Marginal Cost and Marginal Revenue

Why does marginal analysis matter to founders?#

Average margin can reject a good deal. A company carrying heavy historical fixed cost shows a poor average margin. If that average is loaded onto a new opportunity, the team declines business that would have contributed real cash. The sunk portion should never enter the comparison.

Average variable cost can accept a bad deal. The next customer may trigger a capacity reservation, a new support shift, a compliance audit, or a custom integration. Averages smooth those thresholds away and make the deal look cheaper than it is.

Pricing decisions reach backwards into the installed base. A discount designed to win incremental units usually applies to renewals and existing accounts too. Marginal revenue has to net out that cannibalization, or the model will recommend a price cut that shrinks the business.

Runway converts timing into economics. A capacity investment that is profitable across three years can still be unfinanceable this quarter. Cash timing, downside demand, and exit penalties belong inside the decision, not in a footnote. See Burn Rate and Runway.

How do you run the analysis?#

1. State the counterfactual. Compare "accept this block" against "do not accept it," holding unrelated plans constant. If a cost appears in both branches, it is not incremental.
2. Choose a meaningful output unit. Orders, successful tasks, customer-months, workloads, locations, or a capacity tranche. Separate segments with materially different price or cost.
3. Build incremental revenue. New recurring or usage revenue, minus discounts extended to existing volume, plus or minus expected expansion and churn effects, net of refunds, credits, and mix — then adjust for probability and timing.
4. Build incremental cost. Direct variable cost, incremental labor, commissions, support, losses, and any step-fixed infrastructure. Add opportunity cost when the capacity could serve a better use.
5. Compare across the right horizon. For a reversible short-run decision, contribution is enough. For hiring, market entry, or capacity, discount the cash flows: incremental NPV = Σ [(inflow_t − outflow_t) / (1 + r)^t].
6. Apply constraints last. Positive expected incremental value is necessary, not sufficient. Cash, quality, capacity, legal, safety, and strategic limits all sit above the arithmetic.

Worked example: the next block of workloads#

A SaaS infrastructure company serves 10,000 monthly workloads. It can sell the next 1,000 at $12 each. Direct compute and support cost $4 per workload, but capacity above 10,000 requires a new pod costing $9,000 per month that supports up to 5,000 additional workloads.

Case A — sell only the 1,000-workload block#

MR per workload = $12 · average incremental cost per workload = $13

Reject. Note the phrasing: $13 is the average incremental cost over the block, not the marginal cost of any single unit. Workload 10,001 has a true marginal cost of $4 + $9,000 = $9,004; workloads 10,002 to 15,000 cost $4 each. Reporting "$13 marginal cost" is a convenient average that hides a discontinuity — which is precisely the thing the decision turns on.

Case B — same pod, more utilization#

If credible demand is 3,000 workloads rather than 1,000:

The identical capacity decision reverses on utilization alone. Break-even volume for the pod is:

$9,000 / ($12 − $4) = 1,125 workloads

So the pod pays for itself somewhere between the 1,000 and 3,000 cases — a threshold worth knowing before the demand conversation, not after.

Case C — the price cut that looks like growth#

Now suppose winning those 3,000 extra workloads requires cutting price from $12 to $10 for all 13,000 workloads, including the 10,000 already served.

MR per added workload = ($130,000 − $120,000) / 3,000 = $3.33

Marginal revenue is $3.33, against an average incremental cost of $7 for that block (or $4 if the pod is already paid for). Revenue grew $10,000 and the decision still destroys value, because the $2 discount on 10,000 existing workloads costs $20,000 — twice the revenue the new volume brought in. The new price of $10 comfortably exceeds the $4 direct variable cost, which is exactly why this mistake is so easy to make.

Caveats on the precision. All three cases assume the demand is real and immediate, the pod is the only step triggered, no existing customer renegotiates beyond the modeled discount, and the monthly frame is the right one. Change the horizon to twelve months with churn and expansion, and Case A can turn positive while Case C stays negative. Run the block over the horizon the commitment actually spans.

What are the common mistakes?#

• Using average cost for a forward-looking decision. Ask what actually changes if you say yes. Nothing else belongs in the comparison.
• Treating unit variable cost as marginal cost. Step-fixed capacity, incremental headcount, onboarding, and support tiers are all caused by the decision even though they are not per-unit.
• Allocating sunk R&D to reject positive contribution. Money already spent cannot be recovered by declining revenue.
• Ignoring discounts on the installed base. Marginal revenue can be a fraction of the new price — $3.33 against a $10 list price in the example above.
• Reporting block averages as marginal cost. With step costs, the average across a block and the cost of the marginal unit can differ by three orders of magnitude.

When does marginal analysis break?#

When output changes the product. Adding a marketplace participant changes value for other participants (network effects); adding an ad slot can reduce retention; adding a low-fit customer can raise support load and drag the roadmap. These are real costs and benefits that a per-unit table does not capture.

When the commitment is irreversible and demand is uncertain. A data center, a regulated-market entry, or a specialized hire creates option value and abandonment value that a one-period comparison misses. Use scenarios and, where the stakes justify it, an explicit option framing.

When capacity is shared or contended. If the pod also serves another product line, its cost is not fully caused by this decision, and the opportunity cost of the capacity may exceed its cash cost.

When the price is not yours to set. In a competitive market with close substitutes, the demand curve you face depends on rivals' responses. Marginal revenue computed on a static demand curve is then a fiction. See Game Theory and Price Wars.

When constraints are not economic. Positive expected contribution does not justify serving a prohibited use, breaching a safety limit, or accepting unmanaged legal risk.