In short
Track four operational metrics and resist adding more: cycle time (how long work takes end to end, including waiting), throughput (completed items per period), work in progress (started but unfinished), and rework rate. Together these tell you whether the business is getting faster, whether it is getting more done, whether it is overcommitted, and whether it is getting things right first time. Activity metrics — tasks logged, hours recorded, messages sent — measure effort rather than health and almost always look good regardless of what is actually happening.
Key takeaways
- Activity metrics measure effort. Flow metrics measure whether the business works. Only one changes decisions.
- Cycle time should be measured as a distribution, not an average. Customers experience your worst cases.
- Work in progress is the only common leading indicator: it rises before delivery slows.
- Rework rate is the most expensive thing most businesses do not measure at all.
- A metric with no owner and no threshold is a fact, not a control.
Why most operational dashboards change nothing
The typical operations dashboard reports how much activity occurred. Tasks completed, tickets closed, hours logged, messages sent. It is easy to build, because these numbers are what tools emit by default, and it is comforting, because activity is nearly always high in a business that feels busy.
It also cannot answer the question anyone actually has, which is some version of: are we getting better or worse, and at what?
Three failure patterns
Measuring effort rather than flow. Hours logged tells you people were working. It does not tell you whether work is moving. A team can log record hours during its worst delivery month, and frequently does, because rework consumes hours.
Reporting lagging indicators only. Revenue, churn and delivered projects are all outcomes of operational conditions that existed weeks earlier. By the time they move, the cause is old.
Producing numbers with no threshold. A chart showing cycle time at 11.4 days is a fact. Nobody knows whether to act on it, so nobody does. A metric becomes a control only when someone has said what number is unacceptable and what happens when it is reached.
If a number moves and nothing happens, you are not measuring — you are recording.
The four metrics worth the trouble
These four cover most of what a growing business needs to know about its own operations. Each is cheap to collect and hard to game.
1. Cycle time
Elapsed time from the moment work starts to the moment it is done, including all waiting. Not touch time — elapsed. This matters because waiting is where most delay lives, and touch time excludes exactly the thing you are trying to see.
Track the distribution. A median of six days with a 90th percentile of twenty-eight is a very different operation from a consistent seven, and customers experience the tail. The gap between median and 90th percentile is one of the most informative numbers in the whole set: a wide gap means your process has unmanaged variation, usually a queue or an exception path.
2. Throughput
Completed items per week or month. The only metric that speaks directly to business capacity. Note that throughput and cycle time can move independently, and comparing them is diagnostic: throughput flat while cycle time rises means you are starting more than you finish.
3. Work in progress
Items started but not finished, counted at a fixed moment each week. This is the closest thing operations has to a genuine leading indicator — it rises before cycle time does, because everything started must eventually be finished or abandoned.
Rising work in progress predicts rising cycle time weeks before it appears in delivery. If you can only track one metric, track this one, because it is the only one that gives you warning rather than confirmation.
4. Rework rate
The proportion of items that require redoing after being considered complete. Almost never measured, and the most expensive thing in most operations, because you pay twice for the work and sometimes a third time in the relationship. Even a rough two-week sample, categorised by cause, is more valuable than a precise measurement of anything on the activity dashboard.
Metrics that look useful and are not
Some widely used measures actively mislead. Worth naming them.
| Metric | Why it appeals | Why it misleads |
|---|---|---|
| Utilisation | Feels like efficiency | High utilisation at non-constraint steps increases work in progress and slows everything down. Utilisation of 100% guarantees queues. |
| Tasks completed | Easy to collect, always rising | Counts small and large equally; rewards splitting work rather than finishing it |
| Average response time | Sounds customer-focused | An average hides the outliers that generate every complaint |
| Hours logged | Available from existing tools | Measures input. Rises during your worst months, because rework is hours. |
Utilisation deserves particular caution, because it is the metric most likely to be actively harmful. Managing towards full utilisation removes all slack from the system, and a system with no slack cannot absorb variation — which means every small disruption becomes a delay that propagates. This connects directly to bottleneck behaviour: only the constraint should be run near capacity, and everything else needs slack in order to keep the constraint fed.
Collecting the data without a project
The instinct is to solve measurement with tooling. Start smaller; the tooling question resolves itself once you know what you are measuring.
