AI Carrier Selection & Smart Shipping for ERPNext

We build a proper Frappe app (its own smart_shipping module and app, versioned in a git repo) that turns carrier selection into a scored, auditable decision inside ERPNext. New DocTypes — Carrier Profile, Lane Performance, Shipping Recommendation, and RTO Risk Score — hold your carrier accounts, per-lane reliability history, and the model output. A hooks.py doc event on Sales Order and Delivery Note (validate / on_submit) calls a whitelisted server-side method that assembles the order's features (destination pincode/zone, chargeable weight and dimensions, declared value, payment mode, address-quality signals) and returns a ranked list of carrier + service options, each with an estimated landed cost, expected transit window against the order's SLA, a lane-reliability score, and an RTO-risk probability. The recommendation is written back to the order and surfaced in the Delivery Note UI via a client script, so your dispatch team sees "recommended: Carrier B — Express, ₹X, ~2 days, 4% RTO risk" with the runner-up options one click away — and can override with a captured reason.

The intelligence layer starts as an explainable model (gradient-boosted / logistic scoring on your historical delivery outcomes) rather than a black box, so every recommendation shows the factors behind it. Auto-allocation rules let you set policy — for example, force express when the customer SLA is under 48 hours, cap on cost otherwise, or block COD to addresses above an RTO-risk threshold — while always allowing a manual override. A Frappe scheduler event (scheduler_events hourly/daily) continuously retrains and recalibrates the model from newly closed deliveries, updating each carrier's lane-performance and on-time rates so the system gets sharper as your real outcomes accumulate. Everything is exposed through the Frappe REST API and whitelisted methods, so your storefront, OMS, or a WMS can request a recommendation before an order is even confirmed, and role profiles / permission rules keep model configuration and override rights scoped to the right teams.

Because this is build-to-order, nothing is a generic download. After a short scoping call we confirm your exact carriers, rate cards, lanes, SLA definitions, and the outcome data available for training, then build against your ERPNext v15 or v16 instance. You get the installable app source for your version, deployment on a staging site for UAT with a rollback plan, technical and user documentation, a training session for your dispatch team, and a post-go-live support window. Typical delivery is 2-4 weeks from confirmed scope, depending on how many carriers and how much history we integrate.