics whose performance does track with business success.

Unfortunately, these metrics are typically much harder to measure, so most people take the easy route and use the simple old metrics.

Consider the traffic police: they typically ticket drivers for speeding because it's what they can measure easily. What they really need to patrol is reckless driving, but reckless behaviors like tailgating or weaving are hard to measure, so they ticket drivers for speeding instead.

Androcass said...

I wouldn't be so pessimistic if I had seen metrics work; admittedly, I haven't worked in a manufacturing setting where they can be of use.

But, in the "softer" areas of the company, they seem to cause more trouble (that is, cost more and tie up more time) than more direct ways of managing. Many of us have experienced the downside of metric-based help lines, for example.

I guess there is no perfect system. Over-reliance on metrics can lead to results as pernicious as complete subjectivity; as always, tools are only as good as the people who use them.

Clicky Web Analytics