Lyzr Research · October 2026 · A 35 minute read
Contract-to-Pay
Leakage.
Recovering spend with agentic validation. Procurement negotiates the value and accounts payable pays the invoice, but nobody checks one against the other.
To the CPO and the CFO, who should read this together
You negotiated the savings. Did you collect them?
Procurement negotiates the value.
Accounts payable pays the invoice. In between, nobody compares one against the other line by line.
So part of every saving leaks away.
WorldCC finds contracts deliver 8.6% less than expected on average. Retailers measure shrink to the decimal, and almost no finance team measures contract shrink at all.
Checking every invoice is now cheap enough.
Sampling was the only affordable control. An agent can read the contract and every line of every invoice before payment.
This is the argument
This is a report about money. It starts after the contract is signed, follows one dollar to payment, sizes eight kinds of leakage and shows how to measure your own. Part 7 describes Lyzr and is labeled as Lyzr’s.
Lyzr Research, October 2026
The short version.
- The shrinkRetailers count shrink to the decimal, 1.6% of sales. WorldCC finds contracts deliver 8.6% less than expected, and few finance teams measure it.
- The gapProcurement negotiates the value. AP pays the invoice. Nobody checks one against the other line by line.
- The eight leaksOff-contract pricing, missed tiers, missed discounts, duplicates, surcharges, renewals, service credits and spend outside contracts.
- The checkAn agent reads the contract once and every line of every invoice before payment, instead of a sample after it.
- The sizeOur planning range is 1.3% to 7.5% of addressable spend, and the middle case is 3.2%. Measure your own.
- The partnersFinance is the buyer. Procurement is the partner. Audit sees the evidence.
- Where Lyzr fitsTwo agents from Lyzr’s playbook apply here. Part 7 links to the playbook for the full roadmap.
What this covers
Seven parts, start to finish.
Why it hides
The shrink
Retailers count it to the decimal. Finance mostly does not.
What leaks
The eight leaks
Each leak, its evidence and the check that catches it.
Every invoice
Checking everything
Why samples miss, and what an agent checks line by line.
How much
Sizing it
Planning ranges, a calculator and the back-test.
Who owns what
By seat
The CFO, the CPO, the controller, AP and audit.
Predictions
What next
What we expect, and the strongest argument against.
Two agents
Where Lyzr fits
What applies, a first step, and the playbook for the roadmap.
WorldCC, Contract Management, August 2025
On average, contracts deliver 8.6% less value than was expected when they were signed.
Average value erosion, the gap between expected and realized contract outcomes
The best performers hold it to about 3%. In complex environments it reaches 15% or more.
A gap almost nobody measures
World Commerce and Contracting, Contract Management whitepaper, August 2025.
The argument in one picture
Retailers count shrink to the decimal. Almost nobody counts contract shrink.
A retailer knows its shrink every month. A finance team rarely knows how much of a negotiated saving survives to payment. Procurement measures the saving agreed. AP measures the invoice paid. The difference is the shrink, and it sits between two functions.
The shrink.
What retailers measure and finance does not.
This part sets out the metaphor, the evidence that contract value erodes after signature, and why the gap between negotiation and payment has gone unchecked.
Instrument 01 of 03 · A model worth borrowing
Shrink
Retail loss prevention
What is not counted cannot be recovered.
Why it matters here
Retailers track inventory shrink to a tenth of a percent. NRF’s survey puts the average at 1.6% of sales, which was $112.1 billion in fiscal 2022. WorldCC puts average erosion of contract value at 8.6%. The bases differ, but the contrast is plain. One loss is counted shelf by shelf, and the other is rarely counted at all.
NRF, National Retail Security Survey 2023. WorldCC, Contract Management whitepaper, August 2025.
WorldCC, Contract Management, August 2025
The best performers lose about 3% of contract value, and the worst lose 15% or more.
Value erosion among the best performers, against 15% or more in complex environments
The spread matters more than the average. The same kind of contracts, run with discipline, lose a fifth as much.
Discipline is the difference
World Commerce and Contracting, Contract Management whitepaper, August 2025.
Follow one dollar
One dollar, five steps, and one comparison that nobody makes.
Three-way match compares the invoice with the purchase order and the receipt. The PO price was usually keyed from the contract once, by hand, at the start. Rebate tiers, discount windows, renewal terms and service credits never reach the PO at all, so nothing downstream ever checks them.
Why the gap exists
Three reasons nobody checks.
Procurement and AP
One negotiates and one pays. Each assumes the other has checked.
Contracts live elsewhere
WorldCC finds contract data spread across 24 systems on average, and rarely inside the one that pays.
Sampling was affordable
Reading every contract against every invoice took people, so audits looked at a few.
Ardent Partners, State of ePayables 2025
Processing one invoice costs $9.84 on average, so nobody checks each one against its contract by hand.
