---
title: The Error You Cannot See: What Manual PO and Invoice Matching Really Costs You — QuoteToMe
description: Manual purchase order and invoice matching fails quietly. A plain look at the research on data entry error rates, invoice exceptions, processing cost, and the payment delays they cause on a construction job.
url: https://quotetome.com/blog/cost-of-manual-po-invoice-matching
markdown_url: https://quotetome.com/blog/cost-of-manual-po-invoice-matching.md
image: https://quotetome.com/qtm/assets/the-error-you-cant-see-blog-header.png
author: QuoteToMe
published: September 2026
tag: Procurement
read_time: 9 min
---

# The Error You Cannot See: What Manual PO and Invoice Matching Really Costs You — QuoteToMe

Manual purchase order and invoice matching fails quietly. A plain look at the research on data entry error rates, invoice exceptions, processing cost, and the payment delays they cause on a construction job.

[Blog](https://quotetome.com/blog) / Procurement

# The Error You Cannot See: What Manual PO and Invoice Matching Really Costs You

A purchase order, a delivery, and an invoice have to agree before you pay. When a person does that matching by hand, most of the time it works. The problem is what happens the rest of the time, and how rarely anyone finds out.
**QuoteToMe**_|_September 2026_|_9 min read

![A purchase order, a delivery slip, and a vendor invoice laid side by side with mismatched quantities and prices circled, showing where manual three way matching breaks down](/qtm/assets/the-error-you-cant-see-blog-header.png)

0.83%
Manual transcription error rate per keystroke, measured under controlled laboratory conditions.
JAMIA, 2019

14%
Average share of invoices that stop as an exception. The bottom tier runs 22%.
Ardent Partners, 2024

17.4 days
Invoice cycle time for organizations outside the top tier. Best in class clear one in 3.1.
Ardent Partners, 2024

## The short version

Manual matching does not fail loudly, it fails quietly and late. Published benchmarks put the average invoice exception rate at 14%, the average cost to process one invoice at $9.40, and the average invoice cycle time at 9.2 days. For organizations outside the top tier, those numbers run to 22%, $12.88, and 17.4 days. The error rate on any single keyed field is small, but a purchase order, an invoice, and a cost coding pass together carry dozens of fields, and the odds compound. In construction, where cash moves on progress payments, a stalled invoice at the bottom of that chain becomes a payment problem at the top of it.

Nobody sets out to key a wrong number. That is exactly why this cost is so hard to see. Every individual step looks fine, every person involved is doing their job properly, and the failure only becomes visible weeks later when an invoice does not agree with an order and somebody has to stop and work out why.

So it is worth putting real numbers on it. Not the numbers a software company would like to be true, but the ones that have actually been measured, in studies you can go and read.

## It starts with one field

The smallest unit of this problem is a person reading a number off one document and typing it into another. That has been measured properly, and repeatedly.

A peer reviewed study published in the _Journal of the American Medical Informatics Association_ measured manual transcription in a clinical laboratory and found an error rate of [0.83% per keystroke](https://pmc.ncbi.nlm.nih.gov/articles/PMC6913214/). Related laboratory studies cited in the same work found 1.14%, and under more complex conditions, between 3% and 5%. High confidence

A separate study published in _PLoS ONE_ compared entry methods head to head. Single key manual entry produced [0.37 errors per thousand fields](https://pmc.ncbi.nlm.nih.gov/articles/PMC3320865/), while double entry, where a second person keys the same data so the system can flag every disagreement, cut that to 0.046 per thousand. That is 0.037% for single entry and 0.0046% for double entry. Automated capture matched double entry exactly. High confidence

Read those together and you get the honest picture. The error rate for a single field is not one number, it is a range, and where you land in that range depends almost entirely on the conditions: how tired the person is, how complex the document is, how many interruptions there were, and whether anything checks the work afterward.

### Now look at where those numbers came from

Both studies were run in about the most favourable conditions a person can key data in. Trained staff doing one repetitive task they perform every day. Purpose built forms with consistent field order. Validation rules catching impossible entries. Quiet rooms. And even there, under all of that, the error rate never reached zero.

