Of the last 30 models sent to us for review, 21 built revenue top-down from market size. It is the most common single reason a model fails its first serious conversation, and it is also the easiest to fix.
The 2 constructions
Top-down starts from the size of the market and works inward. The market is $4 billion, we capture one percent, revenue is $40 million. Every number after that is a percentage of a number you did not choose.
Bottom-up starts from the smallest unit you control and multiplies outward. 12 sales people, each closing 4 deals a quarter at an average of $18,000, ramping over 5 months. Every number is something you can act on tomorrow.
Both produce a revenue line. Only one produces a plan.
| Top-down | Bottom-up | |
|---|---|---|
| Starts from | Market size | Units you control |
| Answers | What we hope for | What has to happen |
| Testable | No | Yes, driver by driver |
| Links to hiring | Rarely | Necessarily |
| Survives a sensitivity | Moves everything at once | Isolates the failure |
| Time to build | An hour | A day, once |
Why a top-down number ends the conversation
An investor cannot argue with one percent of a market, which sounds like an advantage and is the opposite of one. There is nothing to interrogate, so the interrogation moves to whether you have thought about it at all.
The follow-up questions are always the same four, and a top-down model answers none of them: how many customers, at what price, sold by how many people, closing at what rate. When those go unanswered the meeting stops being about the business and becomes about the file.
A defensible model is not one where the number is right. It is one where every number can be traced to a decision you made on purpose.
Three bottom-up structures, and when each fits
- Volume and price. Units per month at a stated price, with mix visible. Fits products, marketplaces and anything transactional. The driver to argue about is volume growth, and it is a single, testable line.
- Cohorts. Customers acquired per month, retained on a stated curve, spending a stated amount. Fits subscription and usage businesses. The driver to argue about is retention, and it should be visible as a curve rather than a churn percentage buried in a cell.
- Capacity. Sales heads or delivery capacity, with a quota and a ramp per person. Fits services, enterprise sales and anything where a human closes the deal. The driver to argue about is ramp time, which is almost always modelled too short.
Most companies are one of these. A few are two, in which case build them as 2 blocks that sum, not as one blended line.
The cost side has to move with it
The most common failure in an otherwise decent bottom-up model is revenue that grows while headcount does not. If a capacity model says revenue triples, 12 sales people cannot stay twelve. If a cohort model says acquisition doubles, marketing spend or the team acquiring them has to move.
Revenue that grows without cost following is the tell that the 2 halves of the model were built by different logic, and it is visible in about forty seconds.
The runway consequence
When we rebuild a model bottom-up, the runway date usually moves earlier by 2 to 3 months. This is not pessimism. Top-down revenue arrives sooner than sold revenue does, because nothing in the construction accounts for ramp, sales cycle or collection timing.
2 to 3 months is frequently the difference between raising from a position of choice and raising from a position of need. It is better to find it before the timetable is set than during it.
How to rebuild yours this week
- Find your first revenue month and write down, in a sentence, what physically has to happen for that number to occur. If you cannot, that is the whole finding.
- Search the file for constants sitting inside formulas. Each one is a decision nobody recorded.
- Check that hiring is driven by the plan rather than typed in next to it.
- Cut your best driver by 20% and see whether the story survives. If it does not, that is the assumption to defend, and you now know to defend it.
None of that requires a rebuild to test. It requires an hour, and a willingness to find something you would rather not.