SettleFlow
AI-powered settlement automation for natural gas trading. Turning a month-end nightmare into a hands-off pipeline.
A spreadsheet problem at trillion-dollar scale
During my internship at Citadel, I worked on trade settlement reconciliation and was surprised by how manual the process was. Even at one of the most sophisticated trading firms in the world, I watched analysts spend entire days matching counterparty invoices to internal trades in spreadsheets, cell by cell, chasing down penny discrepancies across hundreds of rows.
I assumed Citadel was the exception. Surely smaller shops had figured out something more elegant. They hadn’t. When I started talking to people at mid-sized gas marketers, I found it was worse: legacy systems that treat settlements as an afterthought, teams that grow headcount instead of improving process, and a lot of late nights at month-end. The industry moves trillions in notional value but still closes the books like it’s 2005.
That gap stuck with me. So I started building.
The industry moves trillions in notional value but still closes the books like it’s 2005.The problem
Manual reconciliation doesn’t scale
Natural gas marketers reconcile hundreds of counterparty invoices against internal trades every month, manually. Analysts verify quantities, cross-check pricing against published indices, and chase down discrepancies buried in emails, PDFs, and disconnected systems. It’s slow, error-prone, and doesn’t scale.
The target customer is the Director of Operations or Controller at a natural gas marketer with 50 to 500 trades per month. The person who owns back-office efficiency and feels the month-end crunch personally.
End-to-end settlement automation
SettleFlow automates the full settlement workflow. Upload a counterparty invoice, and the system extracts structured data, matches it against your house trades, detects breaks, explains root causes, and generates settlement memos. Clean matches get auto-approved. Edge cases get escalated for human review.
Upload counterparty invoices as PDF, EDI, or email. AI extracts structured line items, dates, and pricing.
Multi-strategy reconciliation maps invoice lines to house trades by reference, date, quantity, and delivery point.
AI flags breaks and classifies root causes: price variance, timing difference, quantity mismatch, or data entry error.
Clean matches auto-approve. Breaks surface for review with side-by-side comparisons and settlement options.
AI that explains what went wrong and why
When a discrepancy is found, SettleFlow doesn’t just flag it. It classifies the break type, shows a side-by-side comparison of your records versus the counterparty invoice, highlights the variance, and generates a plain-English root cause analysis. Analysts see exactly what’s off and can approve, reject, or escalate with one click.
Price Variance: index pricing difference flagged at $4,750
Timing Difference: 10-day flow date mismatch detected
Quantity Variance: AI identifies data entry pattern as root causeFull audit trail and reporting
The Break Log tracks every discrepancy with severity levels, variance amounts, and resolution time. Reports show approval tier distribution, processing summaries, and counterparty performance. Everything is exportable for compliance and auditing.
Break Log with severity classification and variance totals
Reports with STP rate, approval tiers, and reconciliation exportsFrom break resolution to debit memo in one click
Once breaks are resolved, SettleFlow generates settlement memos and debit/credit documents automatically. Each memo includes itemized adjustments with trade references, corrected amounts, and payment instructions. Documents are generated as PDF, ready for counterparty distribution.
Settlement memo generation from reconciled breaks
Auto-generated debit memo with line-item adjustmentsSlack notifications in real time
SettleFlow pushes processing updates and break alerts directly to Slack. When an invoice is uploaded, your team sees extraction progress, match results, and break summaries without leaving their workflow. Every notification links back to the break detail view for one-click resolution.
Slack channel receiving automated break alerts with invoice details and variance summariesTech stack
Document extraction, intelligent matching, and root cause analysis. Handles messy PDFs and ambiguous line items.
Multi-agent orchestration pipeline. Each stage (ingest, match, detect, resolve) runs as an independent agent with shared state.
PostgreSQL for pipeline state, trade data, and full audit trails. Row-level security for multi-tenant isolation.
Currently in customer discovery. Running conversations with ops leaders at gas marketers to validate the pain and refine the product before building out a pilot.