Construction procurement has almost none of those controls, and the gap is not small.

#### Conditions in the studies

- Trained staff doing one repetitive task daily

- Standardised forms, consistent field order

- Validation rules catching impossible entries

- A second person re-keying to catch disagreements

- A quiet room, few interruptions

- Verification happening immediately

#### Conditions on your job

- A coordinator or super doing it as the fourth task of the day

- Every supplier formatting quotes and slips their own way

- The same material carrying three part numbers at three vendors

- Nothing checking the entry at all

- A trailer, a truck cab, or the middle of a phone call

- Verification weeks later, if it happens

Most importantly, the control that mattered most in the research is the one construction almost never has. The _PLoS ONE_ result was not really a finding about typing, it was a finding about verification. The reason double entry beat single entry by a factor of eight is that something independent checked the work immediately. In a manual procurement chain, nothing does. The check, if it happens at all, happens weeks later when an invoice does not match an order, and by then the error has stopped being a typo and started being a dispute.

We are not going to invent a construction specific error rate, because nobody has credibly measured one and we are not going to be the first to pretend otherwise. But the direction is not ambiguous. Every condition the research identifies as driving the rate upward, meaning document complexity, unfamiliar formats, interruption, fatigue, secondary duty, and absent verification, is standard in this industry.

## Then it compounds

A 1% error rate sounds like a rounding error. For one field it is. A purchase order does not have one field.

Walk the chain. Somebody prices the material and keys the line items into a purchase order. Somebody issues it. A truck arrives and somebody records what came off it. An invoice arrives and somebody keys that. Somebody compares all three. Somebody codes the result to a job and a cost code. Every one of those steps involves fields, and every field is a fresh roll of the dice.

The arithmetic is unforgiving. The chance of a clean document is the chance that every field is right, multiplied together. This is the finding Raymond Panko has documented across decades of human error research: individual accuracy is high, but accuracy across a chain of dependent steps is not, and people are far worse at catching their own errors than at avoiding them. High confidence

#### Run it on your own numbers

Move the sliders. This is the chance that at least one field is wrong somewhere in a single order, from the purchase order through the vendor invoice to the job coding.

33 %

Roughly one order in three carries at least one bad field by the time it reaches your books.

Fields keyed by hand across the order **40**

Error rate per field **1.0%**

This is a QuoteToMe model, not a measured finding. It applies published per field error rates to a typical order. It opens at 1% rather than the 0.0046% the laboratory achieved with double entry, because a trailer on a Friday afternoon is not a laboratory. For construction, 1% is very likely the generous end.

Fields keyed by hand At 0.5% per field At 1% per field At 3% per field

20 fields 10% 18% 46%
40 fields 18% 33% 70%
60 fields 26% 45% 84%
100 fields 39% 63% 95%

QuoteToMe model. Per field rates drawn from the published research cited above. This is applied arithmetic, not a measured industry finding, and we label it that way deliberately. Modeled

## Where the errors surface

An error that stays hidden is not the expensive kind. The expensive kind is the one that surfaces at the worst possible moment.

Accounts payable has a name for this and a number attached to it. Ardent Partners, which has run an annual benchmarking study of accounts payable for nearly two decades, puts the [average invoice exception rate at 14%](https://ardentpartners.com/ap-metrics-that-matter-in-2025/). Top performing teams hold it to 9%. Everyone else averages 22%, which is better than one invoice in five stopping dead. High confidence

The more telling finding is what practitioners now say their biggest problem is. In the 2024 study, invoice exceptions ranked as the number one accounts payable challenge for the first time in nineteen years of the survey, named by 53% of respondents. Not volume. Not fraud. Not headcount. The invoices that do not match.

53%

Of accounts payable teams now name invoice exceptions as their single biggest challenge, ranking first for the first time in nineteen years of the study.

