Driver-based models

A startup financial model is not a prophecy. It turns explicit assumptions into P&L, cash and runway so you can see what must be true for the plan to work. Good models are driver-based: a small set of operational inputs generate the statements. Bad models are plug-based: someone types the revenue number they wish were true and reverse-engineers a story.

The big idea

Think of a machine, not a forecast. You turn the knobs you actually control — leads, conversion, price, churn, headcount — and the statements fall out. If ending cash looks bad, you can ask which knob did that. A plugged revenue cell has no knob.

The analogy breaks when a driver is itself a fantasy (100% quota attainment forever). Then the machine is honest and the input is not.

What belongs in an early model

• Monthly granularity for 24–36 months (or quarterly for years 3–5).
• One primary revenue motion modeled honestly.
• Headcount plan as the main cost engine.
• Three scenarios sharing the same structure, different drivers.
• Outputs: revenue, gross margin, opex, EBITDA-ish, cash, runway, and key SaaS metrics.

Avoid 40 tabs of unused complexity, precision to the dollar on year-5 enterprise expansion, and macros nobody can audit. A clear Google Sheet beats a fragile cathedral.

The model should expose the same metrics you pitch: implied CAC = S&M / new customers, payback ≈ CAC / monthly gross profit per customer, logo and revenue churn as drivers. If the model needs LTV/CAC > 3× to clear cash in 24 months, but the last two quarters imply LTV/CAC ≈ 1.5×, the plan is fiction. Recalibrate to trailing actuals.

Common mistakes

• One scenario only — then surprise when the world is the downside case.
• Orphaned model — never updated after the fundraise deck ships.
• Ignoring cash timing. Spreadsheet revenue that skips payment terms overstates runway. Always output cash, not just accrual profit.

Check yourself

1. The deck says Q4 revenue is $2M, so that cell is typed in. Driver or plug? Plug. There is no mechanism to debug when Q4 misses.
2. The model needs LTV/CAC > 3× but actuals are 1.5×. What do you do? Recalibrate drivers to trailing actuals, then show what must improve.

Key takeaways

• Prefer driver-based models over plugged outputs you cannot debug.
• Keep the sheet simple enough that you still believe it in six months.
• Reconcile to bank and billing actuals monthly.
• Keep the operating plan and the fundraising upside as separate cases.

Further learning

How those drivers become ARR: revenue build and cohorts.