Cycle time requires only two timestamps: started and done. Most work-tracking tools record these already. If yours does not, a spreadsheet with two date columns is entirely adequate for the first three months and will teach you what you actually need.
Throughput is a count of items reaching done in a period. Trivial once done is defined — and defining done is often the real work, because in many businesses it is genuinely ambiguous whether something is finished when it ships, when it is accepted, or when it is invoiced. Pick one, write it down, do not change it.
Work in progress is a count taken at the same time each week. Manual counting is fine and has an underrated benefit: the person counting notices the items that have been in progress for a suspiciously long time.
Rework needs a deliberate capture habit, since no tool emits it by default. Add one field to your workflow: was this redone, and why? Three categories are enough — wrong input, missed requirement, late change.
Definitions before dashboards
Write down what each metric means, in one sentence, including the exact start and end points. Ambiguous definitions are how two people produce different numbers for the same metric and the whole exercise loses credibility. Then leave the definitions alone, because a consistent imperfect definition is worth far more than a series of improved ones.
Turning metrics into controls
A number becomes useful when three things are attached to it: an owner, a threshold, and an agreed response.
| Metric | Threshold example | Agreed response |
|---|---|---|
| Cycle time (90th percentile) | Above 3x the median | Investigate the tail cases specifically; they usually share one cause |
| Throughput | Down two periods running | Run the queue walk; the constraint has probably moved |
| Work in progress | Above the agreed limit for the team | Stop starting, start finishing. No new work until it clears. |
| Rework rate | Above 10% of completed items | Categorise causes and fix the largest upstream |
The specific numbers should be yours, set from your own baseline rather than borrowed. What matters is that they exist, that someone owns each one, and that the response is agreed in advance — because a threshold discussed in the moment becomes a negotiation about whether this particular case counts.
Review rhythm
Weekly for work in progress, which moves fast and gives warning. Monthly for the others, which need enough data to distinguish signal from noise. Quarterly, review the metrics themselves: is each still changing a decision? A metric that has never triggered its threshold in a year is either well-controlled or badly set, and it is worth knowing which.
What genuine operational visibility feels like
It is less dramatic than a dashboard suggests. The markers:
- Someone can answer “how long does this usually take?” with a number and a range, without opening anything.
- Problems are noticed before customers report them. The single clearest test.
- Arguments are about causes, not about numbers. When people dispute the data rather than the interpretation, the definitions are ambiguous.
- Commitments are made from the distribution, not the best case. Quoting the median as a promise means missing half of them.
- Nobody has built a private spreadsheet because the official numbers are not trusted.
That last one is worth watching for. Shadow reporting is the most reliable indicator that your official measurement has failed. If a manager maintains their own version, the official numbers are either wrong, too slow, or measuring the wrong thing — and the manager has already worked out what should be measured instead. Ask them.
The Mayim Ops assessment scores measurement and visibility as one of ten operational dimensions, and reports where you are flying on impression rather than evidence.
Frequently asked questions
What operations metrics should a small business track?
Four: cycle time, meaning elapsed time from start to done including waiting; throughput, meaning completed items per period; work in progress, meaning items started but unfinished; and rework rate, meaning the proportion of completed items that had to be redone. Together they show whether the business is getting faster, getting more done, overcommitted, and getting things right first time.
Why should I measure cycle time as a distribution rather than an average?
Because customers experience your worst cases, not your average. A median of six days with a 90th percentile of twenty-eight days is a different operation from a consistent seven, and the gap between the two reveals unmanaged variation, usually a queue or an exception path that has never been examined.
Is utilisation a good operations metric?
Generally no, and it can be actively harmful. High utilisation at non-constraint steps increases work in progress and lengthens cycle time, and managing towards full utilisation removes the slack a system needs to absorb variation. Only the constraint should run near capacity; everything else needs slack to keep the constraint supplied.
What is a leading indicator in operations?
Work in progress is the most practical one. Because everything started must eventually be finished or abandoned, rising work in progress predicts rising cycle time weeks before the slowdown appears in delivery. Most other operational numbers, including revenue and delivered projects, are lagging indicators of conditions that existed earlier.
How do I stop building dashboards nobody uses?
Attach three things to every metric: a named owner, a threshold that defines unacceptable, and a response agreed in advance. A number with none of these is a record rather than a control. Review quarterly whether each metric has actually changed a decision, and remove the ones that have not.