Source: [Ardent Partners, The State of ePayables 2024](https://ardentpartners.com/ap-metrics-that-matter-in-2025/)

## What the stoppage actually costs

The gap between the best and the rest is not explained by wage rates. It is explained by touches.

Ardent Partners benchmarks the all buyer average cost to process a single invoice at $9.40. Top performing teams do it for $2.78. The organizations outside that top tier average $12.88, roughly four and a half times what the leaders pay for the same piece of work. High confidence

A clean invoice gets looked at once. An exception gets looked at repeatedly, by more than one person, often across days, usually with a phone call to a vendor somewhere in the middle. The Hackett Group reported in November 2025 that teams running manual accounts payable spend roughly 75% of their time on data capture and matching, which is to say three quarters of a salaried role spent moving numbers between documents and reconciling the disagreements. Medium confidence

Cross reference

#### Why this article says $9.40 and another of ours says $2.07

In _The State of Procurement in Construction_ we cited APQC's top quartile threshold of roughly $2.07 per invoice. Here we cite Ardent Partners at $9.40. Both are correct and they are not in conflict.

The difference is what each number measures. APQC's $2.07 is a top quartile threshold, meaning the level the best performing quarter of organizations beat. Ardent's $9.40 is an all buyer average across the full sample, and their own best in class figure is $2.78, which sits much closer to APQC's. Whenever you see a cost per invoice figure anywhere, the first question to ask is which tier it describes, because the spread between the top and the bottom is wider than the spread between industries.

[Read: The State of Procurement in Construction](https://quotetome.com/blog/state-of-construction-procurement)

## What never shows up at all

Exceptions are the errors that announce themselves. The harder category is the spend that never entered the process.

Nothing can flag a document that was never created. Ardent Partners puts maverick spend, meaning purchasing that happens outside the approved process, at [roughly 30% of total spend](https://www.esker.com/blog/source-to-pay/what-is-maverick-spend/) for the average organization. Medium confidence The Hackett Group has found this leakage can erode up to 16% of the savings a procurement team negotiated, and, importantly, that most organizations attribute it not to people deliberately going around the rules but to the absence of an easy way to follow them. Medium confidence

On a jobsite, that is a company card at a supply counter at 7am, because waiting for a purchase order means the crew stands around.

Then there is the money that leaves twice, or leaves for the wrong amount. APQC benchmarking places duplicate and erroneous payments at between 0.8% and 2% of total disbursements. Medium confidence On a contractor spending four million dollars a year on materials, the bottom of that range is thirty two thousand dollars. The top is eighty thousand.

#### A gap worth naming

We went looking for a credible published rate for how often a delivery quantity disagrees with the purchase order and the packing slip, which is the single most common source of a construction matching failure. It does not exist in any independent, methodologically transparent form. Plenty of vendors publish numbers; none of them disclose how they were measured. So we are not going to put one here. What the operations literature does establish is that top quartile receiving accuracy sits above 99% while typical performance sits several points below it, and that the gap between those two is almost entirely a documentation problem rather than a warehouse problem.

## Why this bites harder in construction

Every industry has accounts payable. Construction has accounts payable sitting on top of progress payments, retainage, and thin margins.

Start with the cash. In British Columbia, the 2025 BC Construction Association industry survey of 858 employers and tradespeople found that 91% of employer respondents had been paid late at least once in the past year for completed work, and 69% had not been paid at all at least once. High confidence That is the environment a stalled invoice lands in. When an invoice sits in exception for two weeks, it is not an administrative delay, it is two weeks of a subcontractor's payroll financed by the subcontractor.

91%

Of surveyed BC construction employers were paid late at least once in the past year for completed work. 69% were not paid at all, at least once.

Source: BC Construction Association, 2025 BC Construction Industry Survey, 858 respondents

This is well understood at the policy level, which is why the law changed. A 2015 Canadian Construction Association survey, cited by the federal government when it justified prompt payment legislation, found roughly $46 billion in payments outstanding past the conventional 30 day period, about 16% of the estimated $285 billion in Canadian construction contracts that year. High confidence That figure is a decade old now and we cite it only as the history it is. What matters today is the legislation it produced: Ontario, Alberta, and federal construction work now run on statutory payment clocks of 28 days, and British Columbia's prompt payment legislation received Royal Assent in November 2025.

