Capital Spending: The Approval Gates Before Anything Is Ordered

By the time a capital invoice reaches accounts payable, every decision worth controlling has already been made. The specification was set by whoever wanted the thing. The supplier was chosen by the same person. The price was agreed. The commitment exists. Accounts payable can confirm that the goods arrived and that the invoice matches, and that is genuinely worth doing, but it is a check on a decision rather than a decision.

So capital discipline is not a budget line and it is not an approval limit. It is a sequence of gates, and the useful ones all sit before a purchase order exists.

Here is the claim that companies at this size argue with most: being in the approved capital budget is not authority to spend. The budget said the money was set aside eleven months ago on the basis of a case somebody made at the time. The release is a separate act, taken when the money is about to leave, by someone who has confirmed the case still holds. Collapse those two into one and you have a company that spends its capital budget in the fourth quarter on the things it decided it wanted in the previous fourth quarter, whether or not any of it still makes sense.

The five gates

Gate	What happens	Who moves it	What exists at the end
0. Intake	The request is written down on a fixed form. Anything not on the form does not exist and is not in the plan.	Requester	A numbered request, in one register
1. Case	The business case is built to the depth the band requires. Depth is set by size, not by enthusiasm.	Requester, with finance on the numbers	A case document, proportionate
2. Decision	Approve, decline, or defer to a named point. A deferral without a date is a decline that nobody wanted to say.	Approver for the band	A dated decision, recorded against the request number
3. Release	Immediately before commitment. Confirm the case still holds, the cash is there, and the conditions attached at gate 2 have been met.	Finance lead, plus the gate 2 approver above the middle band	A release note, and only now may a purchase order be raised
4. Post-completion review	After the asset is in use. Did it cost what we said and did it do what we said.	Finance lead, presented to the approver	A short written answer, filed against the same request number

The request number carries through all five. That single detail does more than it looks like it should: it means the case, the decision, the release, the purchase order, the invoice and the review all carry the same reference, and a year later somebody can reconstruct a decision without interviewing anyone.

The size bands, which you set

I am not going to publish thresholds. The right band boundaries depend on the size of your balance sheet, the shape of your cash cycle and how much a mistake would hurt, and a figure printed here would be invented and then copied into companies it does not fit. What follows is the rule for setting them, which is the transferable part.

Call them Band A through Band D, A being smallest.

Set Band D first, from the top. Band D is the amount at which a single commitment would change the cash plan for the quarter or would need to be mentioned to your lender. That is not a preference, it is a fact about your business, and you can read it off your own cash forecast in about ten minutes.

Set Band A second, from the bottom. Band A is the amount below which the cost of the paperwork exceeds the risk of the purchase. If a two-page business case for a purchase costs more management time than the purchase itself, the band is set too low and people will route around it, which is worse than not having it.

Then split the middle into two. The boundary between B and C is where you want a second person involved. There is no science to it. Put it somewhere and move it after a year if it turns out wrong.

Three rules about how the bands are measured, and all three exist because each has been used to get around a gate.

Measure the total committed cash over the life of the commitment, not the first payment. A three-year arrangement at a modest monthly figure is a Band C or D commitment wearing a Band A costume. This is the most common bypass and it is almost never deliberate.

Aggregate related items into one request. Six workstations bought in the same week for the same team are one request, not six. If you have to ask whether two items are related, they are.

Do not band by accounting treatment. Whether something will be capitalised or expensed is a reporting question, answered later, under whatever framework you report on. It has nothing to do with whether the decision needs a case. A large multi-year software subscription may never appear in the fixed asset register and is still one of the largest commitments the company will make that year. Band on cash committed, not on where the entry lands.

What each band has to produce

Band	Business case depth	Approver at gate 2	Release at gate 3	Post-completion review
A	The intake form and nothing more. What it is, why now, what it costs, what happens if we wait.	Department head	Not required. The gate 2 decision is the release.	Not required
B	Intake form, plus a written alternative considered and rejected, plus the operating cost the asset brings with it	Department head plus finance lead	Finance lead confirms cash and that the price has not moved materially	Only if actual cost exceeded the approved amount by more than the tolerance you set
C	All of Band B, plus a cash profile by month, plus the effect on the covenant calculations if you have any, plus who owns the asset once it is in service	Owner or CEO, on the finance lead’s recommendation	Finance lead plus the gate 2 approver, in writing	Required, at a fixed interval after the asset is in use
D	All of Band C, plus the financing decision made explicitly rather than by default, plus what happens to the plan if the benefit does not arrive	Owner or CEO, with a written note to whoever else is entitled to one under your shareholder or lending arrangements	Owner or CEO, in writing, immediately before commitment	Required, and presented rather than filed

The Band A row is the one to read twice. Requiring nothing beyond an intake form for the smallest band is deliberate. A gate structure earns its authority by being cheap where the risk is small, because that is what stops people treating the whole thing as an obstacle. Band A stays in the register for completeness rather than for control. The register is how you know what the year’s capital spending actually was.

