Invoice Matching Tolerances: Setting Price and Quantity Thresholds
Every accounts payable team that runs purchase-order matching eventually hits the same wall: the invoice and the PO almost never agree to the penny. A supplier rounds freight differently. Sales tax shifts because a line shipped from a different warehouse. A field crew accepted 47 sheets of plywood against a PO for 50. None of these are errors worth a human's time, yet a strict three-way match treats every one of them as an exception that has to be researched, explained, and cleared before the invoice can be paid.
Matching tolerances are the release valve. A tolerance is a defined band of acceptable variance — in price, in quantity, or in extended amount — within which a mismatch is allowed to pass without human review. Set them well and the trivial noise disappears while genuine overbilling still gets stopped. Set them badly and you either pay invoices you should have questioned or bury your team in exceptions they learn to wave through. This post is about getting them right for construction AP.
In a three-way match, the system compares the invoice against the purchase order and the receipt. Three things get checked: the unit price billed against the price on the PO, the quantity billed against the quantity received, and the extended line total. A tolerance defines how far each of those can drift before the line is flagged. A price tolerance of 5% means an invoiced unit price up to 5% above the PO price passes silently; a quantity tolerance of 2 units means the crew can receive two more or two fewer than ordered without tripping a hold.
Tolerances are not permission to overpay. They are a statement about which variances are economically rational to investigate. Researching a $4 freight discrepancy on a $9,000 lumber invoice costs more in AP labor than the $4 itself. The tolerance encodes that judgment so the system can make it consistently, on every invoice, without a person having to.
It is tempting to run a strict match — everything must agree exactly — on the theory that catching every variance is the safest posture. In practice, zero tolerance is one of the most reliable ways to weaken a control. When the system flags every invoice, the exception queue swells with items that are 95% noise. Reviewers cannot triage that volume carefully, so they develop a habit: open the exception, see another rounding difference, approve, move on. The dangerous variance — the one that actually matters — arrives in the same queue, wearing the same clothes, and gets the same reflexive approval.
Roughly 0 in 5
Share of invoices that hit an exception or require manual intervention in a typical AP operation (Ardent Partners)
A high exception rate is not a sign of a vigilant process. It is a sign of a process that has not decided what it cares about. The goal is not zero variance; it is a small, high-signal exception queue where every item genuinely deserves a human look. Tolerances are how you get there.
A useful gut check: if your reviewers can clear the exception queue without slowing down to think, your tolerances are too tight. The queue should feel like it requires judgment. If it feels like data entry, it has stopped being a control.
These two tolerances guard against different failure modes and should be set independently. A price tolerance protects against being billed more per unit than was agreed. A quantity tolerance protects against being billed for more units than were received. They have different natural sizes and different risk profiles.
Price variance tends to be small and one-directional. Most of it comes from commodity price movement between when a PO was cut and when the material shipped — rebar, copper, lumber, and fuel surcharges all move. A modest price tolerance, often in the 3% to 5% range for volatile materials and tighter for fixed-price contract items, absorbs that legitimately. The key control is that price tolerance should usually be one-sided: a price below the PO is not a problem and can pass freely, while a price above the PO is what you want to bound.
Quantity variance behaves differently. On a jobsite, partial deliveries are normal — a supplier ships 80% of an order now and the balance next week — and each shipment generates a receipt and an invoice. The right answer here is rarely a generous quantity tolerance; it is accurate receiving. If quantities are captured properly when material hits the site, the invoice should match the receipt closely and the tolerance only needs to cover small counting differences. A wide quantity tolerance is a way of papering over weak receiving, and weak receiving is exactly how a supplier bills you for material that never arrived.
If you are constantly widening quantity tolerances to get invoices to pass, the problem is upstream. Fix receiving — make sure field crews record what actually showed up — before you loosen the match.
A tolerance can be expressed two ways: as a percentage of the line value or as a flat dollar amount. Each breaks down at one end of the spend range, which is why mature AP setups use both at once and pass a line only if it clears the tighter of the two — or, on small lines, the looser.
A pure percentage tolerance is too permissive on large lines. Five percent of a $200,000 structural steel line is $10,000 — far too much variance to wave through. A pure absolute tolerance is too permissive on small lines and too strict in reverse: a flat $50 tolerance does nothing useful on a $40 line of fasteners. The standard pattern is a combined rule: the variance must be within X percent and within Y dollars. The percentage controls small and mid-size lines; the absolute cap controls large ones.
