Operational KPIs: how to choose metrics that help you make decisions
Example indicators in four groups, a simple method for choosing a few KPIs for a single process, and the mistakes that stop measurement from leading to action.
In many of the companies I talk to, there is no shortage of metrics. There is a monthly report with dozens of items, there are charts in the board presentation, there are spreadsheets kept by managers. Even so, the question “what did you do differently last week because of these numbers?” is met with silence. When I ran operations departments, I learned to judge metrics by a single criterion: does anyone make a decision based on them. Below I describe how I choose operational KPIs so that they meet that criterion.
How a KPI differs from an ordinary metric
A metric is any number that can be counted: the number of orders, hours worked, the number of emails in the inbox. A KPI is a metric we have chosen as key because it says something important about achieving a goal and has a response attached to it. If the indicator falls, it is known who deals with it and what steps they consider. If it rises, it is known whether that is good news.
A simple test: ask what happens when the value crosses the threshold. If the answer is “nothing, we will note it in the report”, you have an informational metric, not a KPI. Such numbers are sometimes needed too, but they should not take up space on the board the team looks at every day.
Outcome indicators and process indicators
It is worth distinguishing two kinds of indicator. Outcome indicators (lagging) show the effect: on-time delivery for the month, the number of complaints, the cost of handling an order. They matter, because they are what the customer and the board care about, but they have one drawback: by the time you see them, the problem has already happened.
Process indicators (leading) measure what happens earlier and influences the outcome: the number of jobs in progress, the backlog at the preparation stage, the percentage of jobs sent back for rework. If the queue in front of one workstation is growing, on-time delivery will fall in a few days. A good set of KPIs for a process combines both: the outcome we want to achieve, and two or three indicators that let us react before the outcome deteriorates.
Four families of operational KPIs with examples
I divide operational indicators into four groups: time, quality, cost and productivity, and flow and stability. For most processes, from order fulfilment to complaint handling or invoicing, a sensible indicator can be found in each of them. The table below gives examples with simple ways of calculating them. The response thresholds have to be set for the specific process, based on its history.
| Indicator | What it measures | How to calculate | When to react |
|---|---|---|---|
| Lead time | Time from accepting a job to delivering it to the customer | delivery date minus job acceptance date | When it grows for several weeks in a row or approaches the time promised to the customer |
| Cycle time | Time spent working on a job at one stage | stage completion date minus stage start date | When the spread between jobs grows, not just the average |
| On-time delivery (OTD) | Percentage of jobs delivered by the agreed date | jobs on time / all jobs × 100% | When it falls below the threshold in a given week |
| First pass yield (FPY) | Share of work completed without rework | units without rework / all units × 100% | When it falls or a new, recurring type of error appears |
| Complaints | Errors that reached the customer | number of complaints / number of deliveries × 100% | On every critical complaint and on an upward trend |
| Rework share | How much work we do a second time | rework hours / all hours worked × 100% | When the share grows for a second period in a row |
| Unit cost | Cost of handling one unit (an order, an invoice, a ticket) | process cost in the period / number of units handled | When it grows without a change in the scope of work |
| Productivity | Amount of work per person or per hour | number of units / number of labour hours | When it falls clearly below the baseline |
| Work in progress (WIP) | How many jobs are open at the same time | number of open jobs at a stage, at the moment of measurement | When it exceeds the set limit for the stage |
| Backlog | Jobs waiting longer than they should | number of jobs waiting longer than the set number of days | When it grows for a second review in a row |
| Exception rate | Cases handled outside the standard path | exceptions / all cases × 100% | When it grows, because it is a sign that the standard no longer fits reality |
A few remarks on the table. Lead time and cycle time are often confused, yet they answer different questions: the first says how long the customer waits, the second how long the work itself takes. A large difference between them means jobs mostly sit in queues. The exception rate, in turn, is an underrated indicator. When it grows, it usually means that the process described in the procedure and the process performed in practice are starting to diverge.
How to choose KPIs for a single process
For one process I choose between three and seven indicators. Below three it is easy to miss an important area, for example to watch on-time delivery at the expense of quality. Above seven the team stops remembering what it is looking at, and the review turns into reading out numbers.
