MONTREAL.AI / SKILLOS
Enterprise Ops Market Proof
100% autonomous GitHub Actions proof for procurement invoice reconciliation and payment-risk triage.
Current status
PASSED_AUTONOMOUS_ENTERPRISE_OPS_MARKET_PROOF
No human review. No emails. No customers. No private data. No API keys. Deterministic holdout benchmark.
+75.0 ptsdecision accuracy gain
100.0%critical-risk recall
0.0%false approval rate
$5,977,753.86synthetic risk reduced
Before / after on holdout cases
| Metric | Baseline | SkillOS |
|---|---|---|
| Decision accuracy | 25.0% | 100.0% |
| Critical-risk recall | 19.5% | 100.0% |
| False approval rate | 66.1% | 0.0% |
| Minutes per case | 9.5 | 2.2 |
| Cost per case | $11.88 | $2.75 |
| Synthetic dollars at risk left unblocked | $5,977,753.86 | $0 |
Learned SkillOS rules
- Require a three-way match: purchase order, invoice, and receipt.
- Block duplicate invoice IDs for the same vendor or repeated invoice patterns.
- Escalate vendor identity mismatch or bank-account changes.
- Hold invoices with amount, tax, currency, terms, or delivery mismatches.
- Escalate missing receipts before payment approval.
- Approve clean invoices and preserve early-payment discount opportunities.
- Never approve a payable when a critical risk signal is present.
Proof gates
- ✅ no human review required
- ✅ not an email workflow
- ✅ no emails sent
- ✅ no customers contacted
- ✅ no private data used
- ✅ no api keys required
- ✅ deterministic reproducible benchmark
- ✅ enterprise ops workflow
- ✅ train cases at least 100
- ✅ holdout cases at least 300
- ✅ learned rules created
- ✅ decision accuracy gain at least 25 points
- ✅ critical risk recall at least 99 percent
- ✅ false approval rate zero
- ✅ review time reduction at least 70 percent
- ✅ cost reduction at least 70 percent
- ✅ synthetic dollars at risk reduced positive