How do we decide which time grain to use for a given KPI?

Use the coarsest time grain that still supports the decision you need to make in time.

That is usually the right starting rule. A KPI should be sampled and reviewed at a time grain that matches:

• how quickly the process can materially change,
• how quickly someone can act on it,
• how accurate and complete the timestamped data really is, and
• how much aggregation the metric can tolerate before it hides important variation.

If those factors are not aligned, the KPI becomes either too noisy to manage or too delayed to be useful.

Pick the grain from the decision, not from the dashboard

A practical way to decide is to start with the operating decision the KPI is supposed to inform.

• Sub-minute to minute: use when operators or automated controls can intervene quickly and the source events are captured reliably at that rate.
• 15-minute to hourly: use for shift supervision, line balance, short-interval control, response to downtime, queue growth, or bottleneck monitoring.
• Shift or daily: use for production attainment, first pass yield trends, labor utilization, schedule adherence, and recurring quality loss review.
• Weekly or monthly: use for management review, supplier performance, COPQ trends, capacity planning, or program-level performance.

If no one can make a different decision every five minutes, a five-minute KPI may add cost and noise without adding control.

What to test before locking the grain

• Decision latency: How fast must someone detect and respond to the condition?
• Process rhythm: Does the work happen continuously, by cycle, by lot, by batch, by route step, by shift, or by close period?
• Signal-to-noise ratio: At finer grain, does the KPI reveal meaningful variation or just random fluctuation?
• Data capture fidelity: Are event times synchronized, complete, and attributable to the right asset, order, lot, operation, or operator?
• Denominator stability: At small intervals, does the denominator become too small, making percentages misleading?
• Comparability: Will sites, lines, or vendors calculate the KPI consistently at that grain?

These checks matter because many KPI failures are not mathematical. They are governance and context failures.

Common tradeoffs

Finer grain gives earlier visibility, but it also increases sensitivity to bad timestamps, missing events, clock drift, late transactions, and integration gaps. Coarser grain improves stability and comparability, but it can hide short disruptions, transient quality escapes, and handoff delays.

For example, hourly OEE or downtime views may help a supervisor recover a shift. Monthly OEE is often too slow for execution, but useful for trend review. Conversely, daily scrap rates may be better than hourly scrap percentages if production volume is low and the hourly denominator is unstable.

Some KPIs should exist at more than one grain, but with different purposes. That is acceptable if the calculation logic, timestamp rules, and intended audience are controlled. Without semantic governance, organizations end up arguing over whose number is right instead of acting on the signal.

Brownfield reality

In mixed MES, ERP, historian, SCADA, QMS, and spreadsheet environments, the available time grain is often constrained by system behavior rather than business intent.

Examples include:

• ERP transactions posted in batches rather than at true event time.
• Legacy equipment without reliable state models.
• MES timestamps based on user actions instead of machine events.
• Quality results released after review, not when the condition actually occurred.
• Cross-system clocks that are not synchronized.

In that situation, forcing a very fine grain can create false precision. It may look advanced on a dashboard, but it weakens trust, complicates investigations, and makes cross-system reconciliation harder. In regulated operations, that also raises traceability and change-control concerns if people cannot explain how the KPI was derived at a given point in time.

A practical selection method

1. Define the business question and who acts on it.
2. Set the maximum acceptable delay before action.
3. Map the true event sources and their timestamp quality.
4. Test the KPI at two or three candidate grains.
5. Check whether each grain changes decisions, or only changes chart shape.
6. Document the chosen grain, calculation rules, and exceptions under change control.

If two grains are both needed, treat them as separate governed views of the same KPI, not interchangeable numbers.

The short answer is this: choose the time grain that preserves decision usefulness and data integrity at the lowest operational cost. Not the finest grain available, and not the grain that is easiest for one system to export.