What 21,107 Work Orders Say About Your Maintenance KPIs

The Work Happened All Week. The Paperwork Happened on Friday.

Maintenance KPIs like mean time to repair are only as good as the dates and assets behind them.

A NIST study of 21,107 HVAC work orders found that most were closed on a Friday, long after the work was done, which made jobs look 30% or more longer than they were.

The gaps weren’t random, so more data didn’t fix them.

NIST researchers studied 21,107 HVAC work orders. The work orders opened on every weekday, spread about evenly. But 93% of them closed on a Friday.

When the researchers corrected for that, the median work order took 9 to 10 days, not the 14 days the close dates showed. The close dates overstated the median by 30% or more, and every duration number built on them carried that error.

Nobody lied. The work got done during the week, and the records got closed in a batch at the end of it. The close date shows when someone got to a desk, not when the equipment was fixed.

If you’re about to move to new maintenance management software, your history moves with you. Your duration KPIs start out wrong before anyone closes a single new work order.

We cover what to do with your equipment data before you switch on Dec 17.

Your maintenance KPIs look precise, but they may be timing the paperwork

Most maintenance metrics are built from two timestamps: when a work order opened and when it closed.

Work order completion time is the gap between them.

Mean time to repair, backlog age, and response-time trends usually lean on the same two fields.

The same two fields feed reports on maintenance performance, maintenance efficiency, and operational efficiency, so the error travels a long way.

Those numbers only mean something if the timestamps record real events. Here is what the NIST data looked like:

Day	Share of work orders opened	Share of work orders closed
Monday	22%	1%
Tuesday	24%	1%
Wednesday	21%	1%
Thursday	19%	3%
Friday	13%	93%
Saturday	1%	1%
Sunday	1%	0%

Table reproduced from Conte, Bolland, Phan, Brundage & Sexton (2021), “The Impact of Data Quality on Maintenance Work Order Analysis: A Case Study in HVAC Work Durations,” PHM Society European Conference. Used under CC BY 3.0 US.

The opens show when work arrived. The closes show when the paperwork got done. If you calculate repair time from the close column, you’re measuring your admin schedule.

The asset list had 10,000 lines, and the right asset was still hard to find

As the founder of a company I formerly owned that remains the leader in HVAC/R Compliance, and is now part of a major CMMS platform, and after decades of working in HVAC, I’ve seen how this happens firsthand.

This is not anecdotal, and the study from NIST confirms concerns about scattered data and complex documentation.

On one account, the client had an asset list with more than 10,000 items.

Some were real equipment. Some were components. Some weren’t even there anymore. But every one of them was listed.

To get paid for a job, a tech had to attach the work to an asset. So you scrolled, and scrolled, and kept scrolling.

There’s only so much scrolling a person can do. At some point you pick something just to pick something.

Then the people behind a desk approve it, because they have no way to know it’s wrong. The data moves around, but it doesn’t describe anything.

→ Read the full Year One case study

The techs weren’t the problem. The list was.

If the right assets aren’t in the system and mapped the right way, nobody can pick them. And if adding a missing asset is harder than choosing a wrong one, people will choose the wrong one every time.

Reading the nameplate was the first generation, and we’re living in the seventh

A lot of tools now promise to fix asset data by reading the nameplate for you.

Point the phone, the text comes off the plate, and the fields fill in. That’s OCR, and it was a real step forward when it arrived.

But I think of OCR as a first-generation tool. It reads what’s printed on the plate, and that’s all it does. It can’t tell you whether that unit is already in your list under a different name.

It can’t tell you whether it’s a component that belongs under another unit, or whether the line you’re about to pick describes equipment that was pulled out years ago.

And if the plate is painted over, worn off, or gone, there’s nothing to read.

We’re now about seven generations past that. The question isn’t “what does the plate say?” anymore. It’s “what is this equipment, where does it belong, and is it already in the record?”

A faster nameplate reader on top of a 10,000-line list still ends in a wrong pick, just a quicker one. The record is only right when the asset itself is right: found, matched to what’s really there, and added when it’s missing.