The three fields on the intake form that do the real work

The form is short. Five or six fields, one page, and it should take fifteen minutes.

What happens if we do nothing. This is the field that separates a want from a need, and it is the one requesters most often leave blank or fill with a sentence that just restates the request. A genuine answer usually names a date, a failure, a capacity limit or a contractual obligation.

What it costs to run, per year, once we own it. Maintenance, licences, insurance, power, consumables, training, and the person whose time it takes. Capital requests are argued on the purchase price and paid for over years on the running cost. A company that buys three things a year without this field ends up with an operating cost base that grew for reasons nobody can trace.

What it replaces, and what happens to that. If nothing is being replaced, say so. If something is, then either it is disposed of, in which case there is a proceeds and a book value question, or it is kept, in which case the running cost of the old thing is still there and the new one did not save what the case claimed.

The other fields are administrative: requester, department, the amount, the band it falls in, and the month it is wanted.

Where the gate leaks

Every company with a capital gate also has capital spend that went around it. These are the four routes, and the fix for each.

Leak	How it happens	Control
The emergency	Something breaks and it has to be replaced today	A written emergency route with a single approver and a hard rule that the intake form is completed within a defined number of days afterwards. The form still gets filled in. What changes is the order.
The subscription	A multi-year commitment signed by someone who does not think of software as capital	Band on total committed cash, and put contract term on the vendor setup form so a long commitment is visible at the point the supplier is created
The accumulation	Twelve small purchases that together are a Band C decision	A same-category aggregation report at close, run against the capital register and the relevant expense accounts
The order placed anyway	Somebody orders it and finance meets the invoice	The invoice goes back to the requester’s own department with the request number required before it is paid, and the exception is logged. The payment run controls are where that stop actually happens, which is the one place accounts payable genuinely is the control.

The last row is worth being firm about. The natural instinct when an unapproved capital invoice appears is to process it, because the goods are already here and the supplier wants paying and refusing feels petty. Process it, and the gate is now advisory. Send it back once, with the request number required, and it stops happening. The log exists to be counted rather than to embarrass anyone. A count of exceptions is the only honest measure of whether the gate is working, and a gate nobody ever bypasses is either working or set too high to be relevant.

The post-completion review

Two questions, asked at a fixed interval after the asset is in service, and both of them answered in writing on one page.

Did it cost what we said? Total, including the running cost in the first period, the installation, the training and everything else the case forgot. Compare to the approved amount, not to the revised amount somebody agreed halfway through.

Did it do what we said? Against the specific claim in the case. If the case said it would remove a bottleneck, ask whether the bottleneck moved. If the case said it would let the company take on a kind of work it could not take on before, ask what work has been taken on.

Two rules about how the review is conducted, and they matter more than the questions.

It reviews the assumption, not the person. A capital review that becomes an examination of the requester’s judgement is a review that will produce optimistic cases forever afterwards, because nobody will write a case they might be asked about. The output is a note about which kind of assumption the company gets wrong, and companies are usually consistently wrong in one direction, which is genuinely useful to know before the next case is written.

Something has to happen with the answer. Feed it into the next capital cycle explicitly. If the last three reviews found that installation and training were consistently understated, then the intake form gets a mandatory field for them and the next case is better. A review that is filed and never referenced is a form of work that looks like governance.

Not everything gets reviewed. Band A and most of Band B do not, because reviewing a small purchase costs more than the information is worth. That asymmetry is the design, not a gap in it.

Fitting the gates to the budget year

The gates run continuously. The budget runs annually. They meet in three places and it is worth being explicit about which is which.

At budget time, the capital plan is a list of gate 0 and gate 1 items, banded, with the cash profiled by month. It is a plan, not a set of approvals. Say that on the page so nobody reads inclusion as consent.

During the year, a request that was in the plan still goes through gates 2 and 3. Being in the plan means the case does not have to argue that the company intends to spend money on this category. It does not mean the case is made.

A request that was not in the plan goes through the same gates and one extra question: what in the plan is being displaced, or has the plan changed. If the answer is neither, the capital budget was not a constraint and next year’s version of it will not be either.

What we do not do

I do not set your band boundaries. They are a function of your balance sheet and your cash cycle, you can derive them in an afternoon from your own forecast, and a number I supplied would be a guess with an authority it did not earn.

I do not decide whether a purchase is a good idea. The gate structure makes the case visible and puts it in front of the person entitled to decide. That person is the owner, and the controls work exists to make sure they are asked at the point where the answer can still change something.

And I do not build a capital approval process for a company that buys two things a year. At that volume the register is a spreadsheet, the gates collapse to intake and decision, and installing anything heavier produces a process that is abandoned by the second quarter and then cited as evidence that controls do not work here. The reporting still has to show what was bought and what it cost to run. The apparatus around it should be the smallest thing that achieves that.