A project came in $22K over budget. The account lead says the client kept changing scope. The delivery lead says the estimate was low. Both are guessing. That single overrun number cannot settle the argument, because it hides two different problems inside one figure. Cost variance analysis is how you pull them apart. It splits the overrun into a price effect and a usage effect — each with a distinct owner and a distinct fix. Done well, it turns "the project went sideways" into "we paid 8% more per hour and used 40 hours we didn't plan for."
That is the difference between blaming a project and fixing the next one.
One number is a symptom, not a diagnosis
The total cost variance is just actual cost minus budgeted cost. On our example project, that's $22K unfavorable. Useful as an alarm. Useless as an instruction.
The problem is that two very different failures produce the identical $22K. Maybe your senior rate rose and you staffed the plan exactly as designed. Maybe your rates held and the team burned hours it never budgeted. Same total. Opposite causes. Opposite owners.
Project accounting exists to separate these threads by tying every dollar to a specific project, task, and resource. As Oracle NetSuite frames it, the discipline is about attributing costs precisely enough to explain them. A total variance skips that step. It reports the weather without telling you which way the wind blew.
Split the overrun into price and usage
Every cost variance decomposes into two parts. Learn this once and it changes how you read every over-budget project.
Price (rate) variance measures what you paid per unit versus what you planned. For a services firm, the unit is usually an hour.
- Formula: (actual rate − budgeted rate) × actual hours
Usage (efficiency) variance measures how many units you consumed versus plan, priced at the budgeted rate.
- Formula: (actual hours − budgeted hours) × budgeted rate
Run the numbers on our project. Budget: 400 hours at $150 = $60K. Actual: 440 hours at $155 = $68,200. Total variance: $8,200 unfavorable (a smaller, cleaner example than the messy one above — use real project data when you do this).
- Price variance: ($155 − $150) × 440 = $2,200 unfavorable
- Usage variance: (440 − 400) × $150 = $6,000 unfavorable
Now the diagnosis writes itself. Three-quarters of this overrun is hours, not rate. This is a scope and estimation problem, not a staffing-cost problem. If you'd assumed "the rates blew up," you'd have fixed the wrong thing. The clean math behind your labor cost rate is what makes this split trustworthy.
Every variance needs an owner
A number without an owner is a complaint. The reason to split price from usage is that they belong to different people.
Price variance is a resourcing and pricing question. Did you staff a senior where a mid-level was budgeted? Did a subcontractor rate change? That sits with whoever controls staffing mix and rate cards — usually a practice lead or ops. Deltek is blunt that project accounting only works when costs map to accountable roles, not just cost codes.
Usage variance is a scope and delivery question. Did the client expand the ask? Did the estimate miss a workstream? That sits with the delivery lead and the account owner. Scoro ties this directly to how firms scope and track work against plan.
When you assign each half to the person who can actually move it, the post-mortem stops being a blame session and becomes a work list.
The raw number lies without thresholds and CV%
Here is the trap that catches disciplined teams: reacting to variances that don't matter. A $600 overrun on a $400K engagement is noise. A $600 overrun on a $4K fixed-fee project is a 15% blowout. The absolute number is the same. The signal could not be more different.
Two guardrails fix this.
Set a materiality threshold. Below a dollar or percentage floor, you log the variance and move on. This keeps your attention on the projects that can actually hurt margin. AICPA & CIMA treats this kind of prioritization as core practice management, not optional polish.
Always report cost variance as a percentage (CV%). CV% = variance ÷ budgeted cost. It normalizes across project sizes so a big project can't drown out a small one bleeding proportionally faster. NetSuite's guide to consulting KPIs puts percentage-based margin and variance metrics at the center of how healthy firms watch their book.
Rank your projects by CV%, not raw dollars. The order changes, and so does where you spend your Monday.
Feed the diagnosis forward, not just backward
Cost variance analysis earns its keep when it stops being an autopsy and starts being an early warning. The split you run at month-end is the same split you should run mid-project.
If your usage variance is already unfavorable at 40% delivery, your remaining budget is a fantasy. Recast it. This is exactly what an estimate-to-complete does — it takes what you've learned about actual burn and reprices the rest of the engagement. A live variance split makes that recast honest instead of hopeful.
The pattern also compounds across projects. A recurring price variance means your rate card is stale. A recurring usage variance in one service line means your estimating model for that work is wrong. Harvard Business Review argues that thriving services firms treat this kind of operational feedback as a strategic asset, not back-office housekeeping. Fix the model once and every future project benefits.
To connect variance back to profitability, pair it with your project margin calculation and choosing the right job costing method for how you deliver. The full picture lives on the project costing hub and the broader professional services finance pillar. If you want to model the margin impact quickly, the margin calculator gets you there, and the contribution margin glossary entry covers the underlying math.
The takeaway
A budget overrun is a headline. Cost variance analysis is the story underneath it. Split every overrun into price and usage. Give each half an owner who can act. Filter with a materiality threshold and rank by CV% so size can't fool you. Then feed the answer into your next estimate.
This is the kind of split a founder should see the moment it happens, not four weeks later at close. CentSight is the intelligence layer on top of QuickBooks and your bank — it reads your live numbers and surfaces the variance while you can still change the outcome. If what it flags confirms what you already knew, you're out $95. If it doesn't, it just found the margin you were leaking.
