Marginal Analysis
Date: 2026-08-16
Decide on the next unit, not the average one. Averages describe what’s already happened; every actual decision is about whether to do one more thing, and the answer is usually different.
What it is
Marginal analysis compares the additional cost and additional benefit of one more unit — one more order, one more pound of spend, one more customer.
The rule: do it while marginal benefit exceeds marginal cost. Stop when they meet.
Why averages mislead
The clearest case is paid acquisition, where returns diminish:
spend customers avg CAC marginal CAC
£100k 3,000 £33 £33
£140k 4,000 £35 £40 ← the extra £40k bought 1,000
£180k 4,500 £40 £80 ← the extra £40k bought only 500
At £180k, average CAC is £40 and looks affordable against a £24 contribution LTV… except it doesn’t, and either way the decision “should we spend the next £40,000” is answered by £80, not £40.
In plain terms: the average includes all the cheap customers you already bought. The question is always about the next one, and the next one costs more than the last.
This is why a channel can look efficient on average and be unprofitable at the margin — and why “our CAC is fine” is not an answer to “should we increase the budget”.
Where it applies
- Paid spend. The example above. Every channel has a point where marginal CAC exceeds contribution LTV — Customer Acquisition Cost, Bid Strategies and Budget Allocation
- Stock holding. Hold more while the marginal contribution of avoided stockouts exceeds the marginal cost of capital and markdown risk — Stockouts and Availability
- Discount depth. Each additional percentage point costs more contribution and buys less incremental volume — Discount Impact on Margin
- Free shipping thresholds. The marginal order unlocked by lowering it versus the margin given up on orders that would have qualified anyway — Shipping Thresholds
- Testing effort. The marginal value of the next test versus the traffic and engineering it consumes — Test Prioritisation
Sunk costs don’t count
The corollary that’s hard to apply in practice: money already spent is irrelevant to the next decision. Why that’s hard to act on is Sunk Cost Fallacy.
£40,000 spent building a feature that isn't working
wrong question "we've spent £40,000, we should
finish it"
right question "what does the NEXT £10,000 buy,
and is that the best use of it?"
The £40,000 is gone either way. It should have no weight — and it reliably does, which is why projects continue past the point of sense. Recognising it is easy; acting on it is the hard part, because someone owns the original decision.
Marginal contribution, not marginal revenue
The same distinction as everywhere else in this domain. The marginal order brings revenue and costs variable cost:
one more order at £50
revenue +£50.00
variable cost −£35.00
fixed cost £0
────────
marginal contribution +£15.00
Fixed costs don’t enter the marginal decision at all — they’re already committed. Which means a marginal sale is worth taking at any price above variable cost, even below full cost, provided it isn’t cannibalising a full-price sale — Fixed and Variable Costs.
That’s the legitimate case for clearance pricing, and the reason it’s legitimate is precisely that the alternative is zero.
The practical habit
When someone asks whether to do more of something, ask what the last unit cost and what the next one will. Not the average. Almost every over-investment in growth comes from applying an average to a marginal decision.