Unit EconomicsIntermediate6 min read

Marginal Cost Analysis

Marginal cost is what it costs to produce ONE MORE unit — not the average cost across all units. Formula: Marginal Cost = ΔTotal Cost ÷ ΔUnits. The KnowMBA POV: average cost is what you report; marginal cost is what determines whether your next decision creates or destroys value. The marginal cost of a 1,001st AWS API request is fractions of a cent. The marginal cost of a 1,001st enterprise customer might require a new account executive ($200K/year). Average cost is a backward-looking accounting metric; marginal cost is the forward-looking decision metric.

Also known asMarginal CostIncremental CostNext-Unit EconomicsMarginal Analysis

Challenge a friend Browse library

The Trap

The trap is using average cost to make pricing or scaling decisions. Your average COGS is $20 per unit, you're considering a bulk order at $25 — average analysis says yes, $5 profit per unit. But marginal cost analysis reveals the bulk order needs a second shift ($50K fixed) plus overtime — marginal cost is $32. You'd lose $7 per unit. Conversely, founders REJECT good deals because average cost looks unprofitable when marginal economics would be highly accretive (e.g., filling unused capacity at any price above marginal cost is profitable).

What to Do

For every major scaling decision, calculate marginal cost separately from average cost. Three questions: (1) What additional fixed costs trigger at this volume? (2) What variable costs scale linearly? (3) Are there step-cost cliffs in headcount, infrastructure, or licensing? Accept any opportunity where marginal revenue > marginal cost (in the short term, to fill capacity). Reject opportunities that look profitable on average but lose on the margin.

Formula

Marginal Cost = Change in Total Cost ÷ Change in Units (typically Δ over the next 1,000 units)

In Practice

Airlines famously price empty seats below average cost because marginal cost is near-zero. The flight is going regardless; an empty seat earns $0, while a seat sold at $80 (even if average cost is $200) generates $80 of contribution to the fixed cost of the flight. This is why Hopper, Hotwire, and other 'opaque' bookings exist — airlines can monetize marginal capacity below average cost without breaking their public price points. The same logic applies to SaaS at scale: filling spare server capacity with low-priced overflow customers can be highly accretive even when average analysis says no.

Pro Tips

01
Pricing rule: short-term, accept anything above marginal cost (it contributes to fixed). Long-term, prices must cover average cost (or you go bankrupt). The two-tier framing prevents both leaving money on the table AND racing to the bottom.
02
Marginal cost analysis is brutal on AI features. The 1,000,000th LLM API call costs essentially the same as the 1,000th — there's no economy of scale. Pricing must reflect this. Many AI startups under-price because they expect software-like marginal cost decay that never comes.
03
Plot marginal cost as a curve, not a number. Most businesses have a U-shape: high marginal cost early (under-utilized fixed infrastructure), low in the middle (optimal scale), then rising again (capacity constraints kick in). Operating in the middle of the curve is the goal.

Myth vs Reality

Myth

“Marginal cost is just variable cost”

Reality

Marginal cost includes ANY incremental cost triggered by the next unit — including step-fixed costs that activate at thresholds. Hiring an AE because of one big customer is a marginal cost, even though salary is technically 'fixed.'

Myth

“If marginal revenue > marginal cost, take the deal”

Reality

True short-term, false strategically. Discount-driven deals retrain your customers (and competitors) on lower price points. Some deals are right to reject even though marginally profitable, because they reset market expectations.

Try it

Run the numbers.

Pressure-test the concept against your own knowledge — answer the challenge or try the live scenario.

🧪

Knowledge Check

You produce 1,000 widgets at $30 average cost each. A customer offers $25 each for 200 more units. Variable cost per widget is $18, but the order requires $1,400 of additional setup. Should you take the deal?

Real-world cases

Companies that lived this.

Verified narratives with the numbers that prove (or break) the concept.

🌐

Hypothetical: Cloud Infrastructure

2024

failure

Hypothetical: A cloud-services startup priced storage at $0.05/GB based on average cost analysis ($0.04 fully-loaded). When a Fortune 500 customer asked for $0.025/GB at 50PB scale, sales reflexively declined. The CFO ran marginal cost analysis: at that scale, marginal storage cost was $0.012/GB (existing infra had headroom). The deal would have generated $1.25M/year of marginal contribution. They lost it to a competitor. They later instituted a 'marginal pricing' framework for enterprise deals above 10PB.

List Price

$0.05/GB

Average Cost

$0.04/GB

Marginal Cost (50PB)

$0.012/GB

Lost Annual Contribution

$1.25M

Average cost analysis can cost you deals that would have been hugely accretive on the margin. Always run both — and price strategically when capacity is unused.

📦

Amazon

1997-2010

success

Bezos famously operated retail near zero margin for 13 years, accepting deals where marginal revenue barely exceeded marginal cost. The strategy: build infrastructure (fulfillment centers, delivery network) whose marginal cost would drop dramatically with scale. By 2010, Amazon's per-unit fulfillment cost was 1/10th of competitors who hadn't invested in fixed infrastructure. The marginal cost advantage became a competitive moat that justified the early-years 'losses.'

Years Near-Zero Margin

1997-2010

Fulfillment Cost Advantage

~10x lower vs competitors

Result

Permanent cost moat

Marginal cost can be DESIGNED, not just measured. Bezos built infrastructure whose marginal cost would fall faster than competitors could match.

Source ↗

Related concepts