The problem it solves
Every shipper above a million parcels a year ends up with the same stack: a contract binder of 20+ carriers, an ops lead who knows which carrier is quietly under-performing this week, a finance team that finds out six weeks later, and a rate-card spreadsheet that nobody has touched since the last RFQ. When an order drops into the system, the allocation logic is usually a combination of rules written years ago, one overworked dispatcher, and a fallback to the cheapest-on-paper option. That fallback is not the cheapest in practice — it is the cheapest until the carrier misses its SLA, triggers a chargeback, and burns through the cost advantage three times over.
The result is a well-known leak: 18-25% of every shipper’s carrier spend is misallocated. Not because the contracts are bad — because the allocation is stuck in rules, not in intelligence. Carrier performance changes weekly. Capacity changes hourly. Allocation logic changes once a year. And the ops lead holding the whole thing together in her head is the single biggest KVP (key-value-person) risk in the freight organisation. When she is on leave, allocation quality visibly drops; when she leaves the company, the operation spends six months rebuilding the same mental model from scratch.
What it is
Multi-Carrier Intelligence is Shipsy’s real-time allocation engine that scores every shipment against every eligible carrier and picks the assignment that balances SLA, cost, capacity, and performance — not the one that matches a static rule. It runs under Astra as the carrier-selection layer and under the multi-carrier orchestration product as the execution layer.
What makes it different: the scoring function is not a weighted checklist. It is a learned model that updates with every delivery outcome — every on-time, every breach, every refusal, every chargeback — and re-prices each carrier on every route, every weight band, every service type, every hour of the day. Allocation that used to run once per order now runs live, on the current state of the network. It is new-age in a specific, measurable sense: it does for carrier allocation what dynamic ad-bidding did for media buying — real-time scoring, real-time execution, continuous learning.
Under the hood it combines a real-time carrier-capacity registry, a performance scorecard that updates per shipment, a contract-enforcement layer, and 240+ pre-built carrier integrations. Failover happens in single-digit seconds when a carrier API goes dark. The engine is vertical-ready across quick-commerce, D2C e-commerce, B2B express, pharma, and cross-border freight — each with a tuned weight profile out of the box.
Core capabilities
| Capability | What it does |
|---|---|
| Composite scoring function | Every allocation candidate is scored on four dimensions — SLA match, landed cost, available capacity, observed performance — weighted per customer, per lane, per service type. Weights are configurable and auditable. |
| Live performance scorecard | Every completed shipment updates the carrier’s rolling scorecard. Breaches are weighted by recency and by the shipper’s own priority. A carrier that was #1 on Monday can be #4 by Thursday if its performance slips. |
| 240+ carrier integrations | Pre-built integrations with global (DHL, FedEx, UPS, Aramex), regional (Delhivery, J&T, Pos Malaysia), quick-commerce (Borzo, Dunzo, Porter), and niche LSPs. New carriers onboard in 2-4 weeks with a standard adapter. |
| Contract enforcement | Rate cards, fuel surcharges, zone skips, accessorial rules, and minimum-volume commitments live in a single contract registry. Every allocation honours the contract; every deviation is flagged to Nexa for audit. |
| Real-time failover | A carrier API timeout or a 4xx response triggers automatic re-allocation within 3 seconds to the next-best carrier. The customer never sees the failure; the dispatcher sees a single audit line. |
| Volume-commitment management | Tracks contracted monthly volumes per carrier. Allocates in a way that fulfils commitments without over-serving — protects the shipper from rebates they will not earn. |
| Hybrid own-fleet + 3PL logic | Scores own-fleet capacity against 3PL capacity on the same axis. If the in-house fleet is 70%+ utilised and the marginal 3PL cost is lower than the marginal own-cost, the engine routes to 3PL — and vice versa. |
| Pincode-level eligibility | Every carrier’s service map is maintained at the pincode / zip level with service-type granularity (COD, reverse, fragile, reefer). A carrier that lost coverage in 12 pincodes last week is excluded from those pincodes this week. |
| Dynamic pricing awareness | For spot-pricing carriers and quick-commerce orchestration, the engine pulls live rates at allocation time rather than relying on stale rate cards. |
| Carrier dispute feed | Delivery exceptions, chargebacks, and POD discrepancies flow back into the scorecard and into Vera for settlement reasoning. |
| Simulation and rate-shopping | Shippers can simulate allocation outcomes across historical shipments under new contract terms — “what would last quarter have cost if we gave Carrier X 30% more volume?” |
| Configurable business rules on top | Customer-specific rules (Amazon orders must go to Carrier A; Dubai deliveries must use a 4PM cutoff) layer on top of the learned scoring. Rules are first-class, not a workaround. |
How it works
The engine separates the allocation decision (live, millisecond-scale) from the scoring model (updated in near-real-time from the performance feedback loop) from the contract and eligibility layer (authoritative, change-controlled). An allocation request hits the decision layer, which asks the scoring model for candidate ranks, filters against the contract layer, and commits the assignment to the carrier API. The loop closes when the shipment status — on-time, breached, refused — flows back into the scoring model.
