Control Charts in FP&A: Which Budget Variances Are Real?

Every month, finance teams across the world ask managers to explain variances. Maintenance is 9% under budget: why? Freight is 12% over: why? The managers write commentary, often plausible, sometimes inventive, and next month the variance reverses on its own. A large part of this effort is spent explaining noise.

Manufacturing solved the same problem a century ago. In the 1920s, Walter Shewhart at Bell Telephone Laboratories developed the control chart to separate the routine variation that every process produces from the exceptional variation that has a specific cause. W. Edwards Deming carried the method to Japanese industry after the war, and Donald Wheeler's book Understanding Variation showed managers how to apply it to the numbers in their monthly reports.

Two kinds of variation

• Routine variation comes from many small causes that are always present: the timing of invoices, the mix of jobs, the weather. It is the voice of the process. Reacting to it is not only useless but harmful, because adjusting a stable process in response to noise makes it more variable.
• Exceptional variation has an assignable cause: a new contract, a breakdown, a price change, an error. It deserves an explanation and, often, an action.

A budget variance cannot tell them apart, because the budget is a target, not a description of the process. A control chart can, because its limits are calculated from the process's own history.

The XmR chart

The simplest version, the individuals and moving range (XmR) chart, needs nothing more than a monthly series. Take a baseline period, compute the average and the average absolute change from one month to the next (the moving range), and draw natural process limits at the average plus and minus 2.66 times the average moving range. Values inside the limits are routine; a value outside them, or a run of eight values on the same side of the average, signals that something has changed.

In the first twenty months, this cost line moved between 457 and 559 thousand against a budget of 500. A traditional report would have demanded explanations for month 14 (9% under) and month 15 (12% over). The chart shows that both are within the natural limits of 434 to 580: they are the process being itself. From month 21, something different happens. No single month looks alarming at first, but the values stay high, month 25 breaks the upper limit, and by month 28 eight consecutive points above the average confirm a shift of about 55 thousand a month. That is the variance worth investigating, and the chart dates its start precisely.

What changes in the monthly review

1. Ask for explanations only for signals. Points outside the limits and sustained runs. Everything else is reported, not explained.
2. Separate the budget question from the process question. If a stable process runs above budget, the budget was unrealistic or the process needs redesign; monthly commentary will not fix either.
3. Recalculate limits after a confirmed shift, so that the new level becomes the new baseline.
4. Use the width of the limits as information. A process whose routine swing is ±15% cannot be managed to ±5% by commentary; it needs structural change.