A Business Case a CFO Can Argue With

One page of business case, taken apart line by line: where each number came from, what halving it does to the decision, and who may challenge it.

Somebody in finance reads your paper for ninety seconds, looking for the line they can push on. When they cannot find it, the paper is not approved but deferred, which is the same outcome with better manners.

A business case is the document that puts a proposed change in front of the person who controls the money, says what it costs, what it is expected to return, and under which conditions the answer should be no. That last part is the one people leave out, which is what makes a case unarguable. A page nobody can push on has not settled the question, it has hidden where the question lives.

I should say what I am not. I have never sat in the finance seat or held a budget of that size. What I do is write the paper finance takes apart, on international projects, and watch which line they go for. It is almost never the total, it is one input three lines above it, and the meeting turns on whether that input has an address.

In short

• The unarguable version is unarguable because no number in it has an owner, a source or a unit, so a reader who disagrees has nowhere to put it.
• Mark every assumption as an assumption, and name the system it would come from in a real project plus the role that would confirm it.
• Print the reversal point, the value at which your recommendation flips, rather than a payback figure that reads like a promise.
• The case below comes only from public sources, so every figure is mine and labelled as mine, which is what honest looks like without access to the books.

What is a business case for, if you already know your answer?

You do know your answer, you would not have written the page otherwise. The case exists so somebody can test it without repeating your work: a reader with a budget is not following your reasoning, they are hunting for the load-bearing number, swapping in their own, and checking whether your recommendation survives.

So the quality question is not whether your numbers are right; written from outside a company, they will not be. It is whether a reader can see which numbers do the work, where they would come from, and how far each has to move before the answer flips.

The case, and what I do not have

The worked example on this site is returns and refunds at a mid-size online retailer, built from a published returns policy, help centre articles, and ten public complaint threads I selected myself. No engagement, no client, no interview, no cost figure or volume report, ever. That is where most people stand writing a first business case: a real process they can watch from outside, no access to the money.

The usual response is to invent a plausible cost per case and hope nobody asks. The better one is to use the number anyway, label it as yours, and name who would replace it. Every figure below is my assumption, not a finding about this retailer, and saying so buys the right to use numbers at all.

The one page

BUSINESS CASE (one page)
Damaged-delivery refunds, web store
Version 1.2. Self-directed case study, built from public sources.
No company data was available to me. Every figure marked (A#) is my
own assumption. Register on page 2.

L1  DECISION REQUESTED
    Approve a named owner and a decision clock for damaged-delivery
    refund cases that stall after first contact.
    Decision owner: [role with authority over the returns process].
    Needed by: [date]. No decision means do nothing.

L2  PROBLEM, COUNTED
    Four of the ten damaged-delivery complaint threads I read end
    with no named person owning the case after first contact.
    Source: ten public threads I selected and read between 12 and
    21 March. A count of the threads I chose, not a rate.

L3  SCALE (A1, A2)
    Assumed 500 cases a month (A1), of which an assumed 20 per cent
    stall with nobody's name on them (A2), so 100 a month. A2 sits
    at half what my ten threads showed, because the threads people
    post are the failures.

L4  COST OF A STALLED CASE (A3)
    Assumed 25 EUR per stalled case (A3), mostly repeat contacts
    and manual chasing. My figure. I have seen no cost data.

L5  EFFECT OF THE CHANGE (A4)
    Assumed the change removes 60 per cent of stalls (A4).
    60 cases x 25 EUR = 1 500 EUR a month, gross.
    A4 is the one number nobody can source before the change ships.

L6  COST OF THE CHANGE (A5, A6, A7)
    Assumed 9 000 EUR one-off build (A5) and 300 EUR a month to run
    (A6). Net 1 200 EUR a month. Over a 24-month horizon (A7, a
    convention I do not own), 28 800 EUR against 9 000 EUR.
    Payback at 7.5 months.

L7  REVERSAL POINT
    The recommendation survives any ONE of A1 to A4 being half of
    what I assumed. It dies when two halve at once: the 24-month
    return then falls below the build cost. Break-even sits at
    45 per cent of assumed volume x stall rate x removal rate
    x unit cost.

L8  WHAT I WOULD MEASURE IN WEEK ONE
    Cases with no assignee after first contact, from the queue.
    One measurement, replacing A1 and A2.

L9  DO NOTHING
    The published policy already sets a deadline for closing a
    refund case, and nothing public says who owns a case that
    misses it. Doing nothing keeps a rule with no keeper.

