Variance analysis is where reporting earns its keep, and where most models show their weaknesses. The difference between a fast month-end and a slow one is usually structural.
Model first
Actuals and plan must share dimensions and grain. If budget lives in a separate table with different keys, every variance measure becomes a workaround. Transaction keys on plan rows solve this permanently.
Measure design
- One base measure per fact, then variance and variance-percent derived from them.
- Handle divide-by-zero explicitly rather than in the visual.
- Sign conventions agreed once: is a cost overspend positive or negative?
- Separate measures for budget, latest forecast and prior year.
Visual design
Lead with the variance, not the components. Rank by absolute impact, colour only the exceptions, and keep the detail one drill away rather than on the page.
A good variance page answers what changed before it shows how much of everything there is.
Close the loop with commentary
A variance without an explanation generates a meeting. accoCOMMENT stores the explanation against the exact figure, with approvals and history, so the next reader gets the answer without the meeting.
Keep the plan current
Variance against a stale budget loses meaning by Q3. accoPLANNING lets owners reforecast in the same report, so comparisons stay against a plan someone still stands behind.
