Field Guide · Measurement

Forecast a range, not a number

A point forecast is a wish with a decimal. The operator's version triangulates three models that should disagree — then tracks its own error so it sharpens every cycle.

By Khalid HamadehUpdated June 20268 min read
The short version

A single number hides its assumptions and its uncertainty. Build a range, from three lenses:

The premise

Why single-number forecasts lie

"We'll do 9,000 subs next quarter" tells a CFO nothing about how or how sure. A point estimate buries its assumptions and erases its uncertainty — so when it misses, it reads as a failure instead of a known risk that came true. A range fixes all three problems at once: it surfaces the assumptions, makes the uncertainty explicit, and gives the business something honest to plan against.

KeyA forecast you can't decompose into assumptions isn't a forecast — it's a wish. The range is where the assumptions become visible and arguable.
The framework

The three-lens forecast

Build the same forecast three different ways. They have different blind spots on purpose — so their agreement is signal and their disagreement is a map of your risk.

LensHow it's builtBest atBlind spot
Bottoms-up funnelSpend × conversion rates down the funnel to subsTying the forecast to controllable inputsAssumes CVRs hold; misses saturation & market shifts
Top-downMarket size × share × seasonality & trendCatching demand ceilings and seasonalityToo coarse to action; slow to see channel-level change
Cohort / LTVRetention & expansion curves projected forwardRevenue durability and the installed baseBackward-looking; new-cohort behavior can shift

Three forecasts, one range

Where the lenses agree, the band is tight. Where they diverge, it widens — and that width is the honest uncertainty.
subs months ahead → now Bottoms-up (base) Top-down (high) Cohort (low) ↕ the forecast range widens with time

Early on, the three lenses sit close — near-term momentum is hard to argue with. The further out you go, the more they fan apart, and the band between the lowest credible and highest credible lens is your forecast. Don't collapse it to a point; report the low / base / high and the assumption that separates them.

Build it

The bottoms-up funnel, step by step

This is the lens you can actually steer, because every stage is a number you influence. Each conversion rate is an assumption you can defend — or that a skeptic can attack.

Spend → subscriptions, one rate at a time

Illustrative figures — the point is the structure, not the numbers.
Ad spend $1.2M Installs / clicks 240K Signups 96K Paid subs 9,600 Retained @ M3 7,300 ↓ $5 / install ↓ 40% signup ↓ 10% to paid ↓ 76% retain

Read it top to bottom: spend buys installs at a cost per install, a share of installs become signups, a share of signups convert to paid, and a share of those retain. Change any one rate and the forecast moves — which is exactly why this lens is the one you manage to. Your low/base/high cases come from flexing the two or three rates you're least sure about.

TipDon't forecast a flat conversion rate at scale. As spend grows, the cost per install rises and CVRs soften — borrow the saturation curve from your mix model so the funnel bends like reality does.
Make the call

Turn three forecasts into a decision range

You don't average the lenses — you assemble a range from them, and you treat disagreement as information rather than noise to be smoothed.

L

Low = the most conservative credible lens

Usually the cohort/LTV view or a pessimistic funnel. This is your "we can commit to this" number — the floor you'd stake a plan on.

B

Base = where the lenses converge

The consensus of the three, with your best-guess assumptions. This is what you publish as the plan.

H

High = the most optimistic credible lens

Often the top-down view in a strong market. This is the "if things break our way" upside — useful for capacity and stretch planning, dangerous as a commitment.

Disagreement is the most valuable output

When the funnel says 9,000 and top-down says 13,000, don't split the difference — find out why. Maybe the funnel hasn't priced in a seasonal tailwind; maybe top-down is ignoring a saturating channel. The reconciliation is where you actually learn something about the business. A tight range means you understand the quarter; a wide one tells you exactly what to go investigate.

Compound it

Track your error so the model improves

The difference between a forecaster who gets better and one who repeats the same misses is a single discipline: logging forecast versus actual, every cycle.

1

Record the prediction and the assumptions

Before the period, write down the low/base/high and the key rates behind them. A forecast you didn't write down can't be graded.

2

Separate bias from variance

Bias = consistently high or low (a broken assumption). Variance = noisy both ways (genuine uncertainty). They demand different fixes — recalibrate the assumption vs. widen the band.

3

Trace each miss to one assumption

"We missed because paid CVR dropped from 10% to 8%." Now that rate gets more scrutiny next cycle. Misses with a named cause make the next forecast sharper; misses you shrug off repeat.

TL;DRForecasting accuracy isn't a talent — it's a feedback loop. Grade yourself, attribute the error, fix the assumption. The forecast that learns from its own misses compounds.
Go deeper

Tools & references

The budget allocatorSaturation curves to make your funnel bend realistically as spend scales. Source: this site.
Scrappy MMMWhere the response curves that feed the funnel's CVR assumptions come from. Source: this site.
The measurement field guideHow forecasting connects to attribution, MMM, and incrementality. Source: this site.
Cohort & retention analysisThe backward-looking base for the LTV lens — survival/retention curves by signup cohort. Standard practice in any subscription analytics stack.
Keep going

The rest of the stack

Pillar

How to measure in 2026

The triangulation framework this forecast borrows its logic from.

Read the field guide →
Method

Scrappy MMM

The directional model that gives your funnel its saturation curves.

Read it →
🎛️ Free tool

The budget allocator

Response-curve optimization you can drive by hand.

Open the tool →
Method

The experimentation agenda

How to choose the handful of clean tests you can run in a year.

Read it →
Quick answers

Common questions

Why forecast a range instead of a single number?
A single number hides both its assumptions and its uncertainty, so a miss looks like a failure rather than a known risk that materialized. A range forces you to state the assumptions behind the low, base, and high cases, makes the uncertainty explicit, and gives decision-makers something honest to plan against.
What are the three forecasting lenses?
Bottoms-up funnel (spend to installs to signups to subscriptions, built from conversion rates), top-down (market size, seasonality, and trend), and cohort/LTV (projecting forward from how existing cohorts retain and expand). They have different blind spots, so where they converge is your confidence and where they diverge is your risk.
How do you turn three forecasts into one range?
Use the spread between the lenses to set the band: the lowest credible lens anchors the low case, the consensus anchors the base, and the most optimistic credible lens anchors the high. When they agree, the range is tight. When they diverge sharply, that's information — you investigate the disagreement rather than averaging it away.
How do you make forecasts more accurate over time?
Track forecast error every cycle. Log what you predicted versus what happened, separate bias from variance, and trace each miss back to the specific assumption that broke. Forecasts that record and learn from their own error compound in accuracy; forecasts that don't repeat the same mistakes.
This is how I operate

I build forecasts you can actually defend

If your number is getting questioned in the room — or you want a three-lens range instead of a single brittle figure — let's talk.

Read the measurement guide → Work with me

Last updated June 2026 · Part of an in-progress series on growth measurement & budget allocation.