L10 WHAT THIS PAGE IS NOT
    Not a costed proposal. An argument with its working shown from
    outside the company, every figure meant to be replaced.

The teardown, line by line

Three questions per line, always the same three. Where did this come from, what does the decision do when it is halved, who is entitled to argue with it. L1 takes them quickly: the ask is mine to write, it cannot be halved, and the only person who can argue with it owns the returns process, which is why that slot stays empty rather than holding a title I invented.

L2, the counted problem. Source: ten public threads I picked, on stated dates, with the selection method written next to the number. Halved: two of ten still shows the branch exists, and nothing below rests on it. Who can argue: anyone with twenty minutes and the same kind of public sources, and a reader calling my ten threads unrepresentative is right.

L3, the scale. Source: nowhere, both figures are mine. In a real project A1 is a monthly count by return reason from the order management system, A2 is cases with no assignee after first contact, from the queue. Halved: 750 EUR a month gross, payback at twenty months rather than seven, still positive but no longer obvious. Who can argue: whoever owns that queue, in ten minutes, with a report they already run. This is the pair I most want destroyed, because destroying it is cheap.

L4, the cost of a stalled case. Source: me. In a real project it is a small model: contacts per stalled case from the service desk, times a fully loaded cost per contact from finance. Two owners, one number. Halved: same arithmetic as halving volume, twenty months to payback. Who can argue: a finance business partner, and they will, because a cost per contact from outside their model is the fastest thing here to reject. Hand me theirs and the case gets stronger even if the number worsens.

L5, the effect. Source: judgement, and this is the dangerous one. A4 has no system behind it, because it describes a world that does not exist yet. The only honest sources are a pilot on part of the queue, or a comparable change already made here. Halved: twenty months again, the case still holds. Who can argue: everybody, which is the problem, and only whoever runs that pilot could confirm it. A number nobody can source is one nobody can refute, so it slides through while the room argues about build cost.

L6, the cost of the change. Source: the delivery team, as an estimate with a range. A7, the horizon, is not mine to pick, and choosing it yourself is a quiet way of setting your own exam. Halved: payback under four months, which tells you the build cost is the line everyone attacks and the one with least influence on the answer. Who can argue: engineering on A5 and A6, finance on A7.

L7, the reversal point. Source: arithmetic on the four inputs above, done once and written down. Halved: not applicable, which is why the line exists. Who can argue: anyone who thinks two of my four inputs are optimistic at once, a claim they can make in one sentence.

Notice what the walk exposed. A1 to A4 multiply, which makes them one number wearing four labels, so halving any of them does the same thing to the answer. That never shows up in a table of estimates, only when you write what each line does to the decision.

What happens when every number is half?

Then the case is dead, and the page should say so before the reader gets there.

Halve all four inputs and the monthly return falls to about a tenth of what I claimed, below the assumed running cost, so the change loses money before anybody pays for the build. That is the most useful sentence here: it says how wrong I have to be for the answer to change, and lets a reader judge how likely that is.

A payback figure cannot do this: a single number reads as a promise, and a promise can only be believed or disbelieved. A reversal point reads as a test, and a test can be run by somebody who does not trust you. There is a cost to this, and I would rather name it: a page this full of labels can read as a list of things you failed to find out, which is what L8 answers.

The assumption register you can copy

Page two, and the artifact worth more than the page in front of it. Copy it, keep the columns, fill the last one first.

Three rules for filling it. The last column names something somebody could observe, never "if the assumption is wrong". A row with a blank confirmer gets flagged on the case rather than quietly averaged in. And every row of type "effect of the change" says so, because no system can confirm that number before the change ships, and it is the row that gets waved through.

For the document this page sits inside, there is the BRD piece. The scorecard holds the checks I run against my own work first. The full returns case, register and review notes included, is in the Proof Pack. The business case above is not in it, on purpose: the one-pager in that pack carries no cost figures at all and says so on its face, because the assumptions I am willing to publish inside an article are not the same as the ones I would put in a decision document with nothing to source them from.

What can you do with one hour?

Take a business case you have written, or any recommendation you made in writing, and find the numbers that multiply. Give each an owner, a source it would come from in the real world, and a value. Then halve them one at a time and write one clause about what your recommendation does.

Somewhere in that hour you will hit a number you cannot attribute to anybody. That number is the whole meeting, and until now it was the one line nobody could argue with.

All field notes · The portfolio guide · More in The Document That Gets Read