Nexa is Shipsy’s autonomous settlement agent. She reconciles every freight invoice line against the contracted rate card, the executed shipment legs, and the accessorial rules — at scale, in seconds per invoice, with 90%+ auto-match rates. Lines that don’t match are routed to a dispute queue with Nexa’s full reasoning trace attached, so the finance team works only the exceptions.
Why we built this
Carrier settlements are a margin sink in every logistics enterprise we’ve audited. The root problem is not the rate card — it’s the reconciliation. A single invoice from a national carrier can carry 20,000 lines. Each line has a declared rate, a declared distance or weight, an accessorial code, and a fuel surcharge. Each line must be matched against the shipment record, validated against the rate card, and priced for accessorial correctness. A typical enterprise doing this manually reconciles 30–40% of lines, accepts the rest on trust, and loses 2–5% of freight spend to overbilling.
Nexa was built to flip that ratio. Instead of reconciling 30% of invoice lines, enterprises reconcile 100% — because the agent can.
How it works
Nexa runs as a four-stage pipeline per invoice:
Stage 1 — Invoice ingestion. Nexa ingests PDFs, EDI 210s, CSVs, and carrier-portal exports. She uses an LLM-assisted extraction layer to normalize every line into a canonical schema (origin, destination, service level, weight, volume, accessorials, claimed rate, claimed total). The extraction confidence score per field is retained — low-confidence extractions are auto-flagged for review.
Stage 2 — Rate-card matching. Each line is matched against the digital rate card stored in Shipsy’s freight procurement module. The match uses a rule tree: lane (origin-destination pair), service level, weight/volume bucket, effective date window, contract addendum, and accessorial eligibility. The rate card itself is structured — not a PDF — so the matching is deterministic.
Stage 3 — Execution validation. Nexa cross-checks each line against the actual shipment execution record in Shipsy. She confirms the shipment was executed, the claimed weight matches the captured weight, the lane matches the executed origin and destination, and the accessorial (e.g., residential delivery, liftgate, inside delivery) was actually performed per the driver-app or ePOD evidence. A line claimed as “liftgate delivery” with no ePOD evidence of a liftgate attempt is flagged.
Stage 4 — Decision and routing. For each line, Nexa produces one of three outcomes: auto-approved (rate matches, execution matches, pay as invoiced), auto-disputed (rate exceeds contract, accessorial unjustified, duplicate line, or lane doesn’t exist in the contract — Nexa drafts the dispute with the specific reasoning), or review (low extraction confidence, edge case, contract ambiguity — routed to a human finance lead with the full trace).
The full pipeline closes an invoice in minutes, not weeks.
Here’s the flow at a glance:
Early results
Enterprises deploying Nexa typically report, within 90 days:
- 90%+ of invoice lines auto-matched against rate cards, with deterministic reasoning trails.
- 2–4% recovery of freight spend via disputed overbilling — compounding across a year, this is 8-figure savings for large shippers.
- Days-to-close-invoice drops from 30–60 to under 5.
- Finance team effort on settlements drops 70%+, refocused on vendor negotiations and strategic procurement.
A global alco-bev leader operating across 70+ countries uses Nexa (alongside sister agent Vera) to autonomously handle settlements across millions of distribution trips a year.
What’s next
Three upgrades in flight: multi-tier rate card support for 3PLs managing tenant-specific contracts, natural-language contract parsing so Nexa can ingest a new PDF master agreement and extract a digital rate card automatically, and predictive billing anomaly detection — Nexa flagging rate drift or unusual accessorial patterns before the invoice closes.