How to Build a Real-Time KPI System for Your Factory
DBR77 Team7 min read
On this page
A real-time KPI system is a short list of plant measures, each with one written definition, one automatic data source, one owner and a refresh rate that matches the decision it supports. Build it in four steps: write a definition card for each KPI, agree the elements behind each formula, decide how fast each one must update, and give each level of the plant one view with owners and thresholds. The software comes last.
Most plants have too many KPIs, defined three ways and updated from spreadsheets the next morning.
Why the same KPI has three values
Illustrative example: in a weekly review, a plant director sees OEE for line 2 at 78% from production, 71% from maintenance and 64% from controlling.
- Production leaves changeovers out of planned production time, because the team sees them as planned.
- Maintenance counts changeovers as availability loss, but leaves out stops for preventive work.
- Controlling counts only parts that passed first time as good, so every reworked part lowers its quality factor.
Each version can be defended, and together they cost the first 20 minutes of every review. The Real Cost of Siloed Systems and Spreadsheets in a Plant looks at the wider cost of each department keeping its own truth.
- Production
- Changeovers left out
- Maintenance
- Preventive stops left out
- Controlling
- Rework not counted as good
- Owner
- Production engineering
- Threshold
- Below 65% for one shift
- Action
- Review with the shift lead
Line 2 OEE reads 78% in production, 71% in maintenance and 64% in controlling.
Step 1: Write one definition card per KPI
ISO 22400 is the international standard for KPIs in manufacturing operations management. Part 1 sets out a framework for defining, composing, exchanging, and using key performance indicators. Part 2 presents each KPI by means of their formula and corresponding elements, together with its time behavior, its unit and the user group it serves.
That structure makes a good template for a one-page card per KPI:
| Field | What to write | Example for OEE |
|---|---|---|
| Name and purpose | What decision the KPI supports | Find and rank time losses on a line |
| Formula | The exact calculation | Availability × Performance × Quality |
| Elements | Where each input comes from | Run time and stops from the machine, counts from the line |
| Exclusions | What is left out, and why | Breaks and shifts with no planned production |
| Unit and level | Percentage, count, time; machine, line, plant | Percent, per machine and per line |
| Time behavior | How often it updates and over what window | Live during the shift, closed per shift and per week |
| Owner | One role that answers for the definition | Production engineering |
| Threshold and action | When it turns red and who acts | Below 65% for one shift: review with the shift lead |
Keep the list short: fifteen KPIs on cards beat sixty in a workbook nobody can explain.
Step 2: Agree the elements behind each formula
The formula is rarely the problem: in the OEE example, all three teams used the same formula and still got three numbers.
Vorne's guide to calculating OEE shows where the choices sit. Planned production time is shift length minus time with no intention of running production, typically breaks. Ideal cycle time is the fastest cycle the process can achieve in optimal circumstances. Vorne also gives a simple test: performance should never be greater than 100%. If it is, the ideal cycle time is set wrong.
Agree each element in writing:
- Planned production time: are changeovers, trials and cleaning inside or outside?
- Ideal cycle time: from the machine's real best rate per product, not from an old routing.
- Good count: first-pass good only, or including rework?
- Stop categories: one list, shared by production and maintenance.
Step 3: Decide how real time each KPI needs to be
Not every KPI needs to update by the second. Match the refresh rate to the decision:
- Within the shift: OEE, stops, scrap and output against plan. The shift lead acts on these in minutes, so they should come straight from machines and the line.
- Daily: schedule adherence, open quality holds and overdue work orders, reviewed in the morning meeting.
- Weekly or monthly: cost per unit, energy per unit, maintenance cost and on-time delivery, reviewed with controlling.
The gain from live data is largest in the first group. A spreadsheet OEE arrives after the shift that could have acted on it, and Live OEE vs OEE from a Spreadsheet compares the two. For the other groups, an automatic daily or weekly close is enough, as long as it is built from the same records.
Step 4: Give each level one view, with owners and thresholds
Build the KPI view as a tree. The plant view shows a few measures across all lines, each line view the same measures plus its main losses, each machine view states, stops and counts. A director clicks down, while an operator sees the level they act on.
Every number on a view needs an owner and a threshold, and every threshold needs a defined action. When a threshold is crossed, the signal should go to its owner as a task, as From Signal to Task to Closure describes.
Finally, put the definitions under change control. If someone wants to change an element, the owner approves it, the date is recorded, and history is recalculated or marked. OEE Is Not Enough: What to Measure Instead covers which measures belong on the list in the first place.
How this works in DBR77 IRIS
DBR77 IRIS feeds IRIS KPI from every connected module: OEE, cycle times and counts from IRIS MES, MTBF and MTTR from IRIS CMMS, defect counts and inspection results from IRIS QMS, lost-time incidents from IRIS HSE. Indicators that no module produces, such as energy per unit, are entered by hand or in a batch. There are no manual exports. OEE in IRIS MES is calculated from production data, with availability, performance and quality shown separately.
IRIS KPI keeps the definition of each measure with its target, warning and critical thresholds and alert rules, and groups measures into scorecards. Standard measures such as OEE are calculated the same way for every line, and users drill down from the plant summary to line and machine.
IRIS KPI stays a deterministic layer: it measures, compares with targets and alerts. Correlations across modules and recommendations come from Data & AI, and people decide which ones to act on.
FAQ
How many KPIs should a plant track?
Ten to fifteen at plant level is enough for most mid-size plants, each with a card and an owner.
Who should own the KPI definitions?
One role per KPI, usually production engineering for OEE and controlling for cost measures. The owner approves any change to the elements.
Do we have to follow ISO 22400 exactly?
No. Use its structure: formula, elements, time behavior, unit, user group. Where your plant uses a different element, write it down on the card.
Conclusion
Start with the KPI your teams disagree on most. Write its card, settle planned production time, ideal cycle time, good count and stop categories in writing, and check that performance never goes above 100%. Then move to the next KPI, and keep the plant list to ten to fifteen.