Budget vs. actual — analyzing spend variances in natural language with a Dust agent
Budget vs. actual analysis by cost center typically takes a morning in Excel. With a Dust agent connected to the Spendesk MCP, pulling the data, calculating variances, and identifying significant overruns takes 30 seconds — shared with the whole FP&A team.
The problem
Every quarter, same scenario: pull the Spendesk export, open it in Excel, build pivot tables by cost center, calculate variances against budget, then spend an hour investigating abnormal lines.
And if someone asks for a variation — by department, by category, by quarter — you're back to another 45 minutes.
One limitation to know before you start
The Spendesk MCP has no get_budgets tool: Spendesk isn't a budgeting system, it's a spend management system. There's no native budget-vs-actual comparison. What a Dust agent can do is an approximation via spendesk_analyze_spend: actual spend is pulled live from Spendesk, but the budget figures have to be supplied upfront — in the prompt, in a document shared with the agent, or in the agent's own instructions. The agent doesn't "know" your Q2 budget; you have to give it that number (or have already provided it via its instructions or an attached document).
Budget vs. Actual analysis in a single prompt
@Dust here's the Q2 budget by cost center: Engineering €845,000, Marketing €320,000, Sales €180,000, G&A €95,000, R&D €210,000. Compare against actuals and identify significant variances (> 10%).
Budget vs. Actual Analysis — Q2 2026 by Cost Center
| Cost Center | Budget | Actual | Variance € | Variance % |
|---|---|---|---|---|
| Engineering | €845,000 | €891,200 | +€46,200 | +5.5% |
| Marketing | €320,000 | €398,700 | +€78,700 | +24.6% |
| Sales | €180,000 | €165,400 | -€14,600 | -8.1% |
| G&A | €95,000 | €94,100 | -€900 | -0.9% |
| R&D | €210,000 | €187,300 | -€22,700 | -10.8% |
| Total | €1,650,000 | €1,736,700 | +€86,700 | +5.3% |
3 variances to investigate (> 10%)
Note: the budget figures above come from what you gave me — the actuals come straight from Spendesk. Would you like me to analyze Marketing or R&D detail first?
Investigating the Marketing overrun
The previous response identifies the problem. This question finds the cause — in the same Dust conversation, or by mentioning @Dust again in Slack.
@Dust dig into the Marketing overrun. What spending explains the +€78,700?
Marketing Q2 2026 overrun breakdown (+€78,700)
Primary driver — Digital advertising (+€53,000)
| Supplier | Q2 Budget | Q2 Actual | Variance |
|---|---|---|---|
| LinkedIn Ads | €45,000 | €76,200 | +€31,200 |
| Google Ads | €38,000 | €58,100 | +€20,100 |
| Meta Ads | €12,000 | €13,700 | +€1,700 |
Secondary driver — Unbudgeted events (+€22,000)
Offsetting savings (-€2,700)
Summary: 68% of the overrun came from digital (rising CPL), 28% from a last-minute event.
Would you like a summary slide for the board, or a monthly trend analysis for LinkedIn?
Three other everyday use cases for FP&A
Tracking SaaS and recurring subscriptions
@Dust which SaaS subscriptions have increased by more than 20% compared to last year?
SaaS subscriptions with > 20% increase — Jul 2025 vs. Jul 2026 comparison
| Tool | July 2025 | July 2026 | Increase |
|---|---|---|---|
| Datadog | €8,200/mo | €14,800/mo | +80% |
| Figma (Enterprise) | €1,200/mo | €2,100/mo | +75% |
| Notion | €890/mo | €1,340/mo | +51% |
| Slack | €2,400/mo | €3,200/mo | +33% |
| Salesforce | €9,800/mo | €12,400/mo | +27% |
Annualized impact of increases: +€165,600/year
Spend per employee for benchmarking
@Dust what is the average spend per employee in Engineering vs. Marketing this quarter?
Spend per employee — Q2 2026
| Department | Headcount | Total spend | Spend/employee |
|---|---|---|---|
| Engineering | 48 FTE | €891,200 | €18,567/employee |
| Marketing | 12 FTE | €398,700 | €33,225/employee |
| Sales | 22 FTE | €165,400 | €7,518/employee |
| G&A | 8 FTE | €94,100 | €11,763/employee |
Would you like me to project these ratios against the H2 budget accounting for hiring plans?
A shared FP&A agent, not a personal spreadsheet
Published on the workspace, the Dust agent stays accessible to anyone on the FP&A team who needs to ask the same question — no spreadsheet to track down, no stale version circulating by email. To turn this analysis into an automatic monthly review posted in the #fpa channel, see the guide on Dust Triggers.
Key takeaways
- The Spendesk MCP has no native budget-vs-actual comparison — it's an approximation via
spendesk_analyze_spend, built from budget figures you supply - Budget vs. Actual analysis goes from 3 hours to 30 seconds once the agent has the budget numbers
- The agent automatically identifies variances and suggests likely causes
- Each answer can generate the next question — investigation in conversational mode, in Dust or in Slack
- A published agent prevents a budget-vs-actual spreadsheet from circulating in multiple versions by email
Related skills
Analyze spending over periods exceeding 366 days by splitting into sub-periods and aggregating results.
Break down spending by cost center with per-employee detail, resolving cost center names from IDs.
Analyze month-over-month spend trends by cost center, detect accelerating or declining budgets, and flag anomalies.