Web Analytics · Marketing Measurement

Marketing Forecasting & Scenario Planning

A range with named drivers behind it plans better than one confident number.

A single-point forecast hides the assumptions carrying the plan. We build a range, show which drivers move it, and model scenarios around the assumptions most likely to shift.

A backtested forecast with a defensible range, named drivers, and scenarios tied to the planning decision.

Plan a marketing forecastAll Marketing Measurement & Attribution tasks
A Zeo navigator plotting three scenario routes on a horizon chart

Some of the 500+ brands we've worked with

See all references
  • Decathlon
  • Watsons
  • D&R
  • Doğtaş
  • Cyberpark
  • Jollytur

How we run it

The model must hold up against historical periods withheld from its inputs.

The model shows how its assumptions shape the range instead of presenting one precise-looking answer. Four steps connect the planning question to a forecast range, scenarios, and refresh triggers.

How we hold ourselves to it

  • Establish the baseline from real history — We establish the baseline from your historical performance, including known seasonality and past interventions.
  • Identify the drivers that matter — We isolate the two or three assumptions, such as spend, conversion rate, or market conditions, that move the forecast most.
  • Build named scenarios — We produce a base case with upside and downside scenarios, each tied to specific named assumptions.
  • Backtest before you trust it — We test the model against periods with known outcomes to show how it would have performed historically.

  1. Frame the planning question

    We agree what decision this forecast needs to support and over what time horizon.

    Planning brief

    AI assist
    Drafts the planning question and horizon from stakeholder notes.
    Human gate
    Planning owner confirms this is the decision being forecast.
    Owners
    Measurement Analyst, Planning Owner
    Illustrated figure holding up a signed agreement page

  2. Build and backtest the model

    We build the model and backtest it against historical periods before using it for planning.

    Backtest results

    AI assist
    Runs the baseline model against periods with known outcomes.
    Human gate
    Analyst checks the backtest error against an accepted threshold.
    Owners
    Measurement Analyst
    Illustrated figure reading an oversized measurement dial

  3. Define scenarios

    We identify the key drivers and define named scenarios around plausible ranges for each one.

    Scenario book

    AI assist
    Drafts upside and downside ranges around each named driver.
    Human gate
    Planning owner confirms the scenario ranges are plausible.
    Owners
    Measurement Analyst, Planning Owner
    Illustrated figure sketching plans at a drafting table

  4. Deliver and set refresh triggers

    We present the forecast and agree which changes, such as a platform shift, new competitor, or pricing change, should trigger a refresh.

    Forecast readout

    AI assist
    Drafts candidate refresh triggers from the named key drivers.
    Human gate
    Leadership agrees which triggers force a forecast refresh.
    Owners
    Planning Owner, Finance Lead
    Illustrated figure turning a loop of arrows above server racks

We build the forecast around named drivers and a plausible range.

Automation frames the planning question from the notes, runs the baseline against periods with known outcomes, drafts upside and downside ranges per driver, and proposes refresh triggers. The judgments belong to your side and ours together: what decision is being forecast, whether the backtest error is acceptable, and which ranges are actually plausible.

What you get

You get a planning range with the assumptions exposed.

The deliverables show the range, the scenarios behind it, and how the model performed in backtesting.

  • Working document

    Forecast model

    The reproducible model and its baseline, documented well enough to update later.

    Accepted when

    Someone else can update it with next month's data without rebuilding it.

    Cadence: Updated every planning cycle

  • Reference document

    Scenario book

    Base, upside, and downside scenarios tied to named, specific assumptions.

    Accepted when

    Each scenario names the assumption that moves it, not just a percentage.

    Cadence: Revised when an assumption moves

  • QA notes

    Backtest results

    How the model would have performed against periods you already know the outcome for.

    Accepted when

    The error is reported against periods whose outcome was already known when the model was built.

    Cadence: Rerun each planning cycle

We call it done when: the model has been backtested against periods with known outcomes, the scenarios bracket a plausible range, and the refresh triggers are agreed.

Fit and readiness

One confident number can hide every assumption carrying the forecast.

This work fits a defined planning decision backed by enough consistent historical data.

A good fit when

  • You need a forecast for a hiring, budget, or inventory decision, and a single-point number does not give you enough to plan around.
  • Past forecasts have missed actual results badly enough that leadership no longer trusts a single-point estimate for the next planning decision.
  • You want to see how spend level, conversion rate, or seasonality changes the plan through named scenarios.

Better handled as other work when

  • You need daily budget pacing and bid management. That responsibility stays with your paid-media team, while this work builds the forecast.
  • Your historical data is too broken or inconsistent to support a forecast. A GA4 audit or reporting validation should come first.

If one of these is closer to your situation, start here instead: All Marketing Measurement & Attribution tasks

We call it done when: the planning owner has confirmed the decision the forecast informs, and everyone is using the same planning horizon.

People who build your measurement system

Zeo designs measurement systems that connect a business decision to governed collection and reporting you can check. The people shown here work on the part of that system this page covers.

Tools we use

Tools behind this work

  • Jupyter

    runs the backtest against withheld history so the model is judged on what it never saw

  • Prophet

    the forecasting library that outputs a range by default, not a point estimate dressed up as one

  • Looker Studio

    carries the delivered range and named scenarios past the initial handover, watched for refresh triggers

Next step

Plan with a range and named scenarios

Tell us the decision, planning horizon, and historical data available. We will build a forecast range with scenarios tied to the assumptions that matter most.
Plan a marketing forecast

Before we start

Questions teams ask before booking

A single number implies more certainty than any forecast actually has. A range with named drivers tells you what to watch and what would change the picture, which is more useful for planning than false precision.