Over the past months, I read and heard various things around the overused analytics terms: metrics and Key Performance Indicators (KPIs).
What is a metric? What is a KPI? And what are the differences between both? This is what I want to talk about today.
A metric is actually the result of a process. To highlight my point, I'll take the non-Business example of temperature to explain the different stages needed to build a metric.
The thermometer was not a single invention, but a development. 
First, you have the observation step. Very early, Humans were able to notice temperature differences / changes. "It's hot" or "It's cold". Later on, some Greek philosophers were first able to (very) broadly quantify and compare temperature. They built instruments (thermometers) that would translate the temperature into a numerical value they could use. Finally, in the first decades of the 18th century, Daniel Fahrenheit invented a thermometer able to precisely measure temperature. This measure associated to the eponym Fahrenheit scale defined the temperature as a metric - that will then be wildly used and mostly standardized.
According to the original mathematical definition, a metric is a function that represents a distance between 2 points. A metric can then be the comparison of a measure to a reference but it can be pretty much any combination of multiple measures. When you define a Business metric, those Observation, Quantification and Metric definition steps always happen. Most of the time without you noticing (or without you caring).
Let's take the Monthly Recurring Revenue example.
Observation: You see existing and new customers paying for the month but also some customers churning out from your product.
Measure: You quantify the dollars for each of these categories giving you 3 measures: new monthly revenue, existing monthly revenue, churned monthly revenue.
Metric: You then define Monthly Recurring Revenue (MRR) as a metric with:
MRR = existing monthly revenue + new monthly revenue - churned monthly revenue
In your day to day, you might not always need to mentally go through all those various steps. What is I believe important to remember is that "a metric is always the result of an observation" which is a derivative from "a metric should always have Business meaning".
Key Performance Indicators (KPIs) are the bread and butter of Data Analysts. Very often, it's even listed as priority #1 in job offers... Not sure I agree but it's still what I see.
KPIs are supposed to determine how a company's doing and how it can improve. You can build thousands of metrics and yet you should only have a handful of KPIs to help you pilot your company. If you have too many of them, it will be hard to make sure the truth resurface. Some called them the "North star metrics", I prefer the simple KPIs' mention. While there are many things to take into account, I'll list some important characteristics of the metrics you'll pick as KPIs.
A good KPI should be measurable. As a KPI is a subset of metrics, it stands to reason that it should be measurable. So I will refine and say that a KPI should be easily measurable, and that the measure should be consistant over time!
A good KPI should be objective. A KPI should help your company understand when they're doing good or not. If the KPI is not objective, then it won't be trusted, and worst, it will be torn apart to support different/opposite rationales.
A good KPI should provide evidence of progress against a goal. This is one of the main differentiator compared to a simple metric: all KPIs should have variation expectations for the coming weeks/months...
A good KPI should be transparent and easy to grasp. This is not so much about the metric itself than it is about management. Top-level KPIs should be crystal clear so everyone can understand where the company is and where it is going so decisions are easier to make.
A good KPI should be balanced between leading or lagging metrics. Leading ? Lagging?
Over the past couple of years, governments defined various KPIs to navigate the Covid crisis. The "Number of Deaths" was a critical KPI and maybe one the most objective. However, this KPI was lagging behind the actual risk as deaths occur 3 to 4 weeks after initial contamination. However, the "Number of COVID tests" might trigger alerts whereas the risk was not yet validated (see Omicron variations). It is leading and maybe too much.