- Home
- Skills
- Finance & Accounting
- SaaS MRR Audit: Reconcile invoices before approving
More screenshots
Works with the AI Agents you already use
SaaS MRR Audit: Reconcile invoices before approving
Your MRR number and your invoices disagree and you don't know why. Rebuild MRR from subscription events, flag invoice price mismatches and see the month-by-month bridge before a board or finance review.
$12
SaaS MRR Audit: Reconcile invoices before approving
Example session with this skill installed
Audit the supplied January–March subscription events against bad-invoices.csv. Show the MRR bridge, retention and affected records before I approve the export.
- Read your context and instructions
- Compiled the saas mrr audit
SaaS metrics audit: FAIL
Fail-on level: error. Findings: 2 error, 0 warning, 0 info.
Calendar 2026-01 to 2026-03; 2 customers, 6 movements. Bridge identity checked for every month (holds).
MRR bridge (last 12 months)
| Month | Start | New | Expansion | Reactivation | Contraction | Churn | End | Customers | Logo churn | Net MRR churn | Quick ratio |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026-01 | 0.00 | 200.00 | 0.00 | 0.00 | 0.00 | 0.00 | 200.00 | 0->2 | |||
| 2026-02 | 200.00 | 0.00 | 50.00 | 0.00 | 0.00 | -100.00 | 150.00 | 2->1 | 50.0% | 25.0% | 0.50 |
| 2026-03 | 150.00 | 0.00 | 0.00 | 100.00 | -30.00 | 0.00 | 220.00 | 1->2 | 0.0% | 20.0% | 3.33 |
Cohort retention (pooled over cohorts old enough)
| Months since start | Cohorts | Logo retention | Net revenue retention |
|---|---|---|---|
| 0 | 1 | 100.0% | 100.0% |
| 1 | 1 | 50.0% | 75.0% |
| 2 | 1 | 100.0% | 110.0% |
Billing cross-check
invoices: 5, customer_months_checked: 5
Findings
| Severity | Check | Count | Meaning | First example |
|---|---|---|---|---|
| error | invoice_list_price_mismatch | 1 | amount_due differs from unit_amount x seats. | {"line": 3} |
| error | invoice_mrr_mismatch | 1 | Invoice normalised to a month differs from MRR. | {"customer_id": "EXAMPLE-A", "month": "2026-02", "invoice": "EXAMPLE-I2", "invoice_monthly": "180.00", "mrr": "150.00"} |
FAIL — both errors concern invoice EXAMPLE-I2 at invoice CSV line 3. Confirm the February amount against billing before correcting the export. If the difference is valid proration or an adjustment, agree a treatment outside this model. No reported bridge was supplied; external bridge comparison was not performed.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem Someone hands you a monthly MRR bridge to sign off. The dashboard says February ended at 150.00, but one invoice works out to 180.00 a month, and you can't tell which number is wrong or how the total was built. Rebuilding it by hand in a spreadsheet takes an afternoon and is easy to get wrong.
See it before you buy: real input, real output Command (from the package folder, exit code 1, 0.027 s):
python3 -I -B scripts/mrr_audit.py --events references/good-events.csv --invoices references/bad-invoices.csv --format markdown
Before: Invoice export (excerpt of references/bad-invoices.csv)
invoice_id,customer_id,period_start,period_end,amount_due,billing_interval,unit_amount,seats,status
EXAMPLE-I2,EXAMPLE-A,2026-02-01,2026-03-01,180,month,150,1,paid
After: lines copied from the script's output (excerpt):
# SaaS metrics audit: FAIL
| 2026-02 | 200.00 | 0.00 | 50.00 | 0.00 | 0.00 | -100.00 | 150.00 | 2->1 | 50.0% | 25.0% | 0.50 |
| error | `invoice_list_price_mismatch` | 1 | amount_due differs from unit_amount x seats. | {"line": 3} |
| error | `invoice_mrr_mismatch` | 1 | Invoice normalised to a month differs from MRR. | {"customer_id": "EXAMPLE-A", "month": "2026-02", "invoice": "EXAMPLE-I2", "invoice_monthly": "180.00", "mrr": "150.00"} |
What the numbers show
- Synthetic data only: a generated set of 5,000 subscriptions, 10,000 events and 50,000 invoices; all 84 bridge numbers (12 months x 7 columns) matched the expected answer to the cent.
