Put a gauge on every joint

Plumbers don't put pressure gauges in the middle of a straight run. They put them at the joints, because joints are where pressure changes and where failures start.

The same is true of businesses, and it's the single most common instrumentation mistake I see. Companies measure activities — calls made, tickets closed, features shipped — because activities happen inside a team and a team can report them. Almost nobody measures the seams between teams, which is precisely where the work gets stuck.

What separates a gauge from a metric

Plenty of numbers get called metrics. Very few of them are gauges. Four properties make the difference.

A gauge is automatic. If a human assembles it, it isn't a gauge — it's a report, and reports have an author, a schedule, and a mood. Numbers produced by hand get produced late, get produced selectively under pressure, and stop entirely when the person who produced them is busy. Which is exactly when you need them.

A gauge sits at a joint. It measures the flow between two segments, not the activity inside one. "Tickets closed" is inside. "Median hours from ticket created to first human response" is a joint — it spans the seam between intake and resolution, and it's the one a customer would recognize.

A gauge has an owner. Not a viewer — an owner. A named role that is accountable for the reading and expected to act when it moves. An unowned dashboard is decoration; it will be watched attentively for two weeks and then never again.

A gauge is readable by the people it affects. Internally, at minimum. Ideally by customers too. A number that only leadership can see will be optimized for leadership, and you will get a metric that goes up while the underlying thing gets worse.

Pick the number a customer would recognize

The best gauges describe an experience someone is actually having.

Not "pipeline velocity" — how many days from the demo to a signed contract. Not "support efficiency" — what fraction of tickets get a real answer the same day. Not "model performance" — how often does the agent's recommendation survive human review.

The test: read the number out loud to a customer. If they'd recognize it as a description of their own experience, it's a good gauge. If they'd need a glossary, you're measuring your internal process rather than the flow through it — and internal-process numbers are the ones that drift furthest from reality before anyone notices.

Three gauges per segment, maximum

The failure mode after "no instrumentation" is "too much instrumentation," and it arrives faster than you'd think. A dashboard with forty tiles is a dashboard nobody reads, which is functionally identical to no dashboard while costing considerably more to maintain.

Three per segment, chosen deliberately:

Throughput — how much is getting through per unit time.
Latency — how long a unit takes to get through, measured at the median and at the 90th percentile. The median tells you about a normal day; P90 tells you about the customer who's about to leave.
Quality — the rate at which output comes back. Rework, refunds, reopened tickets, escalations, human overrides of an agent's decision.

Three numbers per segment across eight segments is twenty-four numbers, which is roughly the ceiling of what an organization can actually hold in its head. Beyond that, you're not measuring more — you're diluting attention, and attention is the scarce input that makes measurement worth anything.

Publish the readings

Here's where the law of clarity earns its keep, and where most companies quietly opt out.

A gauge that leadership reads privately is an early-warning system for exactly one group of people. A gauge published where the whole company can see it becomes something better: shared context. The person upstream of the clog finds out about it at the same moment as the person downstream, without a meeting, and the conversation starts from an agreed number instead of two competing impressions.

Publishing to customers is the harder version and the more valuable one. When an operational number is public, you can't quietly let it slide. You'll fix it, because the alternative is watching it sit there in public. That pressure is not a bug — it's the mechanism. Systems that are visible get maintained. Systems that are hidden get explained.

Start with three gauges, not thirty

If you're beginning from nothing, don't build a measurement program. Build three gauges this week:

Time from a customer asking for something to them getting it. Any customer, any request, measured end to end across every internal handoff.
The size of your largest queue. Whatever is waiting on a human right now, counted. Queue length is the earliest signal of a clog and almost nobody watches it.
Your rework rate. How often does completed work come back. This is your quality gauge, and it's usually the one nobody has ever computed.

Three numbers, produced automatically, posted somewhere everyone can see, owned by a named role.

You'll learn more about your business in the first two weeks of watching those three than in a year of quarterly reviews — because a quarterly review is an argument about what happened, and a gauge is just what's happening.