Optimize for the Global Max (not local max)
Cautionary tale for setting your goals
Overview
What is strategy? It is about choosing to do an action over another action to maximize results. What drives you to choose an action over another? There are many answers, but the one I’ll focus on today is a goal, an incentive, a metric.
Why do we have metrics? Measure performance. Pretty simple right? That was my thinking on the topic when I was early in my career. The reality is that metrics are complicated. Set the wrong metric and you won’t get the desired outcomes. Metrics are a key part of the OODA loop, in that they are the observations and often form the framework around a decision. The companies with fast OODA loops can outmaneuver competition and the market.
Take the example of a system alert. On one of my products last year we had alerts for the engineering team if a metric was out of the threshold. This helped us to quickly respond and address the issue. Unfortunately the lower bound didn’t have a similar alert and when the system stopped processing a specific request the alerts didn’t trigger. Fortunately we had other metrics with more lag that triggered us to address the issue.
Here are a few considerations when setting metrics:
- Global max vs. local max — Do you want to increase conversion on the page or full-funnel conversion? Full-funnel right? It depends. Do you have the structure and telemetry to look full funnel? Often you can optimize a local max faster and thus if there are large improvements is a good place to start.
- Speed of decision making vs. precision — How accurate and precise do you need to be? In a previous post I described the OODA loop. One way to increase decision making speed is through getting signal on your metrics in a timely manner. I would argue that “directionally correct” and consistent is better than high levels of precision.
- Leading vs. lagging — Let’s say you wanted to maximize the lifetime value of a customer. This is a great metric to monitor, but if you are making decisions on this metric you may have to wait months or years to see the full results. Thus you may look at another metric tied to engagement.
- Shared metrics vs. individual metrics — Imagine a sports team where each individual was more concerned with their goals than that of the team. What if individuals on a rowing team were more interested in maintaining the highest stroke rate than being the first team to cross the finish during a race.
Designing A Metrics System
My professor of Managerial Accounting at UC-Berkeley started his discussion on incentives with this quote:
“One of the fundamental principles / assumptions of economics is that individuals act in their self-interest to maximize their utility.”
If you start with this assumption then you’ll want to make sure that you are designing metrics that if achieved accomplish the desired outcome. Unfortunately though it isn’t that simple, as you also need to take a system view to evaluate the unintended consequences.
Take for example an ecommerce company that wants to increase transaction volume. It is possible they have a goal that is something like Monthly Unique Users (MUU). This is not a bad metric, but given that you don’t get paid for an MUU the metric could be achieved and the company could fail to achieve their other goals.
During my consulting days we started with first laying out the strategic themes for a company. For each theme we defined value drivers and performance indicators.
To setup the organization to achieve the Global Max we started by looking top down on the organization. Starting top down ensures you are thinking about the Global Max. More often than not I find that bottoms up metric definitions result in Local Max situations given that it is harder to bubble up a metric than cascade a metric down through an organization.
Using the above model we then laid out a set of core company metrics that we cascaded through the organization. Below is an illustrative example of the metric architecture we used. While you could stop with Profitable Growth and define metrics there, we found it useful to get specific for each team to have a goal that would contribute directly to the above.
There is a debate to be had on single vs. multiple metrics for a team. I have seen both in action and work/not-work at the same time. Generally I feel that you have a primary metric and a set of forensic metrics that go along with that core metric. For example, your core metric might be ecommerce purchases. The forensic metrics with this would be the formula that delivers this result, like [# of visitors] x [conversion to cart] x [conversion to purchase] x [purchase size in $]. Even from here you could break down [# of visitors] into Daily Active Users (DAU), Monthly Active Users (MAU), Monthly Unique Users (MUU), and many more.
Local Max Example
Windows Vista is an example of a product that failed to meet the market needs due to a local max problem. My take based on research and personal experience is that the teams were setup to optimize their part of the code base to deliver the best experience (a couple retros from various Microsoft employees like this one and this one). This on its own does not mean a product will fail, but due in part to the complexity of the solution there wasn’t an easy way to make tradeoffs on features. The end result was Vista was a worse product than its predecessor XP (write up here and another one) and many individuals didn’t make the upgrade or looked to Apple.
Another example that I saw a lot in my consulting days was that the goals for an individual (mostly promotion and raises) were not always aligned with making the company stronger. For example, if I sold a project where the revenues were from another part of the company I would only get full credit for the revenues that were from my part of the company. Thus there was no incentive for me to want to look at broader solutions, instead I would focus on my local max.
Global Max Example
Apple is a great example of solving the Global Max. They have setup the organization in a way that they are driving for the best integrated product. Apple hasn’t been first to market, but because of their Global Max mindset they create the best products, capture the most market value and wow customers. The iPhone is a great example of solving for the Global Max. They built killer hardware, the OS is designed for the hardware that is being used (unlike Android phones), and the App Store design of a marketplace ensures 3rd parties are continuing to develop solutions that make the phone even more valuable. Since Apple launched the iPhone in 2007 their stock is up 14x (vs. the S&P which is up 2x).
Another example that I have seen personally is from my experience at Porch. As we were undergoing a transformation the leadership team laid out a shared milestone plan for all employees. With this plan the team would achieve a bonus every time a milestone was hit over a 2-year time horizon. This included financial and business metrics, but also employee retention. The main bonus was more options, but a secondary bonus was an extension on the exercise period for the options. This plan helped to catalyze the teams energy around a common goal and pursue the actions that were most likely to deliver the end results.
So What?
You will face a point in your career where you are setting goals. You can choose to set goals that optimize for Global or Local max. Hopefully the above points to why you should evaluate what challenges you are looking to solve and the options you have to solve them. Setting out a great metric framework can strengthen your decision making (OODA loop) and position you / your company for success.
When I was in consulting the first question I would ask is to understand their incentive system. Based on their metrics I could accurately hypothesize 80% of the problems they were facing and the root causes. The reason is that many companies are not willing to make tough choices to setup their companies to be successful, but unfortunately then have to address these unintended consequences later.
In Closing
Designing a strong dynamic system starts with putting in place the metrics that drive the desired behavior. Make it easy for your employees to do the right thing and hard to do the wrong thing. The toughest thing for leaders to do is to say no and make hard choices. If you don’t do this you are essentially creating organizational debt that you’ll have to face everyday. Finally, while you here a lot about SMART objectives, to me the key is to evaluate if there are potential unintended consequences from the design.
Don’t shortcut incentive setting. This is a critical action for your company. It can free up the teams to make decisions faster and build products and companies that can challenge the best in the world.
References
Management Accounting is a great course to learn these principles