Some missing data is only noise, and some changes the answer

Every CMMS has gaps, and your maintenance KPIs are only as good as the records underneath them.

Most teams assume the gaps even out. Sometimes they do.

The NIST team tested this. They took a corrected dataset and removed records two ways.

When they removed records at random, the averages barely moved. The numbers got fuzzier, but they stayed centered on the truth.

Then they removed the rare, unusual jobs first, the ones that are harder to describe and more tedious to close.

This time the averages moved a lot, and not in a straight line. How far off the KPI got depended on which work went missing, not how much.

→ Read the full Year One case study

Figure reproduced from Conte, Bolland, Phan, Brundage & Sexton (2021), “The Impact of Data Quality on Maintenance Work Order Analysis: A Case Study in HVAC Work Durations,” PHM Society European Conference. Used under CC BY 3.0 US.

That’s what we call MNAR drift. MNAR stands for “missing not at random.”

It means the gaps in your data are tied to the kind of work being done. The easy picks near the top of the list get chosen again and again, and the rare equipment slowly drops out of the record.

Your KPI drifts away from reality, and adding more records of the same kind won’t pull it back.

When that happens, your KPI trend stops showing the real work in your maintenance operations, including where your unplanned downtime comes from.

Some maintenance KPIs hold up against bad records, and some quietly break

Most lists of essential maintenance KPIs cover the same maintenance key performance indicators.

They don’t all fail the same way. Some depend on when a work order closed. Some depend on which asset the work landed on.

Some depend on which work got recorded at all.

Here’s how the most common maintenance performance metrics hold up when the records behind them look like the NIST data, and these are the key metrics most teams track because effective maintenance metrics usually include MTTR, MTBF, OEE, and Planned Maintenance Percentage.

KPI	What it’s built on	Verdict	What happens
Mean time to repair (MTTR)	The average time from open to close on repair work	BREAKS	MTTR calculates average time to repair equipment after failure, so batched closes count paperwork days as repair time. Lower mean time to repair indicates faster repairs and reduced downtime, which is why this 30%+ overstatement from the NIST study matters.
Mean time between failures (MTBF)	Failures recorded against each asset	BREAKS	If techs pick the wrong line, failures land on the wrong equipment. A higher MTBF indicates better equipment reliability and supports uptime and productivity, and typical benchmarks range from 500 to 2,000 hours, but bad records make some units look worse and others better when neither is true.
Planned maintenance percentage (PMP)	Planned maintenance hours ÷ total maintenance hours × 100	BREAKS	Rare, reactive jobs are the ones most likely to go unrecorded. When they drop out, your proactive mix and your preventive maintenance program look stronger than they are, even though an ideal Planned Maintenance Percentage benchmark is 85% or higher and a high PMP suggests effective preventive maintenance scheduling.
Reactive maintenance percentage	Unplanned work as a share of all maintenance work	BREAKS	Same cause, opposite direction. The emergency work that goes missing is the work this KPI is meant to count, even though a reactive maintenance percentage below 20% is ideal and organizations relying on reactive maintenance experience 3.3 times more downtime.
Preventive maintenance compliance and schedule compliance	The share of scheduled preventive maintenance tasks completed on time	BENDS	If the close date stands in for the completion date, a PM done on Tuesday but closed after its due date on Friday counts as late.
Maintenance backlog	Open work orders and how long they’ve been open	BENDS	Work that’s finished but not yet closed still counts as pending maintenance work. The backlog looks bigger than it is, even though an ideal backlog is two to four weeks, and going beyond that range may point to a need for more resources.
Equipment downtime, equipment availability, and overall equipment effectiveness (OEE)	OEE = availability × performance × quality, with availability coming from how long equipment was out of service	BENDS	Asset availability is the percentage of total scheduled time that equipment is operational. If downtime comes from work order times, machine downtime runs long and availability runs short, even though aiming for less than 5% equipment downtime is ideal. If it comes from equipment sensors or production logs, it holds.
Total maintenance costs	Every cost recorded in the period	HOLDS	Totals stay right as long as every job gets recorded.
Maintenance costs per asset	Costs recorded against each asset	BREAKS	Wrong picks move costs onto the wrong equipment, so the repair-or-replace decisions built on them go wrong too. Tracking Cost as a Percentage of RAV helps, because Maintenance Cost as a Percentage of Replacement Asset Value compares annual maintenance spending to replacement cost to justify whether to replace aging assets or continue repairs.

