Budget variance analysis: what actually moved versus plan

You set a budget. The quarter closes. Actuals come in different. The gap between the two is a variance, and explaining it is one of the most common asks in any finance or operating review. Most people do it badly. They paste a table of plan, actual and variance, highlight the big red numbers and call it analysis. That is reporting, not analysis.

Budget variance analysis is the work of explaining what actually moved versus plan and why. Done properly it tells the reader where reality diverged from the assumptions, which of those gaps matter and what is driving each one. Done properly it changes a decision. This is the method, step by step, and where an AI analyst speeds it up without inventing a single number.

What budget variance analysis is and why it matters

A budget is a set of assumptions written as numbers. Volumes, prices, costs, timing. When actuals land, every assumption is either confirmed or wrong. Variance analysis is the systematic comparison of actual results against the plan, line by line, to surface which assumptions held and which broke.

It matters because the variance is your early warning system. A budget that is on track in total can be quietly broken underneath. Revenue might be exactly on plan because a price increase you pushed through is masking a fall in units you have not noticed. The total looks calm. Two real problems are hiding inside it. Variance analysis is how you find them before they compound.

Actual vs plan, favourable vs adverse

Two pairs of words do most of the work. The first is actual versus plan. Plan is what you committed to. Actual is what happened. The variance is the difference, and you state it both in absolute terms and as a percentage, because a £50k variance means one thing on a £200k line and nothing on a £20m line.

The second pair is favourable versus adverse. A favourable variance improves your result. Higher revenue than planned, lower cost than planned. An adverse variance worsens it. The trap is treating the sign of the number as the label. A cost that comes in higher than plan is a positive number in your variance column but an adverse outcome. Always label by effect on profit, not by arithmetic sign, or your narrative will read backwards.

Rule of thumb: never report a variance as just a number. Report it as actual vs plan, in absolute and percentage terms, tagged favourable or adverse by its effect on profit.

Decompose the variance into price, volume and mix

This is the step that separates analysis from reporting. A revenue variance is almost never one thing. It is the net of three forces, and until you split them you cannot say what moved.

• Price variance: you sold at a different price than planned. Hold volume and mix at plan, flex price to actual, and the difference is your price effect.
• Volume variance: you sold a different quantity than planned. Hold price and mix at plan, flex volume to actual, and the difference is your volume effect.
• Mix variance: you sold a different blend of products or segments than planned. Even if total units matched plan, selling more low-margin and fewer high-margin items shifts the result. That shift is mix.

The standard approach is to flex one factor at a time so each effect is isolated and the three add back to the total variance. Price effect is the change in price multiplied by actual volume. Volume effect is the change in volume multiplied by the planned price. Mix effect is what is left once price and pure volume are accounted for, the result of the basket of what sold being different from what you assumed. The same logic applies on the cost side: an input cost variance splits into the rate you paid and the quantity you used.

Now you can write a real sentence. Not "revenue was £180k under plan" but "revenue was £180k under plan: volume cost us £260k as units fell 9%, partly offset by a £90k price gain from the April increase, with a small adverse mix as the cheaper tier grew". That is what actually moved.

Set a materiality threshold so you focus on what counts

A full chart of accounts can have hundreds of lines, every one with a non-zero variance. Explaining all of them is a waste of everyone's time and buries the three that matter under noise. Set a materiality threshold before you start, and only investigate variances above it.

A workable threshold combines two tests, and a line clears the bar if it passes either:

1. 1.An absolute floor. A figure below which a variance is not worth a sentence, set relative to the size of the result. On a £10m budget that might be £50k.
2. 2.A percentage trigger. A line that moved more than, say, 10% versus plan, even if the absolute number is small, because a large proportional swing often signals a broken assumption worth catching early.

Materiality is a judgement, not a formula, and the right level depends on your audience and the decision at hand. The point is to set it deliberately and apply it consistently, so the analysis you present is the handful of variances that change something, not a wall of every line that moved.

Build the 'what moved and why' narrative

Numbers do not explain themselves. Once you have decomposed the material variances, the deliverable is a short narrative that walks the reader from the headline gap to its causes. Structure each material variance the same way:

1. 1.The headline. The total gap, actual vs plan, favourable or adverse, in absolute and percentage terms.
2. 2.The decomposition. How much of the gap is price, how much is volume, how much is mix. This is the 'what moved'.
3. 3.The cause. The business reason behind each effect. A delayed launch, a competitor price cut, a supplier increase, a channel shift. This is the 'why', and it is the only part the numbers cannot give you.
4. 4.The implication. What it means going forward and what, if anything, to do about it.

Lead with the largest material variance and work down. Quantify everything, attribute every effect and stop at the materiality line. A reader should be able to recreate your headline number from the parts you give them, which is the test that you have explained the variance rather than just described it.

Common traps that quietly ruin the analysis

Most variance analysis fails in predictable ways. Watch for these.

• Offsetting variances that net to zero. A line that is on plan in total can hide a large favourable variance cancelling a large adverse one. Two real problems, both invisible, because you only looked at the net. Always decompose before you conclude a line is fine.
• Ignoring mix. Total volume on plan and total revenue on plan does not mean nothing happened. If the blend of what sold shifted, margin moved even when the top line did not. Mix is the variance people skip and regret.
• Comparing against the wrong baseline. Variance is only meaningful against the plan you actually committed to. Comparing against last year, against a reforecast, or against a stale version of the budget produces a number that answers a different question than the one being asked.
• Restating prior periods. If you quietly reclassify or restate the budget to make the variance smaller, you break the comparison. The plan you measure against has to be the plan as it was set. Adjust the presentation, never the baseline.

How Nexlyr AI computes the variance from your files

The method above is the same whoever does it. What slows it down is the manual work: reconciling an actuals export against a budget file, building the flex calculations, splitting price, volume and mix without an arithmetic slip, then writing it all up. This is where Nexlyr AI fits.

You give it your actuals and your budget files, a spreadsheet, a PDF or a document and a short brief. It reads the data and computes on it with a real analysis engine that writes SQL and runs it against your actual numbers. The variance is calculated from your source, line by line, not estimated by a language model guessing at what looks right. The figures are grounded in your own data, and if a figure is not supported by the source it is not shown.

From there it drafts the 'what moved and why' narrative for you, with the decomposition done and every figure traceable back to the file it came from. It applies standard variance frameworks where the material supports them, never forced, and a post-build review pass reads the finished analysis like an analyst would, raising the questions and risks worth checking, including the what-ifs you would want to test. The output is a fully editable, branded PowerPoint deck you can take into the review and change as you go.

The point is operational trust. Because every variance is computed from your own actuals and budget, and nothing is invented, you can stand behind the numbers in the room without re-checking each one by hand.

You still own the judgement. Materiality is your call, the business causes are yours to confirm and the recommendation is yours to make. The analysis engine does the exact, slow, error-prone arithmetic and gives you a grounded first draft, so you spend your time on the part that needs a human: deciding what it means.

Keep exploring

• Writing an investment memo that stands up to scrutiny
• Market-entry assessment: which frameworks, in what order
• Budget variance analysis