Average all-inclusive cost to process a single invoice, against $2.65 for best-in-class teams
Adding a manual contract check to every invoice was never going to happen. That is why checks are sampled and happen after payment.
Hand checks do not scale
Ardent Partners, State of ePayables 2025.
The change
What has changed.
An agent can read a contract and extract its terms in seconds.
The same agent can read an invoice line by line against those terms.
The check can run before payment, on every invoice, not after payment on a sample.
A person still approves every hold, dispute and claim.
The eight leaks.
Where the shrink goes.
Eight kinds of leakage account for most of what slips between a negotiated contract and a paid invoice. Each has evidence, a check that catches it and a person who decides.
Eight leaks, and where each hides.
Leak 1 of 8, off-contract pricing
Off-contract pricing is a price keyed once and billed thousands of times.
The contract rate is keyed into the system once, by hand, and suppliers bill from their own price lists. A forty cent difference on a unit bought 12,000 times a year is $4,800. Contract audits tend to find it. SC&H reports that its clients recover 2% to 4% of audited spend on average, and 10% to 20% on some supplier relationships.
Leak 2 of 8, missed tiers and rebates
Volume tiers and rebates are earned at thresholds nobody watches.
A rebate often applies only when cumulative spend crosses a threshold. Spend is spread across invoices, sites and months, so the total is rarely tracked against the tier. Unapplied supplier credits are a known recovery source in rebate-heavy industries. An agent keeps the running total for every supplier and flags the one that stopped just short.
Leak 3 of 8, missed early-payment discounts
A 2/10 net 30 discount is worth 36.7% a year, and slow approvals forfeit it.
Taking 2% for paying twenty days early equals 2 over 98, times 365 over 20, or 36.7% annualized. The discount only exists if the invoice is approved inside the window, and an approval cycle longer than ten days makes it impossible. An agent reads the terms when the invoice arrives and routes it for speed.
Leak 4 of 8, duplicate payments
Duplicate payments hide behind a different reference.
APQC benchmarks put duplicate and erroneous payments at 0.8% to 2.0% of disbursements. Recovery audits reclaim 0.05% to 0.5% of audited spend after the money has gone. Duplicates rarely match exactly. They differ by a character, a date or a suffix. An agent compares supplier, amount, purchase order and content, and holds the second invoice before it is paid.
Leak 5 of 8, unauthorized surcharges and fees
Unauthorized surcharges and fees are small, plausible and everywhere.
Accessorial charges such as liftgate, detention and residential delivery are consistently the largest source of freight overcharges, because they are easy to add and hard to verify. Estimates of invoice error rates vary, and one advisory firm puts them at 5% to 10% of freight invoices, mostly in the carrier’s favor. An agent compares every line with the fees the contract allows.
Leak 6 of 8, auto-renewals at an uplift
Auto-renewals at an uplift are decided before anyone looks.
Zylo finds that 79% of IT leaders met a price increase at renewal, and a typical company faces 211 renewals a year. Vertice puts average SaaS inflation at 12.2%. A notice window of 30 to 90 days closes long before the renewal date. An agent tracks every notice date and holds the renewal invoice when the uplift exceeds the contract cap.
Leak 7 of 8, unclaimed service credits
Service-level credits go unclaimed because they are rarely automatic.
Most service credits must be claimed, with your own logs, inside a short window. Microsoft’s documentation says providers do not usually apply them automatically, and AWS requires a claim by the end of the second billing cycle after the incident. No reliable benchmark exists for how many go unclaimed, which is part of the problem. An agent compares service logs with the SLA and drafts the claim.
Leak 8 of 8, spend outside any contract
Spend outside any contract buys at whatever price the supplier names.
Hackett finds that off-contract buying costs organizations up to 16% of their negotiated savings. Its leaders reach 91% on-contract compliance, against 74% for typical peers. An agent cannot stop every purchase, but it can match every paid invoice to a contract and show which categories escape one.
What the checks need.
Checking everything.
Why samples miss, and what an agent checks.
A sample finds a share of the leaks equal to the share it samples. This part sets out why, borrows two ideas on sampling, shows what an agent checks line by line, and sets the guardrails.
Sampling
A 5% sample finds about 5% of the leaks.
If leaks are scattered at random, a sample finds them in proportion to its size. Auditors aim samples at large invoices and risky suppliers, which helps, and still leaves most of the population unread. Checking every invoice changes what is found, and when.
Instrument 02 of 03 · A model worth borrowing
Wald’s bullet holes
Abraham Wald, Statistical Research Group, 1943
The damage you can see is not where the damage is.
Why it matters here
Wald studied bombers that returned with bullet holes and advised armoring the places with no holes, because planes hit there did not come back. Invoice checks have the same blind spot. You only find the leaks in the invoices you looked at, and every sample is chosen by what you already suspect.
