Fixed vs. Variable Costs

Why does cost behavior matter to founders?#

It sets your downside. High committed cost means the spend stays when demand leaves. A company can have an excellent gross margin and a cash break-even it will not reach for two years. The relevant number in a downturn is not margin — it is how much cost you can actually remove, and how fast.

It determines whether scale is worth anything. Fixed cost spread over more successful units lowers average cost, right up to the next capacity step. Variable-cost improvement is a different project entirely: process, supplier negotiation, architecture, or a change in what you charge for.

It prices your discounts correctly. Contribution funds fixed cost, so a discount is repaid out of a smaller unit contribution and pushes break-even further out. A 10% price cut at a 77.5% contribution margin requires a ~15% volume increase just to stand still. See Price Elasticity of Demand.

It disciplines the "we're asset-light" claim. Cloud, contractors, and partners genuinely convert commitments into variable expense — at a higher rate. Flexibility has a price, and the honest comparison is the option value of walking away against the premium you paid for it. See Build vs. Buy vs. Partner.

How do you build the cost model?#

1. Choose the activity driver. Customers, orders, successful tasks, transactions, GB stored, locations, support cases, productive hours. One cost can be variable against one driver and fixed against another — support cost may track tickets, not revenue.
2. Choose the horizon. Next month, next contract term, next planning year. Payroll is fixed next week, partly adjustable next quarter, and a strategy question over three years.
3. Record behavior per cost line. Base or committed amount, variable rate and unit, tier or capacity threshold, notice or cancellation period, minimum commitment, allocation rule, and cash-versus-accrual timing.
4. Use quality-adjusted units. If a failed task consumes the same compute as a successful one, break-even on successful units or you will understate variable cost.
5. Model operating leverage where it is valid. degree of operating leverage = total contribution / operating profit, evaluated at a point comfortably above break-even. It becomes unstable as profit approaches zero (the ratio diverges) and invalid across a capacity step.
6. Run a capacity and runway table. Downside, base, and upside demand; the fixed commitments each triggers; monthly cash burn; time required to adjust; and exit penalties. See Burn Rate and Runway.

Worked example: a step that eats a quarter of your profit#

A software-enabled service charges $20 per successful unit, with $4.50 unit variable cost and $95,000 monthly base fixed cost. Above 10,000 units, a capacity pod costing $25,000 per month is required; each pod supports 5,000 additional units.

unit contribution = $20 − $4.50 = $15.50

At 8,000 units — inside the first capacity range#

break-even = $95,000 / $15.50 = 6,129.03 → 6,130 units

Round up. At 6,129 units the business is 50 cents short; break-even quantities are never fractional in practice.

The DOL of 4.28 says a 1% revenue change moves operating profit about 4.28% — only near this volume, only if price and cost behavior hold, and only within this capacity range. It is a local approximation, not a property of the company.

Crossing the step#

One extra unit costs $24,984 of operating profit. Profit does not recover to its pre-step level until 11,613 units — ($60,000 + $120,000) / $15.50 = 11,612.9, rounded up. Break-even itself moves from 6,130 units to $120,000 / $15.50 = 7,741.9 → 7,742 units.

At 12,000 units, average total cost is ($120,000 + $54,000) / 12,000 = $14.50, better than the $16.38 at 8,000 — and DOL has fallen to $186,000 / $66,000 = 2.82. Average cost improved across the step even though the step itself was brutal. Reporting only the endpoints hides a 1,600-unit dead zone. Plan the threshold; do not extrapolate the smooth 8,000-unit economics through it.

The outsourcing alternative — and a formula that lies#

Now compare an outsourced option that adds $2.75 to unit variable cost on overflow above 10,000 and requires no pod. The tempting crossover calculation is:

$25,000 / $2.75 = 9,090.9 → the pod wins above ~9,091 overflow units

That number is wrong in the way that matters, because it silently assumes one $25,000 pod has unlimited capacity. It does not — it holds 5,000 units.

At these rates the pod never wins, because the crossover of 9,091 overflow units per pod exceeds the 5,000 units a pod can serve. The pod only becomes rational if its capacity rises above ~9,091 units, its cost falls below 5,000 × $2.75 = $13,750, or it buys something the table does not price — latency, data residency, service level, or control. Any break-even formula built from two rates and no capacity constraint will produce a confident number that cannot happen.

What are the common mistakes?#

• Labelling a ledger account permanently fixed. Classification is a function of driver, horizon, and reversibility. The same account gets different labels in a monthly forecast and a three-year plan.
• Calling payroll variable because headcount can change. Notice periods, severance, hiring lead time, and lost institutional knowledge make it committed within any realistic decision horizon.
• Calling cloud fully variable. Minimum billing increments, idle reserved capacity, storage that persists after compute stops, egress, and multi-year commitments are all fixed in practice.
• Ignoring step-fixed costs. Capacity thresholds create the discontinuities that break every linear projection — and they are usually the reason a "profitable at scale" model never arrives.
• Reporting DOL near break-even. As operating profit approaches zero the ratio explodes toward infinity. A DOL of 40 is not a signal of extraordinary leverage; it is a signal that you are barely profitable.

When does the fixed/variable model break?#

Across capacity thresholds. The linear equation holds inside a relevant range and nowhere else. Model the cost structure piecewise, with the thresholds named, rather than fitting one average rate through them.

Under tiered supplier pricing and learning curves. Volume discounts, committed-use tiers, and process learning all make the "variable rate" a function of volume, so the variable cost line is not actually a straight line.

When cutting cost cuts quality. Reducing support headcount, redundancy, or QA lowers spend on the model and raises churn, outages, and refunds off the model. Put the customer consequence in the same table as the saving.

When contracts change the label without changing consumption. A three-year cloud commitment makes cash and obligation fixed while workloads remain variable. Show resource use, invoice structure, and controllability as three separate columns, or you will confuse a billing artifact for an operating fact.

When customers are heterogeneous. A single unit variable cost across enterprise and self-serve customers, or across easy and pathological workloads, describes an average customer nobody serves. Segment before you average.