The agent reads your P&L the way a skeptical CFO would
Line by line, category by category, every billing cycle - catching the patterns that manual review misses.
How the scan engine reads your general ledger
Rivvun ingests a general ledger export or reads directly via accounting API. The agent normalizes cost category names across naming conventions, groups line items by vendor cluster, and builds a trailing 3-month rolling baseline per category.
- Reads: transaction date, vendor name, cost category, amount, invoice reference
- Normalizes category naming across chart-of-accounts variations
- Flags statistical outliers against per-category rolling baseline
- Does not store raw P&L data beyond the current flagging window
| Category | Baseline Avg | This Period | Delta |
|---|---|---|---|
| COGS - Software | $41,200 | $48,900 | +18.7% |
| COGS - Infrastructure | $28,400 | $29,100 | +2.5% |
| S&M - Agency | $15,800 | $15,200 | -3.8% |
| G&A - Legal | $8,100 | $14,600 | +80.2% |
| Vendor Name | Amount | Date | Confidence |
|---|---|---|---|
| Acme Software Inc | $4,200.00 | Jun 3, 2026 | High |
| ACME SOFTWARE INC. | $4,200.00 | Jun 4, 2026 | High |
| Acme Softwr Inc | $4,198.50 | Jun 5, 2026 | Med |
| Zenith Analytics | $1,850.00 | Jun 8, 2026 | High |
Fuzzy clustering catches vendor name variations
Duplicate vendor charges rarely arrive as identical line items. The agent uses fuzzy string matching to cluster name variations - "ACME SOFTWARE INC.", "Acme Software Inc", "Acme Softwr Inc" - into a single vendor group before comparing invoice amounts and dates.
- Groups vendor name variations using edit-distance clustering
- Identifies identical or near-identical amounts within a rolling window
- Scores each match with a confidence level before flagging
- Legitimate recurring charges are baselined and excluded from flags
Silent rate increases surface before they compound
The agent reads per-unit rates from invoice line items and tracks them vendor-by-vendor across billing periods. When a rate changes without a corresponding contract amendment event in the ledger, Rivvun surfaces a pricing drift signal.
Example: a vendor bills $12.40 per seat in Q1 and $13.80 per seat in Q2 with no amendment. That 11.3% increase is below the noise threshold for manual review but compounds to meaningful margin erosion at 40-seat scale across four quarters.
- Tracks per-unit rate per vendor per billing period
- Detects rate changes above a configurable threshold (default 3%)
- Flags include the vendor, old rate, new rate, and period-over-period delta
| Vendor | Q1 Rate | Q2 Rate | Change |
|---|---|---|---|
| Vendor A | $12.40/seat | $13.80/seat | +11.3% |
| Vendor B | $0.008/API | $0.009/API | +12.5% |
| Vendor C | $420/mo | $441/mo | +5.0% |
Connects to the accounting systems finance teams already use
Read-only OAuth connection where supported. No write permissions requested. Your data stays in your accounting system.
How Rivvun handles your financial data
Read-only access scope declared at connection. P&L data processed in-scan and not retained beyond the flagging event.
Read-only OAuth scope
Connection requests read-only permission at the OAuth consent screen. Rivvun cannot modify, create, or delete records in your accounting system.
No raw data retention
P&L data is processed during the scan window and not stored beyond what is required to generate and deliver the flag. Aggregated baselines are retained; raw transaction records are not.
Encryption in transit
All data in transit between your accounting system and Rivvun's processing layer uses TLS 1.3. No plaintext transmission at any point in the data pipeline.
India data residency
Processing infrastructure is operated within India for design-partner accounts. Enterprise deployment options include on-premise configuration for organizations with specific data residency requirements.
REST API and webhook delivery for design partners
Design-partner accounts get access to the flag feed via REST endpoints and can subscribe to webhook delivery for real-time events. Build your own alerting workflow or pipe flags into your existing finance tooling.
- REST endpoints to query flag feed by date range, category, or signal type
- Webhook delivery with configurable endpoint URL and signature verification
- Event payload includes category, signal type, variance percentage, and flag reason
- Available to design-partner accounts during the early access program
{
"event_type": "spend-flag",
"category": "vendor-duplicate",
"vendor_cluster": "Acme Software Inc",
"variance_pct": 18.7,
"flag_reason": "Duplicate invoice detected
within 72h window, high
confidence match",
"invoice_count": 3,
"total_amount_usd": 12600.00,
"detected_at": "2026-07-08T09:14:22Z"
}