Maintenance Capacity Planning: The Probability You Finish the Program

Maintenance capacity planning should tell you how likely you are to finish the program of work. Here is how we estimate it, and why P80 and P90 matter.

Maintenance Capacity Planning: The Probability You Finish the Program

Maintenance capacity planning should answer one question: how likely are we to finish the program of work? I've been giving this a lot of thought. We have been working with a couple of our clients on maintenance planning and scheduling, defining what best practice looks like and what a maintenance plan and a maintenance schedule actually are, measured against the standards and against what the business expects of them.

We arrived at a simple rule. Every maintenance plan and every maintenance schedule should come with a probability estimate of completion. Across a thousand possible futures, or ten thousand, how many times would you actually deliver this plan?

Due dates hide the real question

Most maintenance plans are a list of work with due dates. They say what should happen and when. They say nothing about whether the people, tools and fixed plant exist to make it happen.

A bad maintenance plan is one that never explores the capacity of its resources, whether that is trade and labour resources or fixed plant resources. It looks fine on paper until the week the crane is booked twice and the electricians are short.

Supply and demand in maintenance capacity planning

Every maintenance task is demand. Every fitter, electrician, crane, pit and wheel lathe is supply. A plan works only when the two meet, week after week, at every site.

That is what maintenance capacity planning is for. It sets the demand created by the maintenance program against the supply of trades, tools and fixed assets, and shows where they will not meet. Our rostering and workforce planning work starts from exactly this view.

How we estimate the probability of completion

The method has five steps. None of them needs perfect data to begin.

  1. Give every task a resource recipe. Each task lists the trades and hours it needs, any specialist tooling, any plant, and the fixed asset it occupies, such as a road, a pit or a crane.
  2. Estimate durations with three points. Planners give a best case, a most likely case and a worst case for each task. They know these numbers even when the maintenance system does not hold them.
  3. Schedule the most constrained work first. Jobs that need the most fixed assets and the scarcest trades go into the schedule before the flexible work, which fills the gaps around them.
  4. Simulate thousands of futures. A Monte Carlo simulation draws a duration for every task, schedules the work against the resources available, and records whether the program finishes. Then it does it again, thousands of times.
  5. Read the result as a probability. The runs build a curve that gives the chance of finishing by each date. We report it for the whole program, then break it down by site, by trade and by asset.

P80 and P90

A good maintenance plan states its P80 or its P90. P80 means that in 80 per cent of the forecast futures, we deliver the plan. P90 means we deliver it in 90 per cent of them. If we can do that, it is a good plan.

P80 vs P90 on a cumulative probability curve for maintenance capacity planning: an 80 per cent chance of finishing by day 32 and a 90 per cent chance by day 35
An illustrative example. Each point on the curve is the chance of finishing the program by that day. Here the P80 is day 32 and the P90 is day 35, so three more days buys the extra ten per cent of confidence.

A good plan is one you can achieve, and one that makes the most of your trades. A plan that is only ever achieved on its best day is a hope.

Where the probability drops

The breakdown is where the value sits. When the P90 for one site falls away, the breakdown shows why. Perhaps one trade is short in one fortnight. Perhaps two heavy jobs need the same pit.

Each answer points to a decision: change the roster, move people between sites, recruit, or make the case for another fixed asset. Because the plan is a model, you can test each option before committing to it. This draws on the same work order and resource data our asset data analytics and fleet management work relies on.

Questions to ask of your own plan

  • Does each task carry its labour, tooling and fixed plant needs?
  • Do your planners give a range for each duration, or a single number?
  • Can you say how likely this month's program is to finish?
  • When the program slips, can you see which resource caused it?

So consider this. How are you building the demand and supply of maintenance work into your maintenance planning, and into your asset management system?

Frequently asked questions

What is maintenance capacity planning?

Maintenance capacity planning compares the work the maintenance program demands with the trades, tools and fixed plant available to do it, over time and by site. It shows where the program will fall short before it does.

What does P90 mean in a maintenance plan?

P90 is the point at which 90 per cent of simulated futures finish the program of work. A plan with a P90 date has a nine in ten chance of being delivered by that date.

How is the probability of completing a maintenance program calculated?

Each task gets a resource recipe and a three point duration estimate. A Monte Carlo simulation then schedules the program against available resources thousands of times and counts how often it finishes by each date.

Find out how likely your program is to finish

If you would like to know the probability that your current maintenance program will be delivered, and what is holding it back, contact us.

Related reading

Watercolour of a maintenance depot with many possible futures fanning across it, one highlighted path showing maintenance capacity planning by probability

Maintenance Capacity Planning: The Probability You Finish the Program

Maintenance capacity planning should tell you how likely you are to finish the program of work. Here is how we estimate it, and why P80 and P90 matter.
Illustration of an AI agent reusing a cached rulebook to cut the cost of AI agents across many maintenance work orders

How Prompt Caching Cuts the Cost of AI Agents in Maintenance

AI agents make the same decision thousands of times. Prompt caching lets them reread the rulebook at a tenth of the price. Here are the numbers from a real job.