Statutory clocks change the stakes of a matching error completely. When payment timing was a matter of custom, a slow invoice was an annoyance. When it is a matter of statute, an invoice that cannot be matched is an invoice that cannot be certified, and the clock does not care why.

An invoice that cannot be matched is an invoice that cannot be certified. The clock does not care why.
What changes when payment timing becomes a matter of statute

### And then there is the time

The construction specific research on wasted effort is thinner than the accounts payable research, and the best available studies are industry surveys rather than measured cost data, so we flag them as such. Autodesk and FMI surveyed more than 3,900 construction professionals for _Harnessing the Data Advantage in Construction_ in 2021 and estimated that poor data may have cost the global construction industry $1.85 trillion in 2020, with bad data implicated in somewhere between 14% and 16% of all rework worldwide. Medium confidence An earlier survey of roughly 600 industry leaders, _Construction Disconnected_, put the share of time spent on non optimal activities at about 35%, or roughly 14 hours per person per week, of which about five and a half hours went to simply hunting for project information. Medium confidence

Treat those as directional. They are self reported extrapolations, not audited figures, and we would not hang a business case on them. But the direction is consistent with everything above, and it matches what anyone who has worked a project controls role already knows: a meaningful slice of the week goes to finding out what was ordered, what arrived, and what it cost.

Cross reference

#### How this fits with our 27 minute figure

In _The 27-Minute Tax_ we put the full cycle for one purchase order, from creating it through to a coded bill, at about 27 minutes, with a clean order moving in 13 and a messy one passing 54. Elsewhere we have cited independent research putting a manual purchase order at 8 to 12 minutes, and broader cross industry work putting it as high as 90.

Those three numbers measure three different things, and the scope is the whole explanation. The 8 to 12 minute figure covers creating the purchase order only. Our 27 minutes covers the entire chain through matching and cost coding, which is why it is larger. The broader 30 to 90 minute range is a cross industry span that includes far more complex procurement environments than a contractor's. When you compare procurement time figures from any source, check the start and end points before you compare the numbers.

[Read: The 27-Minute Tax](/blog/27-minute-tax)

## Why catching it late is the expensive part

An old principle from quality management, and the one piece of this we are citing as a rule of thumb rather than a measurement.

It is usually written as one, ten, one hundred. Preventing a bad record at the point of entry costs one unit. Correcting it later costs about ten. Letting it through to become a real world failure costs about a hundred. It is generally credited to Labovitz and Chang in 1992, building on the cost of quality work of Crosby and Feigenbaum before them. Heuristic

We are citing it as what it is, which is a heuristic rather than a measurement. Nobody ever ran a controlled study and got exactly one, ten, and one hundred. But the shape of the curve is well supported, and the shape is the point. It costs almost nothing to get a line item right when the person ordering it is standing in front of the supplier. It costs somebody's afternoon to work out why an invoice does not match six weeks later. And it costs real money when the answer arrives after the job closed and the number is already in the final cost report.

We process a high volume of purchase orders, and the Scan Quote to Purchase Order feature has significantly reduced PO processing time while minimizing human error.
**Leo Rizzo**, Purchasing Manager, Monterey Mechanical Co.

## What good looks like from here

The benchmarks give a clear picture of the destination, because the top tier is measured separately from everyone else in the same study.

Best in class accounts payable teams hold exceptions to 9% instead of 22%, clear an invoice in 3.1 days instead of 17.4, pay $2.78 to process one instead of $12.88, and push 49.2% of invoices straight through without a human touching them at all.

None of that is achieved by keying more carefully. It is achieved by keying less. The single highest leverage move in the whole chain is the one the _PLoS ONE_ researchers found almost by accident: automated capture matched double entry for accuracy, which means the reliable way to stop transcription errors is to stop transcribing.