- 9 planted errors (3 duplicate events, 5 invoices off by one cent, 1 wrong ending balance): all 9 caught, shown as 10 findings.
- The small export (813 customers, 3,330 invoices) passed clean; it ran in about 0.1 s and the 50,000-invoice runs in about a second.
What you get A single Python script, small example CSVs, notes on the input format and the formulas, and four saved example reports. You point your agent at your exports. It checks they fit the expected columns, runs the audit, and tells you what failed, which rows to look at, and which checks it couldn't run because an input was missing.
- It rebuilds month-end MRR from subscription events and splits each month into new, expansion, reactivation, contraction and churn.
- It checks every active customer-month against an invoice, if you supply invoices.
- It checks a bridge someone else produced: does each month add up, does each month start where the last one ended, and does it match the recomputed numbers?
- You get logo and revenue retention by months since start, as a JSON or Markdown report.
- The exit code is 0 for pass, 1 for fail and 2 for bad input, so you can use it as a gate.
Measured on real data (full detail) To be upfront, none of this is real customer billing. I tested on two synthetic datasets. One is an export from my own SaaS warehouse simulator. The other is a generated set where I knew the right answer in advance, worked out by a separate script that doesn't share code with the auditor.
| Sample | Size | Result |
|---|---|---|
| Warehouse simulator export | 813 customers, 1,075 events, 3,330 invoices | Clean pass, no findings |
| Generated sample | 5,000 subscriptions, 10,000 events, 50,000 invoices | All 84 bridge numbers (12 months x 7 columns) match the expected answer to the cent |
| Same sample with 9 planted errors | 3 duplicate events, 5 invoices off by one cent, 1 wrong ending balance | All 9 caught. They produce 10 findings, because the wrong balance fails two checks |
The small export ran in about 0.1 s and the 50,000-invoice runs in about a second, Python startup included. The large datasets aren't in the package, but the small examples are, so you can re-run those yourself.
Known limitations
- Each event row has to carry the full new price (not a price change), with one subscription per customer and one currency. If your billing system gives you raw price IDs or webhooks, you'll need to join prices first.
- It doesn't do revenue recognition, tax, FX, discounts, proration or cash reconciliation.
- It looks at the last day of each month, so a customer who cancels and comes back within the same month shows no churn.
- A bridge can add up even when early history is missing. Supplying invoices and a separately produced bridge makes the check much stronger.
- Everything is held in memory. I measured 50,000 invoices, which isn't a size limit, but I haven't tested far beyond it.
Who it's for Finance analysts, SaaS operators and data people who review recurring revenue numbers before they go into a board deck or month-end close, or who need to chase down why billing and the dashboard disagree.
Requires Python 3.8 or newer and an agent that can read local files and run Python. The script uses only the standard library. You give it CSV files. It doesn't need a warehouse connection or dbt.
Runs locally with no network access or telemetry. Writes only when you redirect output or select --json-out.
Check it yourself
22 test cases ship with the package, each running the script on an example input. I ran those plus 28 extra edge cases (empty files, Unicode, Windows line endings, broken UTF-8, oversized cells, same-day events, numeric extremes and bad arguments), compared the four saved reports byte for byte, and re-ran all three large datasets. To check it yourself from the package folder, run python3 -I -B scripts/mrr_audit.py --events references/good-events.csv --invoices references/good-invoices.csv --bridge references/good-bridge.csv --format markdown and compare it with references/sample-output/pass.md. references/benchmarks.md has the failing example and the details of each measurement.
Price and guarantee $12 one time, yours to use with any agent that can run Python. Covered by Agensi's money-back guarantee if it doesn't do what this page promises.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Finance & Accounting.
Requires single-currency CSVs and one subscription per customer. Does not handle tax, FX, or accounting revenue recognition. History must start from the first event for accurate classification.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
Permissions
Uses local shell execution and file reads for the user-selected CSV inputs. Prints reports to stdout; writes only an explicitly selected --json-out file, which replaces an existing file at that path. No network access, no telemetry, and nothing is sent anywhere.