This separation is what lets the engine run change safely at enterprise scale. The contract layer is ground truth and is edited under change control. The scoring model is continuously learning but bounded by versioning. The decision layer is deterministic once the model version is pinned. A shipper can audit any past allocation by replaying the exact inputs and getting the exact output — which is the compliance-team table stakes no rule-based allocator has ever met.
On the execution side, every allocation runs through a fast path that typically settles in under 400ms end-to-end, with a resilient failover path that handles carrier API outages without exposing them to the shipper. The sequence below captures a typical fast-path allocation and the rare failover case in one flow.
Two aspects of the architecture matter operationally. First, the decision layer is stateless per request — which means a node can fail mid-allocation and another node picks it up without a retry penalty. Second, the scoring model is versioned, so a shipper who wants to roll back a scoring update can pin allocation to the previous version for any lane or customer segment without a full rollback. Combined, this lets the engine run with 99.99% availability across peak-hour allocation volumes that sometimes exceed 500 orders per second.
Proven outcomes
| Customer type & scale | Outcome |
|---|---|
| One of Asia’s largest quick-commerce arms, 5M+ orders/month, 200+ dark stores | ~21% reduction in cost-per-delivery via hybrid own-fleet + 3PL orchestration; sub-20-minute promise held as carrier capacity fluctuated hourly |
| Global parcel leader, 65+ countries, 18,000+ drivers | 18% reduction in cross-border allocation cost; 2-4 week carrier onboarding using the standard adapter framework |
| MENA retail conglomerate, multi-brand sports, health, lifestyle | 14% reduction in last-mile allocation cost across 80+ markets; unified scorecard replaced six fragmented carrier reviews |
| Leading ANZ parcel operator, AUD 200-250M annual revenue | 35% reduction in failed deliveries via carrier-aware allocation + performance-weighted scoring |
Integrations
- Carriers: 240+ pre-built — DHL, FedEx, UPS, Aramex, Delhivery, BlueDart, J&T, Pos Malaysia, Yamato, Australia Post, Posti, Correos, DPD, GLS, DPDgroup, IndiGo cargo API, regional LSPs across MENA, SEA, LATAM, EU
- Quick-commerce 3PLs: Borzo, Dunzo, Porter, Loggi, Rappi Cargo, Pandago
- ERP and order sources: SAP, Oracle, Salesforce Commerce Cloud, Shopify, Magento, WooCommerce, custom OMS via REST
- WMS: Manhattan, Blue Yonder, Oracle WMS, Körber, native Shipsy WMS
- Financial reconciliation: Handoff to Nexa for invoice reconciliation and rate enforcement; Vera for dispute handling
- Data platforms: Snowflake, Databricks, BigQuery for historical allocation analytics and contract simulation
Deployment
Phase 1 — Discovery (weeks 1-2). Carrier inventory, contract audit, rate-card ingestion, baseline allocation analysis against the last 90 days of shipments. Identify top leakage patterns and define the target scorecard weights with the shipper’s ops and finance leads.
Phase 2 — Configuration (weeks 3-4). Contract registry loaded, eligibility map validated per lane and service type, scoring weights tuned, carrier APIs connected using pre-built adapters (typically 60-80% of the carriers are pre-integrated).
Phase 3 — Pilot (weeks 4-6). Shadow mode — live allocation decisions compared against the current production allocator, with no change in carrier assignment. Shipper reviews side-by-side on 100% of real orders for 2-3 weeks. Exit criteria: 10%+ projected savings and ops-team sign-off on explainability.
Phase 4 — Scale (weeks 6-10). Cut-over by lane or by customer segment. Governance: a weekly carrier-performance review in the first 60 days, monthly thereafter. Contract-enforcement deltas flow to finance. Most enterprises go live in 8-12 weeks; savings typically visible within the first 30 days of cut-over.
Security & compliance
- SOC 2 Type II, ISO 27001, GDPR-compliant data handling
- Contract registry is change-controlled — every rate-card edit has an owner, a timestamp, and a diff
- Every allocation decision is auditable: which carriers were considered, what each scored, why the winner won, which rule (if any) overrode the model
- Three-tier confidence scoring on scoring outputs — low-confidence allocations (new carriers, new lanes) trigger human-in-the-loop approval for the first N shipments
- PII is not used in the scoring model; only anonymised shipment features
- 21 CFR Part 11 and GDP / GMP-aware for pharma-regulated lanes
Case study callouts
One of Asia’s largest quick-commerce arms · 5M+ orders/month
Shifted allocation from a static “own-fleet first” rule to hybrid dynamic orchestration across own-fleet and 3PLs. Cost-per-delivery dropped ~21% while the sub-20-minute customer promise held through peak hours. The scoring model re-prices every 3PL every 15 minutes against live capacity.
Global parcel leader · 65+ countries · 18,000+ drivers
Replaced region-specific allocation systems with a single engine spanning 65+ country operations. 18% reduction in cross-border allocation cost, and the 2-4 week carrier-onboarding adapter cut new-market launches by an average of six weeks.
Leading ANZ parcel operator · AUD 200-250M revenue
Performance-weighted allocation — paired with driver-level performance data — dropped failed deliveries by 35% and allowed the operator to renegotiate three carrier contracts on verifiable performance data rather than gut feel.