High MTTR or low Availability can also point to issues in spare parts availability or troubleshooting workflows.

The pattern is simple. KPIs that add everything up tend to survive.

KPIs that divide by time, or split results by asset or type of work, tend to break. Most of the metrics maintenance teams use to plan maintenance strategies, set maintenance schedules, and measure continuous improvement are the second kind.

That matters for equipment reliability too.

You can’t improve asset reliability, or trust a predictive maintenance model, when equipment failures are recorded against the wrong unit. Every asset performance report inherits the same error.

Here’s what that looks like with one wrong pick

When the asset list isn’t well known, identifying the right equipment and its parts gets harder, and the work, the parts plan, the maintenance schedule, and the depreciation plan stop lining up.

Say you have two rooftop units on the same roof, RTU-4 and RTU-7. RTU-7’s compressor fails, and a tech replaces it.

When they close the work order, they scroll the list and pick RTU-4 by mistake. Nobody catches it. Run the Friday Test on your preventive maintenance metrics.

Area	What the record says	What’s actually true	What goes wrong
Work history	RTU-4 had a compressor failure. RTU-7 has had no repairs.	RTU-7 failed. RTU-4 is fine.	MTBF looks worse on a healthy unit and better on the one with problems.
Parts planning	RTU-4 has a new compressor, installed on this date.	RTU-7 has the new compressor, and its model, voltage, or refrigerant may not match RTU-4’s.	The next parts order is based on the wrong unit, and the warranty is tied to a serial number the part isn’t in.
Maintenance scheduling	RTU-4 needs follow-up checks. RTU-7 stays on its normal schedule.	RTU-7 is the one that should be watched.	Your team spends time on the wrong unit and skips the one that needs attention.
Depreciation and replacement planning	RTU-4 got a major component. RTU-7 is still running its original one.	The opposite.	If you capitalize major repairs, the value lands on the wrong asset. RTU-7 may get flagged for early replacement, while RTU-4 carries value for a part it doesn’t have.
Refrigerant records	Any refrigerant added during the repair went into RTU-4.	It went into RTU-7.	The leak rate is calculated on the wrong unit, which is a compliance problem, not just a KPI problem.

Notice what still works: your total maintenance costs for the year are right, because the money was spent either way. Everything split by asset is wrong.

That’s one wrong pick. In a 10,000-line list, it’s never just one.

You need one export and about 20 minutes.

It tells you whether the records behind your key performance indicators can be trusted, and anyone tracking kpis can run it as a quick reference for checking the records behind those numbers.

Step	What you do	What it tells you
1	Export open dates and close dates for the last 12 months of work orders.	You now have the two fields most duration KPIs depend on.
2	Count each column by day of the week.	You can see when work arrives and when records get closed.
3	Compare the two counts side by side.	If opens spread across the week but closes pile up on one day, your close dates reflect a paperwork routine.
4	Look at which work orders have blank or late closes.	If the blanks cluster around certain equipment or job types, the gap has a pattern, and your averages are drifting. Reviewing that pattern helps confirm whether KPI trends are on track or biased by late closes. Late closes can also hide real work in your maintenance backlog.
5	Count how many work orders each asset has.	If a few assets collect far more work than makes sense, people are picking what’s easy to find, not what they fixed. That can make equipment downtime look better or worse than it really is. This check supports better resource allocation and helps teams prioritize cleanup on the assets with suspicious work-order volume.

If your closes look like the table above, your duration KPIs are probably running long.

Tracking work order completion rates also helps confirm effective labor allocation. The NIST data suggests the gap can be 30% or more.

A failed test proves your maintenance KPI is off, but passing doesn’t prove it’s right

A clean Friday Test is good news, but it only checks a few fields.