Abraham Wald, Statistical Research Group, Columbia University, 1943, as told by Mangel and Samaniego, 1984.
Instrument 03 of 03 · A model worth borrowing
The law of small numbers
Amos Tversky and Daniel Kahneman, 1971
Small samples are far less reliable than they feel.
Why it matters here
Tversky and Kahneman showed that even trained researchers expect a small sample to look like the whole population. A 5% sample of invoices can show no leakage or 9% by chance, and either answer feels like evidence. Checking every invoice replaces an estimate with a count.
Tversky and Kahneman, Belief in the Law of Small Numbers, Psychological Bulletin, 1971.
What the agent does
An agent reads the contract once and checks every line of every invoice.
The agent extracts the terms from the contract once, price, tiers, payment terms, allowed fees, renewal and service levels. It then reads each incoming invoice line by line against those terms and against payment history, before the invoice is approved. Clean lines pass. Anything else goes to a person with the evidence attached.
A sample, and every invoice.
Guardrails
Guardrails that keep it safe.
Read-only access to the ERP and the contract store.
A person approves every hold, dispute and claim.
Materiality thresholds, so small differences do not flood the queue.
A record of what was checked and why, on every invoice.
Sizing it.
How much leaks, and how to find out.
The honest answer is that nobody knows until they check. This part gives planning ranges for each leak, a calculator you can adjust, and the back-test that replaces estimates with your own number.
What each leak is worth, as a planning range.
Our planning ranges, as a share of addressable spend, to argue with. Replace them with your own back-test.
Try it on your own spend
Size your shrink.
Pick your spend and a case. It is a first estimate to start a conversation, not a finding.
Uses the planning ranges on the previous slide. Nothing you select here is sent anywhere.
Your estimated contract shrink, per year
About 3.2% of addressable spend, in the middle case.
Planning ranges are ours, anchored on published figures where they exist.
Added up
Added up, our range runs from 1.3% to 7.5% of spend.
The middle case, 3.2% of spend, is about what WorldCC’s best performers still lose. The high case, 7.5%, sits just under WorldCC’s 8.6% average erosion, which also counts delivery and scope losses we do not size. The ranges are planning tools. The back-test is the measurement.
The sizing method
Replace the estimate with a count.
A category and a year
Choose a spend category with contracts on file and twelve months of invoices.
Read-only
Contracts and invoices are read without changing your systems or how anyone pays.
The eight checks on every line
The agent reads each invoice against its contract and logs every difference.
Review with procurement and AP
People confirm the findings, so the number can be defended to the CFO.
Who owns what.
Finance buys, procurement partners.
Shrink sits between functions, so no one owns it by default. This part sets out what each seat does and what to show the others.
The CFO owns the number.
The CPO owns the contract and the partnership.
The controller owns the evidence.
The AP lead owns the hold.
Internal audit owns the independent view.
What comes next.
Five predictions, and the strongest argument against.
Predictions are cheap, so each one comes with what would prove it wrong and what to watch for. The strongest argument against this report comes right after, because leaving it out would be dishonest.
Looking forward
Five things we expect to be held to.
A date
Each prediction names the year we expect it by.
A confidence
A rough number, so you can hold us to it.
Signals to watch
What would tell you early that it is happening.
What would prove us wrong
Written down before the fact.
Prediction 01 of 05 · Lyzr’s view
Checking every invoice against its contract becomes a standard control at close.
Why we believe it
Finance wants evidence at close, and the cost of checking every invoice has fallen below the cost of the leaks.
What would prove us wrong
If recovery audit firms keep offering low-effort contingency services, finance may stay with periodic audits.
Signals to watch
- Close checklists add a contract check
- Controllers ask for coverage by category
- AP vendors add contract validation
Prediction 02 of 05 · Lyzr’s view
Early-payment discount capture becomes a finance KPI beside days payable outstanding.
Why we believe it
A return of 36.7% a year is hard to ignore once approvals are fast enough to earn it.
What would prove us wrong
If treasury keeps preferring longer payment terms, discounts stay optional.
Signals to watch
- Capture rate reported to the CFO
- Approval time tracked against discount windows
- Suppliers offer dynamic discounts
Prediction 03 of 05 · Lyzr’s view
Auto-renewal notice dates move from contract repositories into the payment workflow.
Why we believe it
A reminder in a repository is easy to miss, and a hold on the renewal invoice is not.
What would prove us wrong
If vendors drop auto-renewals under buyer pressure, the problem shrinks by itself.
Signals to watch
- Renewal holds appear in AP tools
- Notice dates sit beside invoice dates
- Buyers insist on uplift caps
Prediction 04 of 05 · Lyzr’s view
Auditors begin to expect full coverage on high-volume supplier categories.