Start with the customer and the purpose of the process
The first question is: what does the customer of the process expect? The customer may be an external recipient or another department. If they expect fast, error-free fulfilment, the main outcome indicators are on-time delivery and complaints. Only then do I choose the process indicators that influence them. A process map helps here, because it shows where queues and rework arise. I described how to prepare one in the article on process mapping.
Every indicator has an owner
The owner is one person who has a real influence on the value of the indicator and is responsible for the response when it crosses the threshold. They do not have to fix everything themselves, but they do have to initiate action. If the indicator has a department name or several names against it, in practice nobody looks after it. When settling responsibility, a RACI matrix comes in useful.
The indicator definition sheet
For every KPI I prepare a short definition sheet. It contains the name, the purpose (what question it answers), the formula, the data source, the measurement frequency, the owner, the target value and the threshold beyond which action is required. The sheet seems a formality until two people come to a meeting with different values for the same indicator, because one calculates on-time delivery from the order date and the other from the confirmation date. Then the conversation is about definitions, not about the process.
Review rhythm and data source
An indicator works when someone looks at it regularly and talks about deviations. Process indicators work best in short daily reviews, ten to fifteen minutes at a board, physical or on screen. I discuss outcome indicators weekly, and the monthly review serves to assess trends and change targets. At every review I am interested in three things: where we are outside the threshold, why, and who is doing what before the next meeting.
The second question is how data is collected. A spreadsheet filled in by hand is enough to begin with, because it lets you check whether the indicator is useful at all. In the long run manual collection costs time, delays the data and introduces transcription errors. Once the set of KPIs has stabilised, it is worth pulling the data directly from the systems where it originates and refreshing the board automatically. I describe how to assess whether such a step pays off in the article when to automate a process, and when not yet, and examples of automated reporting are on the Automation and AI page.
The most common mistakes with KPIs in operations
Too many indicators
Every manager adds their own numbers, nobody removes the old ones. After a year the report has dozens of items and nobody knows which matter. Once a year it is worth going through the list and removing the indicators that have not triggered a single decision.
Indicators the team cannot influence
Holding the warehouse to account for company revenue, or the customer service department for the number of tickets, which depends on product quality, breeds frustration. The team should see the link between its work and the number.
Averages that hide the spread
An average turnaround time of three days may mean that all jobs take three days, or that most take one day and some take two weeks. The customer remembers those two weeks. That is why, alongside the average, I show the percentage of jobs above a set time.
Targets with no starting point
A target of “95% on-time delivery” set without knowing what the figure is today is guesswork. First a few weeks of measurement, then the target.
An indicator that can be gamed
When the number becomes an end in itself, people start optimising the number instead of the process. Closing tickets before the problem is solved improves handling time but not customer satisfaction. A pair of indicators that balance each other helps, for example handling time and the percentage of tickets reopened.
A report that arrives too late
A monthly report ready on the tenth day of the following month describes the situation from six weeks ago. That is enough for assessing trends, but not for steering the process.
KPIs and the requirements of ISO 9001
Companies certified to ISO 9001 sometimes ask whether the standard specifies which indicators to measure. It does not. Clause 9.1, on monitoring, measurement, analysis and evaluation, requires the organisation itself to determine what it monitors and measures, by what methods, when it does so and when it analyses the results. Well-defined KPIs with definition sheets and regular reviews are a natural way of meeting this requirement, but it is the organisation that is responsible for choosing indicators appropriate to its processes. A set of charts in the quality manual on its own is not enough if nobody acts on them.
Where to start
Choose the one process that causes the most trouble today. Establish what its customer expects, choose one outcome indicator and two process indicators, prepare definition sheets for them, and review them weekly for a month. After a month you will know which numbers help you make decisions and which merely fill the report. Only then is it worth extending measurement to further processes.
If you want to put measurement in your operations in order, take a look at the Operations optimisation page or try the maturity test, which shows where your processes stand today.
Let us talk about your process.
The first call takes about 30 minutes. Tell us what gets in the way most in your process and we will say whether and how we can help.
Talk to us