
Planner Agent
Delhivery is one of the largest tech-driven logistics companies, backed by FedEx and SoftBank. Know More →
Agentic Workforce/Penny
Invoice & AP Agent
Penny is the Invoice & AP Agent: she checks every invoice against contract to stop overpayments.
Penny
Audits every invoice against contract
Finds the overcharges your audit passed, files the dispute, and wins the cash back.
Skills
Coverage by mode
A truckload invoice, an LTL bill, a container demurrage claim and an LCL revenue-ton charge fail in completely different ways. Penny runs mode-specific logic for each rather than matching totals and calling it an audit.
Rate, plan, execution and paperwork should tell the same story. Penny reads all four and surfaces the charge where they stop agreeing.
45
canonical charge codes
25+
accessorial types validated
4
independent mileage sources
Every charge is tested against four independent sources: the contract rate, the load plan, what the trip actually did, and the documents behind it. That separation is what tells a genuine execution variance, a reroute or an added stop, apart from a rate that was simply billed wrong.
Detention, TONU, layover, liftgate, residential delivery and 20 more, each tested against GPS dwell time, ELD logs, geofence crossings and the free time your contract actually grants. Deterministic checks settle everything they can before a model is asked anything.
Invoice-claimed miles against ELD odometer, raw GPS trace and a routing-engine benchmark. Four numbers that should agree, and when they do not, a variance explained against the execution timeline instead of flagged and left for you.
Fuel surcharge recomputed from the EIA weekly diesel index against the matrix band in force on the shipment date, not the date the invoice happened to arrive. Percent of linehaul, per mile, per hundredweight or flat, whichever the contract specifies.
The dispute is drafted with the evidence attached, sent in a tone matched to that carrier, then tracked. Replies are classified as accept, partial or counter, with follow-up on an SLA clock and a negotiation floor you set.
LTL breaks on class, density and re-measurement, none of which a total-to-total comparison can see. Penny rates the shipment the way the tariff actually works, then asks whether the carrier earned each adjustment.
18
NMFC classes, 50 through 500
4
re-measure failure modes caught
2
specialist reasoning agents
Linehaul rated the way an LTL tariff really works: NMFC class, FAK remapping, class-specific or blanket discount, weight breaks and minimum charge. Lane-rate overrides take precedence where your contract carries one.
Density floors and dimensional weight decide when light, bulky freight pays for the trailer space it occupies. Where both apply, the greater chargeable weight governs. Deficit weight pulls the other way and catches the carrier who failed to bill you the cheaper higher weight break, an overcharge visible only if someone computes it.
A carrier who re-measures after pickup has to show the certificate. A reweigh line with no weight certificate is an automatic dispute, and so is a re-measure that did not move the rating basis. A re-measure that changed nothing is not a billable event.
When billed class differs from the BOL, the evidence packet is built first: commodity, weight, dimensions, cube, computed density and the NMFC named-item lookup. Only then does a reasoning model work the four NMFC characteristics and draft the dispute. Density is the argument that actually wins a reclass.
Detention double-billed across a reconsignment, stop-offs telemetry says never happened, overlapping detention windows. These only appear when the charges are read together against one merged timeline, never one line at a time.
A full container is billed by up to seven parties, and the unit that accrues a charge is a container-day. The question is never just what the invoice says, it is whether the box was takeable and returnable on each day billed.
4
actionability tests per claim
5
chassis reconciliation rules
10
index-linked surcharge codes
Unit basis, then obligation, then identity, and only then amount. Most tools start at the amount. A charge can be perfectly rated and still not be yours to pay under the liner terms, and that is where the large findings sit.
First, how many days were chargeable under the allowance that actually applies. Then, of those days, which ones you could have acted on. Four tests decide it: was the container available, was it free of holds, was an appointment obtainable, could the empty be returned.
Demurrage accrues against a container, per day, so the ledger is keyed that way rather than by invoice. An invoice-level model cannot express the finding that matters most: three of these five billed days are claimable and two are not.
A chassis bill that adds up perfectly is still wrong if it charges for days after the container went back. Billed days are bounded by container possession, and a day you could not act on was also a day you did not control the chassis, so it inherits that day's verdict.
BAF, LSS and eight other codes are recomputed from the index published at or before the strike date, never rate-matched. Where the index-date convention is unstated, it escalates rather than guessing across a fortnight of index movement.
Credits, partial settlements, chargebacks and FX effects are recorded against the entries they actually touched. A partial settlement leaves the residual disputed on the original basis rather than quietly written off.
In FCL the audit unit is a container with a number on it. In LCL it is calculated from measurements, so the unit has to be recomputed before any rate can be evaluated. Most of the money is decided before the tariff is ever consulted.
7
independent revenue-ton checks
3
co-load allocation paths
0
claims filed on inference
Volume against weight, whichever governs, recalculated from the pieces before a rate is looked at. Seven independent checks fall out of that one computation, and none of them need execution data, so they land on day one of an onboarding.
Rounding per piece and rounding on the total give materially different answers. Billing the per-piece figure under a tariff that never specified it is a quiet overcharge, and it recurs on every shipment moving under that tariff.
Where a tariff does not state a field, that silence is recorded as its own finding rather than filled with a sensible default. The default an auditor would reach for is usually the exact point in dispute.
Three gates stand before a draft: a dollar floor, a measurement tolerance, and evidence before inference. A remeasurement with no stated method does not outrank your packing list, but the first move is to request the record, not to dispute it.
A container charge accrues against a box carrying several house bills, and yours is a share of it. With the manifest, the share is computed and any excess is a claim. Without it, the implied share ceiling still catches the worst case, a small consignment billed the full container's demurrage.
A $29 difference is not worth a per-shipment claim, and filing it costs more than it recovers. It goes to root cause instead, because the value was never the $29, it was the tariff defect repeating underneath it.