Your close dates can be spread across the week and still be entered days after the work. A failed test doesn’t mean your team did anything wrong.

It means your KPIs describe a real process, the paperwork, instead of the one you think you’re measuring.

You can’t fix years of records overnight, but you can stop adding bad ones this week

Once the Friday Test shows a problem, the job for maintenance managers is to set specific KPI improvements from here, not to rewrite history. Start with how your maintenance team records the work.

The NIST authors point to a few simple changes, with the goal of making cleaner records available for relevant maintenance metrics. Set a time during the day for data entry instead of saving it for Friday.

Pair new hires with an experienced tech. Encourage maintenance technicians to write more when they’re unsure.

And use an entry screen that doesn’t take a manual to learn.

Until the records are clean, report duration KPIs as a range, and say which ones were affected so teams can take action on those measures rather than treat them as exact.

That’s more honest than one precise number.

It also means that when you work to reduce maintenance costs or reduce unplanned downtime, you’re working on the real problems, not the ones the paperwork created.

None of this needs new preventive maintenance software.

It needs preventive maintenance activities and critical maintenance tasks recorded when they’re done, not when Friday comes.

High PM compliance and strong MTBF scores help prevent costly facility disruptions in the long run.

How do you measure CMMS ROI and equipment downtime when your equipment data is incomplete?

Start by asking whether the gaps are random in your data collection.

If they are, your averages still hold, and you can report them with a wider margin.

If the gaps follow the work, with rare jobs, certain sites, or certain equipment missing more often, your averages are biased, and the company should align KPI reporting with numbers you’ve checked rather than assume more data will fix them.

Run the Friday Test first, then measure the CMMS benefits you’re paying for, including preventive maintenance, against numbers you’ve checked.

Once you have, those numbers help you make informed decisions, justify budgets, and decide which repairs come first to optimize maintenance decisions and support long term success.

Why do maintenance KPIs look worse than the work?

The work gets done during the week, but the record often gets closed later.

Every day between the repair and the paperwork counts as repair time, which is why regular KPI review matters if you want to make informed decisions quickly.

In the NIST data, that lag made the median job look 30% or more longer than it was.

Mean time between failures has the same weakness: if the close date is late or the work lands on the wrong asset, the failure gets recorded in the wrong place.

Late records or the wrong asset assignment can also make unplanned stops and technician errors distort what the KPI seems to show.

You can’t fix the record from behind a desk, but you can fix it at the equipment

A record you can trust is made at the equipment, by the person standing in front of it. That’s why I built Tag Wizard.

You take the picture. Tag Wizard does the rest. Here’s how that answers the problems in this story:

The problem	What Tag Wizard does
A 10,000-line list where the right asset is hard to find	Identifies the equipment from the photo and checks whether it’s already in your record, so duplicates get caught instead of added
Components mixed in with equipment	Puts components under the unit they belong to at the moment of capture
A nameplate that’s painted over, worn off, or gone	Identifies make, model, and category even when there’s no readable tag
The same unit named five different ways by five techs	Uses one consistent naming standard, so every tech’s record lines up
A CMMS you already pay for	Works under any CMMS and feeds it clean asset data, instead of replacing it

It doesn’t make your team do more paperwork. It makes the right asset the easiest one to pick.

Until your records work this way, a dataset can look full and still mislead you.

Good records improve overall effectiveness, especially in manufacturing environments where teams rely on maintenance data to protect uptime.

Good records let you see which equipment needs attention before it fails, because OEE depends on availability performance and quality as the three factors that show where losses begin.

An OEE score of 85% or higher is considered world class for manufacturers, while many manufacturing operations run closer to 60% productivity. If the part you’re missing is the rare, messy work, the part you have only describes the easy jobs.

60% complete is 0% trustworthy.

Explore how cleaner asset data extends equipment lifespan, supports expected replacement timing, and improves visibility into units produced and parts history.

The work orders in the study cover ten years, from 2009 to 2019. They came from HVAC maintenance at a site with mixed-use laboratory and office space.

The paper itself was published in 2021 at the 6th European Conference of the Prognostics and Health Management Society.