A combined tolerance rule in practice
- Percentage band — handles ordinary commodity drift on routine lines (for example, within 4%)
- Absolute cap — prevents a large line from passing a big-dollar variance just because the percentage is small (for example, no more than $250 regardless of percentage)
- Small-line floor — a minimum dollar threshold below which trivial variances always pass, so $3 differences never reach a human
- Pass only when the line satisfies the binding constraint — the cap on large lines, the floor on tiny ones
Get AP insights in your inbox
A short monthly roundup of construction AP + accounting posts. No spam, ever.
No spam. Unsubscribe anytime.
A single company-wide tolerance is a blunt instrument. The right variance band for ready-mix concrete is not the right band for a fixed-price equipment rental, and neither is right for a negotiated subcontractor lump sum. Segment your tolerances by category and by the risk each category carries.
How tolerance bands should vary by category
- Volatile commodities — lumber, steel, copper, fuel: wider price tolerance, because real market movement between PO and invoice is expected
- Fixed-price contract items: tight price tolerance, because the price was negotiated and should not move at all
- Catalog and consumable supplies: moderate price tolerance, tighter quantity tolerance, since these are easy to over-ship and over-bill
- Freight, fuel surcharges, and small fees: a dedicated small-dollar tolerance so these never dominate the exception queue
- High-risk or new vendors: tighter bands across the board until the vendor has a clean billing history
Vendor history is a legitimate input. A supplier with two years of clean, accurate invoices has earned a slightly wider band; a vendor onboarded last month, or one that has produced billing errors before, should be matched more strictly. This is risk-based control: spend your scrutiny where the evidence says it is warranted.
A large share of match noise has nothing to do with the goods themselves. It is tax, freight, and rounding — charges that frequently are not even on the original PO, so a strict line match has nothing to compare them against. Treating these the same as a material line guarantees a noisy queue.
Tax should be validated against the expected rate for the ship-to jurisdiction rather than matched line-for-line, because the correct tax legitimately differs from any estimate on the PO. Freight is best governed by a small standalone tolerance — a flat dollar band — since carriers rarely bill the exact figure anyone predicted. Rounding differences at the line and invoice total, typically a few cents from how a vendor's system computes extended amounts, should be absorbed by a tiny tolerance and never surfaced at all. The principle: match the goods strictly, and govern the ancillary charges with purpose-built small-dollar tolerances so they stop drowning the signal.
Tolerances are not a set-and-forget configuration. They are a hypothesis about your spend, and your spend changes — material prices move, your vendor mix shifts, receiving discipline improves or slips. Treat tuning as a recurring review, ideally quarterly.
The data to look at is your own exception history. Pull every exception from the period and sort by outcome. If a category produces a steady stream of flags that reviewers approve almost every time with no adjustment, that tolerance is too tight and is generating noise — widen it. If exceptions in a category routinely turn into invoice corrections or short-pays, that tolerance is doing real work and should stay, or tighten. The exceptions that lead to actual money recovered tell you where the band is correctly placed. A modern AP platform such as Covinly surfaces this directly — exception volume, approval rate, and recovery rate by category — so tuning is a decision grounded in evidence rather than a guess.
“We were drowning in match exceptions until we looked at what they actually were. Eighty percent were freight and tax noise. We carved those out into their own small-dollar tolerances and the real queue shrank to something a person could review properly.”
— AP Manager, mid-market general contractor
The payoff of well-set tolerances is straight-through processing. When an invoice matches its PO and receipt and every variance falls inside tolerance, there is no judgment left for a person to add — the invoice can post and move toward payment automatically, with the full match result and applied tolerances recorded in the audit trail. Human attention concentrates on the genuine exceptions: lines outside tolerance, missing receipts, no PO at all.
This only works if the tolerances are trustworthy, which loops back to everything above. Auto-approval inside a sloppy tolerance is just automated overpayment. Auto-approval inside a tolerance that has been segmented by category, capped in absolute dollars, and tuned against real exception outcomes is one of the highest-leverage moves in AP — it removes the routine work without removing the control.
Matching tolerances are where AP decides what is worth a human's attention. Zero tolerance flags everything and quietly trains the team to approve everything. The better path is deliberate: separate price and quantity, combine percentage and absolute thresholds, segment by spend category and vendor risk, give tax and freight their own small-dollar bands, and revisit the numbers each quarter against your own exception data. Do that, and clean invoices flow on their own while the variances that actually cost you money still get stopped — which is the entire point of matching in the first place.
Written by
Sarah Blake
Head of Product
Former AP Manager at a $200M construction firm, now leads product at Covinly. Writes about what AP teams actually need from automation — beyond the marketing promises.
View all posts