Building a growth model your CFO will believe

There is a particular meeting that most marketing leaders have had. The plan is presented, the numbers are questioned, and within ten minutes the conversation has moved from strategy to a debate about whether the pipeline forecast is credible. The plan may have been excellent. It did not survive the model.

A growth model is the artefact that prevents that meeting. It is not a forecast and it is not a budget — it is a written statement of how you believe the business grows, expressed in numbers that can be checked. Built properly, it changes the conversation from “do we trust marketing” to “which of these assumptions should we test first”, which is a far more productive place to be.

What a growth model is not

It is not a spreadsheet that takes last year’s revenue and multiplies it by a growth rate. It is not a channel budget with a target return applied. And it is not a bottom-up pipeline forecast built from individual campaign plans, which is the version most marketing teams produce and the version finance trusts least, because every input is an estimate made by the person whose budget depends on it.

A model is a chain of assumptions where each link is either measured from your own history or explicitly flagged as a guess. The value is not the output number. The value is that everyone can see which links are load-bearing.

The structure that works

We build every client model in four layers.

Volume. How many people enter the funnel, by source, per month. This is the layer with the most historical data and the least controversy.

Conversion. The rate at which each source moves to the next stage. Crucially, these rates must be measured per source, not blended. A blended conversion rate hides the entire point of the model — that paid search and content produce different-quality demand.

Economics. Average contract value, gross margin, and time to close, again by source. Most models stop at revenue. Finance thinks in margin and cash timing, and a model that ignores both will be treated as marketing’s number rather than the company’s.

Constraints. The layer everyone forgets. Sales capacity, delivery capacity, support headcount. A model that projects three times the leads with the same number of account executives is projecting a queue, not revenue.

Separate the measured from the assumed

This is the single practice that earns credibility. Every cell in the model gets a colour: measured from our own data, benchmarked from an external source, or assumed. Then you count them. A model where eighty percent of the load-bearing cells are assumptions is not wrong, but it needs to be presented as a hypothesis rather than a plan.

The first time a team does this, the result is usually uncomfortable. Conversion rates that everybody quotes turn out to have come from a single quarter two years ago. Average contract value turns out to be a mean dragged upward by one enterprise deal. Surfacing this is the work; the model is just the container.

Sensitivity beats precision

Finance does not need your central estimate to be right. They need to know how wrong it can be and which input is responsible. So every model we deliver includes a sensitivity table: change each major assumption by a fifth in each direction, and show the effect on the annual number.

The table almost always shows that two or three assumptions dominate everything else. That is enormously useful, because it converts an abstract question — how do we de-risk the plan — into a specific one: these two numbers matter, so this quarter we are going to measure them properly.

Tie it to a review cadence

A model that is built once and referenced at budget season is a document. A model that is reconciled monthly against actuals is an instrument. The monthly reconciliation is short: for each assumption, what did we predict, what happened, and is the difference noise or a signal.

Within two quarters this changes the internal politics of marketing measurement more than any dashboard. The team develops a track record. When the model has been within ten percent for six months, the next plan is discussed rather than defended.

Common ways it goes wrong

• Modelling leads instead of revenue. Leads are marketing’s currency and nobody else’s. If the model stops before closed-won, it will be read as a departmental metric.
• Assuming linear channel scaling. Doubling paid spend does not double paid conversions. Build in a saturation curve, even a crude one, or the model will confidently recommend an impossible plan.
• Ignoring time. A ninety-day sales cycle means spend in Q4 shows up as revenue in Q1. Models that recognise revenue in the month of spend look excellent and cause cash problems.
• Hiding the assumptions. A model nobody outside marketing can open and interrogate will not be believed, no matter how sophisticated it is.

How much detail is enough

Less than most teams build. A model with four sources, three funnel stages and honest economics will out-predict one with twenty-two channels and a tab for every campaign, because every additional input is another estimate compounding the error.

Our rule of thumb: if you cannot walk someone through the whole model in fifteen minutes, it is too complicated to be trusted and too fragile to be maintained.

The version to build first

Start with one page. Four sources, monthly volume, conversion to closed-won, average contract value, and a sensitivity table on the three assumptions you are least sure about. Reconcile it against the last twelve months of actuals before you show it to anyone — if it cannot retrodict the year that already happened, it will not predict the next one.

That single page, reconciled monthly and honest about what it does not know, is worth more than any planning tool you could buy. It gives finance something to argue with, which is precisely what a good working relationship between the two functions is made of.

Getting finance into the room early

The fastest way to build a model finance believes is to build it with finance. This sounds like a platitude and it is a scheduling decision: one hour, before any numbers exist, in which you ask what they need the model to answer and what has made them distrust marketing forecasts in the past.

The answers are usually specific and easy to accommodate. They want cash timing, not just annual totals. They want the plan expressed in a way that maps to the categories in their own model. They want to know which line item to cut first if the quarter is short, and they would rather have that conversation in advance than in week ten.

None of that is difficult to provide, and providing it converts finance from an auditor into a co-author. A model with two authors gets defended in the meetings marketing is not in, which is where most budget decisions are actually made.

What to do when the model is wrong

It will be. The question is what happens next, and the answer determines whether the model survives its first bad quarter.

The failure mode is silent revision: the assumptions are quietly updated to match what happened, the model once again fits the data perfectly, and it has learned nothing. The alternative is to write a short note against the assumption explaining what moved and why, and to leave the original prediction visible. Over a year this produces a document that shows how the team’s understanding of the business has changed — which is far more persuasive to a sceptical executive than a model that has always been right.

We keep a change log at the bottom of every client model for exactly this reason. It is usually the first thing a new CFO reads.

A worked sense of scale

For a business doing a few million in annual revenue, the whole exercise is roughly two weeks of work: three days pulling and cleaning historical data, two days building, a day with finance, and the rest spent discovering that some of the data you assumed existed does not. The monthly reconciliation afterwards takes about two hours.

That is a small enough commitment that the honest question is not whether it is worth building, but why the plan was ever presented without one.

Two questions that expose a weak model in thirty seconds

If you inherit a growth model and want to know quickly whether it can be trusted, ask two things.

What would have to be true for this to be wrong by half? A model built by someone who has thought carefully will have a ready answer, usually naming a single assumption. A model built backwards from a target will produce a defensive answer about market conditions.

Show me the month this model was most wrong last year. If nobody has checked, the model has never been tested against reality and its accuracy is unknown regardless of how detailed it looks. If somebody can point to the month and explain it, you are dealing with a working instrument.

Both questions take under a minute and neither requires you to understand the underlying business. They are, in our experience, better predictors of a model’s usefulness than any inspection of the formulas.

The compounding benefit

The reason to persist through the awkward first quarter is that the value compounds in a way most marketing work does not. Every month of reconciliation narrows the range on an assumption. Every narrowed assumption makes the next plan more specific. Within a year, a team that started by guessing at its conversion rates is negotiating budget with measured ones, and the arguments move from whether marketing works to which part of it should grow.

That shift — from justifying existence to allocating capital — is the actual outcome. The spreadsheet is incidental.