Build a Variance Analysis Model

This guide explains how to perform variance analysis using Model Reef by comparing separate models and periods, rather than relying on in-model lag functions or multi state scenarios.

Because each scenario or version is a separate model, variance analysis is an exercise in structured comparison rather than a special engine feature.

Before you start

You should have:

  • A Budget model or prior forecast model.

  • An Actuals or Latest Forecast model for the same scope.

  • Consistent period settings and broadly aligned structures across the models.

If you have not yet built budget or actuals models, see:

  • Build a Budget vs Actuals Model

  • Build a Multi Scenario Comparison (A vs B vs C)

What you will build

  • A simple structure for comparing two or more models.

  • A variance view for key metrics such as revenue, EBITDA and cash.

  • A process for explaining differences in terms of drivers and assumptions.

Step 1: Select models to compare

Decide which models you are comparing, for example:

  • Budget vs Actuals.

  • Previous plan vs Updated plan.

  • Base Case vs Downside.

Confirm that:

  • They cover the same time horizon.

  • They are expressed in the same currency and periodicity.

  • Their category structures are close enough for comparison.

Step 2: Extract key metrics from each model

1

Export or record key series

For each chosen model, export or record key series for the relevant periods, for example:

  • Revenue.

  • COGS and gross profit.

  • Opex and staff costs.

  • EBITDA.

  • Cash and debt.

2

Store and label

Store these series in an external file or in a structured note, clearly labelled by model and period.

You now have the raw data needed for variance analysis.

Step 3: Calculate variances

Outside the engine, compute:

  • Absolute variance:

    • Actual minus Budget.

  • Percentage variance:

    • (Actual minus Budget) divided by Budget.

You can apply the same method to any pair of models, for example Updated plan versus Prior plan.

Keep the sign convention consistent so that positive always means better or higher, depending on your preference.

Step 4: Attribute variances to drivers conceptually

Model Reef does not perform automated driver based variance attribution, but you can interpret differences by looking at the underlying assumptions, for example:

  • Revenue variance:

    • Volume higher or lower than planned.

    • Price changes.

  • Margin variance:

    • Changes in cost per unit.

    • Mix between products or segments.

  • Opex variance:

    • Delayed or accelerated hiring.

    • Higher marketing spend.

  • Cash variance:

    • Different working capital timing.

    • Changes in capex or funding.

Use the variable and driver views inside each model to understand how differences arose.

Step 5: Summarise variance findings for decision makers

Produce a short summary that answers:

  • Where did we diverge from plan.

  • Which drivers explain most of the variance.

  • Which of those are likely to persist versus one off.

  • What actions you propose as a result.

You can present this alongside charts and tables generated from the models.

Step 6: Feed learnings back into future models

Variance analysis is only helpful if it improves future planning. Use insights to:

  • Adjust driver and variable design where they did not capture reality well.

  • Revise assumptions in future budgets.

  • Build better downside and stress scenarios.

Over time, this should improve both forecast accuracy and the usefulness of your model for decision making.

Check your work

  • Models used in comparisons are clearly identified and documented.

  • Variances are computed correctly and consistently.

  • Explanations map back to specific drivers or assumptions in the models.

  • The analysis leads to practical recommendations rather than just observations.

Troubleshooting

chevron-rightVariance analysis becomes very detailed and hard to communicatehashtag

Focus first on the largest few variances and their main drivers. Detail can sit in an appendix.

chevron-rightDifferent structures make comparison hardhashtag

Align structures in future models, and for now aggregate metrics to a level where structures are comparable.

chevron-rightStakeholders misread the sign of varianceshashtag

Standardise your variance sign convention and explain it clearly in the report.

Last updated