1

### Capture the price once, at the source

The quote becomes the order without anyone retyping it, so the numbers on the purchase order are the numbers the vendor actually gave you.

2

### Record what arrived, where it arrived

The delivery is confirmed on site, against the order, by the person looking at the pallet, rather than reconstructed from a packing slip two weeks later.

3

### Let the invoice check itself

Quote, order, delivery, and invoice are compared automatically, so an exception is flagged the day it appears instead of at month end.

4

### Code it once and send it on

The job, the cost code, and the vendor carry through to the accounting system without a second round of typing.

Do that, and the quiet failure stops being quiet. You find out about a price that crept, a quantity that came up short, or a charge against the wrong job on the day it happens, while it is still a conversation with a vendor rather than a line in a closed cost report.

Field Approved

## Frequently asked questions

### How often does manual data entry produce errors?

Measured rates range from roughly 0.0046% per field under double entry conditions to between 3% and 5% for complex or fatigued work, with a peer reviewed clinical study measuring 0.83% per keystroke. The rate for any single field is small. The odds compound across a document, because a purchase order, an invoice, and a cost coding pass together involve dozens of fields.

### What percentage of invoices fail to match?

Ardent Partners benchmarks the average invoice exception rate at 14%. Top performing teams run 9%, the remaining organizations average 22%. In the 2024 study, exceptions were named the number one accounts payable challenge for the first time in nineteen years, by 53% of respondents.

### What does it cost to process one invoice manually?

The all buyer average is $9.40. Top performers process one for $2.78, the rest average $12.88. The gap is driven mostly by how many invoices need a human to handle them more than once.

### How long does a manual invoice take to clear?

The average is 9.2 days. Top performers clear one in 3.1 days, while the bottom tier averages 17.4 days, meaning committed cost sits invisible for more than two weeks before anyone can act on it.

### Why does a matching error matter more in construction?

Because construction runs on progress payments and thin margins, so a stalled invoice stalls the cash behind it. In British Columbia, 91% of surveyed employers reported being paid late at least once in the past year for completed work and 69% reported not being paid at all at least once. Statutory prompt payment clocks of 28 days now apply in Ontario, Alberta, and on federal work, so an invoice that cannot be matched is an invoice that cannot be certified.

### Is a three way match enough?

A three way match compares the purchase order, the receiving record, and the invoice. It catches quantity and receipt disagreements but it does not check any of them against the price you were originally quoted. Adding the quote as a fourth document closes that gap, which is where price creep between quoting and invoicing tends to hide.

### A note on the numbers

We flag every statistic with our confidence in it. High confidence means the figure comes from a primary source with a disclosed methodology, such as a peer reviewed journal, an independent analyst benchmark, or a government publication. Medium confidence means the direction is well established but the figure is older, self reported, or drawn from a survey rather than measured data. Where a number is our own applied model rather than a research finding, we label it modeled and show the arithmetic.

Several statistics that circulate widely in this subject area were deliberately left out, because we could not trace them to a primary source with a disclosed methodology, or because newer research contradicts them. These include the frequently quoted cost to resolve a single invoice error, the claim that 39% of manually processed invoices contain an error, the $3.1 trillion estimate of the cost of bad data to the US economy, and the commonly cited construction rework percentage, which is the subject of genuine and unresolved disagreement between self reported industry estimates and newer measured data. We would rather publish fewer numbers and stand behind all of them.

## Sources

Every stat above, and where to read it for yourself.

- **Journal of the American Medical Informatics Association**, [Measuring the rate of manual transcription error in outpatient point of care testing](https://pmc.ncbi.nlm.nih.gov/articles/PMC6913214/) (2019). Per keystroke transcription error of 0.83%, with related laboratory studies at 1.14% and 3% to 5%.

- **PLoS ONE**, [Quality of Data Entry Using Single Entry, Double Entry and Automated Forms Processing](https://pmc.ncbi.nlm.nih.gov/articles/PMC3320865/) (2012). Single key entry at 0.37 errors per thousand fields; double entry and automated capture both at 0.046.

