eProcureAI / Resources / Guide
GuideMost approval matrices are inherited rather than designed. Somebody set the thresholds years ago, the business changed, and nobody revisited them. Here is how to build one deliberately.
An approval matrix answers one question. Who has to agree before this purchase can go ahead.
Get it right and nobody notices it exists, which is the point. Get it wrong in one direction and small purchases crawl through three levels of sign off. Get it wrong in the other and material commitments are approved by somebody who cannot reasonably assess them.
Most organisations are wrong in the first direction, and the symptom is not complaints. It is people going around the process.
Almost always because they were set once and never revisited. A five thousand threshold that was sensible when the company had forty people is absurd at four hundred, and by then the number has acquired a kind of authority nobody wants to challenge.
The second cause is designing for the exception. Somebody once approved something they should not have, so a control was added for everybody, forever. Do that four or five times and you have a matrix that treats every purchase as a potential incident.
If your approval process assumes bad faith, the people acting in good faith are the ones who suffer for it.
Across the matrices we see, four structures come up repeatedly and they cover most organisations.
Straight through below a floor. Catalog items under a figure you choose need no approval at all. This is the single most effective change most companies can make, because it removes the majority of transactions from the queue and lets attention go to the ones that matter.
Single approver in the middle band. The department manager, and nobody else. Adding finance to this band is where most matrices start to feel heavy without adding much protection.
Two steps above a ceiling. Manager, then finance. Sequential, because finance should only spend time on things the manager has already agreed to.
A separate path for unusual conditions. New supplier, off catalog, or over budget. These are not about value, they are about type, and mixing them into the value bands is what makes matrices complicated.
| Band | Who approves | Why | Common mistake |
|---|---|---|---|
| Below the floor | Nobody, for catalog items | Removes most volume from the queue | Setting the floor too low to matter |
| Middle band | Department manager | One person who knows the context | Adding finance here as well |
| Above the ceiling | Manager, then finance | Two perspectives on material spend | Running them in parallel and wasting finance time |
| Unusual conditions | Depends on the condition | Risk is about type, not value | Folding these into the value bands |
The bands themselves matter less than having a floor at all. Most matrices have no floor, which is why everything queues.
There is no correct number, but there is a useful test. Take last year's purchases and sort them by value. Your floor should sit somewhere that removes most of the transaction count while leaving most of the value under review.
In most mid sized companies that lands surprisingly low, because purchasing follows a long tail. A large share of transactions are small and a small share of transactions carry most of the money.
The ceiling is a different question. It should sit where a purchase genuinely warrants a second perspective, which usually means where it would materially affect a budget rather than at a round number somebody picked.
Sequential means each approver acts in turn. Parallel means everybody is asked at once. Both have a place and choosing the wrong one is a common source of delay.
Use sequential where one approval depends on another. Finance reviewing a purchase the manager has not yet agreed to is wasted effort, so that chain should be in order.
Use parallel where the checks are independent. A new supplier needs procurement to confirm eligibility and finance to confirm budget. Neither depends on the other, so asking both at once saves a day.
No floor. Everything requires approval, so approvers stop reading and start clicking. A matrix that produces rubber stamping is worse than no matrix, because it creates the appearance of control.
Thresholds nobody can explain. If the answer to why is that it has always been that number, it is worth revisiting.
Approvers who cannot assess. Somebody three levels up approving technical purchases they have no basis to judge. That is a signature, not a control.
No delegation. Work stops for a fortnight because one person is on leave. Delegation should be scoped, time boxed and recorded against both people.
Blocking rather than routing. Refusing off catalog purchases outright pushes spend onto personal cards, where you have no visibility at all. Route them for review instead.
Annually is enough for most organisations, plus whenever headcount or structure changes materially. What to look at is not the thresholds themselves but the behaviour around them.
Two signals matter. If a band produces almost no rejections, it is probably not doing anything and could be raised. If you see repeat purchases sitting just below a threshold from the same requester, the threshold is below how people actually buy and it is generating workarounds rather than control.
Adoption is a design problem rather than a training problem.
Read it AnalysisHow to calculate what eight days actually costs you.
Read it GuideWhat it catches, what it misses, and how to set tolerances.
Read itBring your real thresholds and we will configure them live, then put a request through them.
Book a demoBack to all resources