Why we believe it
When checking everything is cheap, sampling is harder to defend.
What would prove us wrong
If standards do not move, sampling stays acceptable.
Signals to watch
- Audit plans cite coverage, not samples
- Committees ask why a category was sampled
- Auditors use agents themselves
Prediction 05 of 05 · Lyzr’s view
Suppliers start sending invoices with machine-readable contract references.
Why we believe it
Suppliers want clean validation, because it gets them paid faster.
What would prove us wrong
If every ERP vendor builds a proprietary format, adoption stalls.
Signals to watch
- A shared reference standard gains traction
- Large buyers require contract IDs
- Suppliers market faster payment
The strongest argument against this report
Recovery audit firms already do this, on contingency.
The argument
Recovery auditors review payments and reclaim leaks for a share of what they find, with no cost to you up front.
Where it is right
They are good at finding old leaks, and a first back-test may overlap with one.
Where it stops
They look back, often two to four years, after the money has gone. They do not hold an invoice before it is paid, and recovery depends on the supplier agreeing.
What would change our mind
Evidence that recovery after payment matches prevention before it, at the same total cost.
Where Lyzr fits.
Two agents, and the playbook for the roadmap.
This part describes Lyzr from Lyzr’s own pages. It applies two agents from Lyzr’s procurement playbook to this problem, and points to the playbook for the full source-to-pay roadmap, which this report does not repeat.
From Lyzr’s procurement playbook
Two agents apply to contract-to-pay.
Validates spend against contract terms
Lyzr describes it as catching wrong prices and missed negotiated savings by checking spend against the contract.
Matches invoices to orders and receipts
Lyzr describes two-way and three-way matching, with exceptions routed to people.
For the full roadmap
The Strategic Procurement Automation Playbook covers source-to-pay, which this report does not repeat.
Lyzr, The Strategic Procurement Automation Playbook, lyzr.ai/playbook, read October 2026. Vendor descriptions, not independently verified here.
An example run
An example of what a first run looks like.
Lyzr’s pricing page shows an illustrative Agent Test Drive on invoice matching. The agent read 214 invoices, matched 208, flagged six and drafted four replies that wait for finance approval, and none went out without approval. It shows the shape of a run. It is not a result from your data.
Where Lyzr fits, and where to look elsewhere.
A first step, a back-test.
About Lyzr
The full-stack agent infrastructure platform for enterprises and governments that want to own their intelligence.
Founded in New York in 2023. Agent Studio, Architect, Opencontroller, Open Cowork, Openfinetuner, Optimus and Madison cover building, governing, running and owning agents. Deploys in your AWS, Azure or GCP account, a dedicated tenant, SaaS, on bare metal, fully air-gapped, or on the Optimus appliance.
Backed by Rocketship.vc, BGV, Accenture Ventures, GFT, the Partnership Fund for New York City and Plug and Play.
Sources and notes
Every fact, and where it came from.
- World Commerce and Contracting, Contract Management whitepaper, August 2025.
- NRF, National Retail Security Survey 2023, shrink of 1.6% and $112.1 billion.
- Ardent Partners, State of ePayables 2025, cost per invoice.
- APQC duplicate and erroneous payments, as cited by Corpay.
- Recovery audit rates of 0.05% to 0.5%, Transparent Global.
- SC&H, contract compliance audit recoveries of 2% to 4% of audited spend.
- Corpay, early payment discounts and the 36.7% annualized return.
- Zylo and Vertice renewal data, as compiled in Contract renewal statistics 2026.
- Freight invoice error estimates, including Tompkins Ventures. The widely repeated rates have no dated original.
- Microsoft Learn, service level agreements and credits.
- Hackett maverick spend findings, as summarized by Art of Procurement.
- Lyzr, The Strategic Procurement Automation Playbook. Vendor claims, not independently verified here.
- Lyzr pricing page, the example Agent Test Drive, October 2026.
Statistics carry their source. Secondary and vendor figures are labeled, and where no reliable benchmark exists the report says so.
Sources and notes
The ideas, and who had them.
- Retail loss prevention and the shrink metric, NRF.Count it to the decimal
- Abraham Wald, Statistical Research Group, 1943.You only see the holes that came back
- Amos Tversky and Daniel Kahneman, Psychological Bulletin, 1971.Small samples feel like evidence
- Annualizing a cash discount, standard treasury practice.2/10 net 30 is worth 36.7% a year
The eight-leak frame, the planning ranges, the calculator and the first step are our own, offered to be argued with.
Your next step
Bring a year of invoices.
Leave with your own shrink number. Thirty minutes, no slides, and an honest view of what a back-test would find.
01
Pick it
We choose one category with contracts on file.
02
Run it
We read a year of invoices against those contracts, read-only.
03
Count it
We size what leaked and what is recoverable.