- **Raymond R. Panko**, human error research compiled across multiple papers including _What We Don't Know About Spreadsheet Errors Today_, European Spreadsheet Risks Interest Group (2015). Per action accuracy is high while accuracy across chains of dependent steps degrades, and people detect their own errors poorly.

- **Ardent Partners**, [The State of ePayables 2024, reported in Accounts Payable Metrics That Matter in 2025](https://ardentpartners.com/ap-metrics-that-matter-in-2025/). Survey of 212 accounts payable professionals. Cost per invoice $9.40 average, $2.78 best in class, $12.88 all others. Exception rate 14% average, 9% best in class, 22% all others. Cycle time 9.2 days average, 3.1 best in class, 17.4 all others. Touchless processing 32.6% average, 49.2% best in class. Exceptions named the top challenge by 53% of respondents.

- **The Hackett Group**, accounts payable research (November 2025). Teams running manual accounts payable spend approximately 75% of their time on data capture and matching.

- **Ardent Partners**, maverick spend benchmark of approximately 30% of total spend for the average organization, [reported via Esker](https://www.esker.com/blog/source-to-pay/what-is-maverick-spend/). The Hackett Group (2019) separately found maverick spend can erode up to 16% of negotiated savings.

- **APQC**, Open Standards Benchmarking. Duplicate and erroneous payments represent between 0.8% and 2% of total disbursements.

- **BC Construction Association**, 2025 BC Construction Industry Survey and Spring 2025 Stat Pack (April 2025). Survey of 858 employers and tradespeople across seven development regions. 91% of employer respondents paid late at least once in the past year for completed work; 69% not paid at all at least once.

- **Canadian Construction Association** 2015 survey, cited in the Regulatory Impact Analysis Statement, _Canada Gazette, Part II_, Vol. 157, No. 26 (December 20, 2023). Approximately $46 billion unpaid past 30 days, about 16% of an estimated $285 billion in Canadian construction contracts that year. Cited as historical context for prompt payment legislation, not as a current figure.

- **Prompt payment legislation.** Ontario _Construction Act_ in force October 2019; Alberta _Prompt Payment and Construction Lien Act_ in force August 2022; federal _Prompt Payment for Construction Work Act_ in force December 2023; British Columbia prompt payment legislation received Royal Assent November 2025. Statutory owner to contractor payment period of 28 days.

- **Autodesk and FMI**, _Harnessing the Data Advantage in Construction_ (2021). Survey of more than 3,900 construction professionals. Poor data may have cost the global construction industry $1.85 trillion in 2020, with bad data implicated in 14% to 16% of global rework.

- **PlanGrid and FMI**, _Construction Disconnected_ (2018), [reported via USGlass](https://www.usglassmag.com/report-time-spent-on-non-optimal-activities-costs-the-construction-industry-billions-annually/). Survey of approximately 600 construction leaders. About 35% of time, or roughly 14 hours per week per person, spent on non optimal activities, including about 5.5 hours hunting for project information.

- **George Labovitz and Yu Sang Chang**, _Making Quality Work: A Leadership Guide for the Results-Driven Manager_ (1992), building on cost of quality work by Philip Crosby and Armand Feigenbaum. The one, ten, one hundred escalation is an illustrative heuristic and has never been empirically measured as a precise ratio.

## Stop retyping, and the errors go with it

QuoteToMe takes manual data entry out of the buying chain. The quote, the purchase order, the delivery confirmation, and the vendor invoice are checked against each other automatically, a four way match rather than the usual three, so a price that crept or a quantity that came up short is caught the day it appears instead of at month end.

The matched result flows straight into the accounting system or construction ERP your office already runs on, coded to the right job, with the committed cost visible before the invoice ever lands. What your team notices is simpler than that: nobody is retyping anything, and the numbers agree.

[Book a Demo](https://quotetome.com/book-a-demo)

From one builder to another.