Variance Analysis by Account Type: One Method Does Not Fit All

Most variance reports apply one rule to every line: subtract budget from actual, flag anything over a threshold, sort by size. It feels rigorous. It is actually a category error, because a P&L is not a list of interchangeable numbers — it is a stack of lines with completely different behaviours. A materials line should move with revenue. A rent line should not move at all. A commissions line should move with sales but at a fixed rate. Reading all three with the same subtraction treats normal behaviour as news and real news as noise.

We’ve written before about what separates a variance number from variance analysis — decomposition, timing versus trend, materiality, cause attached to effect. This piece extends that discipline one level down: the method of decomposition, and even the definition of “material”, should change depending on what kind of line you are looking at. Here is the framework, line type by line type.

Variable costs: judge the rate, not the rupees

Cost of goods, direct materials, transaction fees — anything that scales with activity — should never be judged on the absolute variance alone, because the absolute variance is contaminated by volume.

A worked example. Materials were budgeted at ₹22,00,000 against budgeted revenue of ₹55,00,000 — a 40% materials rate. Actuals come in at ₹26,00,000. The variance column prints ₹4,00,000 adverse, it is the largest number on the page, and it goes to the top of the flag list.

But revenue came in 20% above budget. At the budgeted rate, 20% more volume should cost ₹26,40,000 of materials. Actual spend of ₹26,00,000 is below that — the materials rate improved from 40.0% to roughly 39.4%. The line that looked like the month’s biggest problem is in fact mildly good news: purchasing held the rate while volume surged. The correct read is a large favourable volume variance on revenue, offset by proportionate materials cost, with a small favourable rate variance on top.

For variable lines, the first move is always to restate the budget at actual volume (a flexed budget) and measure the variance against that. The residual is the rate variance, and the rate variance is the only part that says anything about how well the line was managed.

Semi-variable costs: split the volume from the rate, explicitly

Freight, sales commissions, packaging, payment-gateway charges — lines that move with activity but where the rate is itself a managed quantity. These need the volume/rate split made explicit every period, because either component can hide the other.

Suppose freight was budgeted at ₹3,00,000 and lands at ₹3,90,000 — 30% adverse, apparently alarming. Shipment volumes were up 15%, which explains ₹45,000 of the gap. The remaining ₹45,000 is a genuine rate increase: the freight cost per shipment rose about 13%. Now the question is precise — did the carrier raise prices, did the shipment mix move towards heavier or more distant deliveries, or did expedited shipping creep up because the warehouse kept missing cut-offs? Each answer belongs to a different owner. A single “freight over budget” flag belongs to nobody.

Commissions deserve the same treatment in reverse. Commissions exactly on budget while revenue ran 20% ahead is not a clean line — it means the effective commission rate fell, and someone should be able to say why (mix shift towards house accounts, a threshold structure, or a booking lag that will catch up next month).

Step-fixed costs: a variance is an event, not a trend

Salaries within a team, software licences priced in seat bands, supervisory roles — these costs do not glide with volume. They sit flat, then jump when a threshold is crossed: a hire, a new band, an added shift.

That changes what a variance means. If the customer-support salary line steps from ₹8,00,000 to ₹9,20,000 in June, the wrong instinct is to annualise it as a trend or average it into a run rate. It is neither. It is an event — two support hires started in June — and the analysis has exactly two questions: was the event planned (in which case the variance is a phasing note, not a problem), and what is the new plateau (₹9,20,000 is the base from July onwards, and the budget comparison should acknowledge that).

Step-fixed lines are also where “under budget” needs the most suspicion. Being ₹1,20,000 favourable on salaries usually means a planned hire has not happened yet — which is not saved money but delayed capability, and often a warning about next quarter’s output rather than a win in this one.

Fixed and contractual costs: small variances are the interesting ones

Rent, insurance, subscription contracts, audit fees. These are set by document, not by activity, so the expected variance is precisely zero. That inverts the usual logic: on most lines a bigger variance deserves more attention, but on a contractual line any variance is anomalous, and small ones are, if anything, more interesting than large ones.

