Guardrail metrics

Guardrail metrics (sometimes called diagnostic metrics) help keep a process on track by detecting unintended impacts, unwanted side-effects and avoiding short-term solutions that might cause long-term damage.

For example, a company may want to increase revenue by 10%, but at the same time not increasing operational costs by more than 10%. The revenue is a target metric, the operational costs are a guardrail metric.

A key aspect for setting effective guardrail metrics is that they should be quick, easy and relatively cheap to measure. For aspects that can only be detected after a long delay, or at huge cost, look for indicators or proxy values. For example, for a product currently trying to optimise market-share, profitability might seem like a good guardrail metric (we don’t want to grow at the expense of bankrupting the company). However, if it takes a few months to turn new users into customers, waiting for actual profit is not useful as a guardrail metric. Indicators which could signal that users are likely to convert to paid customers are better as guardrail metrics (for example, frequency of usage). In such cases, it’s perfectly fine to add several cheaper and faster indicators to guardrail metrics. Unlike target metrics, where it’s important to focus on one or two important key measurements, there are typically lots of guardrail metrics protecting against different types of adverse effects.

The same measurements could be at different times targets or guardrail metrics, depending on the current focus. Targets are the things you want to improve, fix or optimize. Guardrail metrics are there to ensure that the general direction of work is consistent with your long-term goals, and to prevent runaway processes from causing too much damage before they are stopped. You should never aim to optimize a guardrail metric, just to keep it in an allowable range.

---