A large variance on rent tends to have a visible cause — a new office, a renewal, an escalation clause landing. It gets noticed and explained. A rent line that runs ₹15,000 over for three quiet months is the one worth pulling, because rent does not drift. Something mechanical is wrong: a duplicated invoice, a miscoded cost, an escalation applied from the wrong date, a deposit booked as expense. Until proven otherwise, a variance on a contractual line is a timing or booking error, not a business event — and booking errors found early are cheap, while booking errors found at audit are not.

The practical rule: on contractual lines, set the flag threshold near zero and treat every flag as a data-quality question first and a spending question second.

Revenue: decompose into volume, price and mix before reacting

Revenue variance is the line most likely to be discussed and least likely to be decomposed. “Revenue ₹6,00,000 ahead of budget” supports whatever story the room wants to tell until it is split three ways: did we sell more units (volume), at better prices (price), or more of the profitable things (mix)?

The split matters because the three components can point in opposite directions inside a favourable total. Volume up 12%, average price down 4%, mix shifted towards a low-margin product line — that combination can beat the revenue budget while quietly eroding margin, and the margin variance two lines down will look mysterious unless the revenue decomposition travels with it. A revenue variance reported without its volume/price/mix split is a headline without an article.

Allocated lines: check the driver before the spend

Department P&Ls carry lines nobody in the department spent: allocated rent, shared services, apportioned leadership costs. These have a failure mode all their own — the variance can be real at the line level while nothing changed in the underlying cost at all.

If marketing’s allocated rent rises ₹40,000, there are two candidate causes. Either the building costs more, or the allocation basis moved — marketing hired three people, rent is apportioned by headcount, and their share of an unchanged total grew. The second is far more common, and it calls for a completely different conversation: not “why are we spending more” but “the driver moved, as designed”. So for any allocated line, the first check is the driver, not the spend — and that presumes the allocation method is written down and applied identically every period, because a variance against an inconsistent method is not information about anything.

Materiality should vary by line nature too

The final consequence of all this: a single materiality threshold across the P&L is miscalibrated almost everywhere. A sensible scheme sets thresholds by line behaviour, not just line size:

• Variable lines — threshold on the rate, not the rupees. Flag when cost as a percentage of revenue moves more than, say, half a point; ignore absolute swings explained by volume.
• Semi-variable lines — threshold on the rate component after the volume split.
• Step-fixed lines — flag any step, regardless of size, and annotate it as planned or unplanned.
• Contractual lines — threshold near zero; every variance is a query.
• Allocated lines — flag driver movements as well as cost movements.

This is what keeps the flags honest: each line is measured against what it could reasonably have been expected to do, rather than against a uniform tolerance that is too loose for rent and too tight for materials.

The decision guide: the first question, by account type

When a variance lands in front of you, the discipline is to ask the right first question for the kind of line it sits on:

• Variable (COGS, materials, transaction costs): what is the variance against the volume-flexed budget? Judge the rate, ignore volume-driven rupees.
• Semi-variable (freight, commissions, packaging): how much is volume, how much is rate? Only the rate component has an owner.
• Step-fixed (team salaries, seat-banded licences): what event caused the step, was it planned, and what is the new base?
• Fixed / contractual (rent, insurance, subscriptions): is this a booking or timing error? Assume yes until the document says otherwise.
• Revenue: split volume, price and mix before anyone tells a story about the total.
• Allocated (shared costs, recharges): did the driver move, or did the spend? Check the basis first.

Six line types, six first questions. None of them is “how big is the number” — because on a P&L read properly, the size of a variance is the last thing that tells you what it means.

Datavrn is management-reporting software for finance teams, virtual CFO practices and enterprise finance functions. Available now: for one standalone Entity, take a Trial Balance through confirmed groupings and disclosures to an Excel working draft with live links or a clean finalised workbook with fixed values. Datavrn prepares the statements; it does not file or certify them. The wider management pack is in active development. Start free with your work email — no invitation needed.