Now accepting projects — Q3 2026

Transportation & Logistics: Custom AI Solution

Transport AI’s real sweet spot is collapsing tracking-event noise into customer-facing clarity — ETA prediction, proactive anomaly alerts, multilingual service triage, complaint document extraction. Design principle: AI proposes, humans decide changes. Rerouting, cancellation, and compensation are never automated — once automated they become irreversible commercial risk.

// AI use cases to start with

ETA prediction + proactive notification

Real-time tracking events + historical timeseries + route complexity feed ML models, accuracy 30–50% better than static ETA; anomalies surface 2–6 hours earlier and pass through a human-review queue before reaching customers.

Service case summary + draft replies

Tracking history + contract excerpts + prior complaints feed RAG; agents get full context + draft in the first second, multilingual / multi-timezone cases included.

Cross-system status fan-out + complaint extraction

One query against TMS + WMS + port + ticketing + billing + 3rd-party tracking; complaint documents (claims, damage photos) AI-extracted for staff confirmation.

// How EKel would deliver it

01Define hard constraints first: enumerate non-negotiable red lines (no auto-reroute, no auto-cancel, no auto-compensate), align dispatch, service, and contract management.
02Choose deployment: negotiation floors / special contract terms / recipient PII go through private deployment; public SOPs and tracking specifications may use commercial LLMs. Models prefer Llama / Mistral / Gemma open-source families; structured tasks like ETA combine classical ML with an LLM explanation layer.
03Build a reference dataset from real scenarios — 100+ cases (strikes, weather, port congestion, customs delay, damage, returns, hand-over failure), specifically test false-positive notification rate (over-sensitivity is costlier than late notification).
04Post-launch: full audit log + monthly human-sampled review + regression tests on every model upgrade. If the client already runs Salesforce, our stronger recommendation is Salesforce + Agentforce.

// Best fit

→Transport / logistics operators with high service volume, multilingual customers, complex tracking integration, wanting AI to free human time for high-judgement work.
→Operators with clear compliance boundaries and data classification (especially recipient PII, contract-sensitive data, cross-border transfer).
→Programs that want to prove ROI on one vertical use case (e.g., ETA prediction + proactive notification) before rebuilding the whole service platform.

// Sources

Salesforce Travel, Transportation, and Hospitality Salesforce Trust & Compliance

Back to industries See transport Salesforce →

// Custom AI architecture

Transport AI is a four-layer event-aware / human-signed stack.

// LAYER L4

User layer

Customers (shippers / consignees / consignors), service agents, operations schedulers, contract managers, partners — interacting via web, mobile, portal, or agent desk. Any AI surface affecting operations (reroute, change ETA commitment, compensation rulings) requires dispatcher / ops sign-off.

Customer / ServiceOps / SchedulingPartner Ops

// LAYER L3

Application layer

Built with **Vibe Coding** — ETA prediction + proactive notification, service-case summaries, complaint document structuring, partner-portal smart assistants. **Agentic workflow** fits cross-system status checks, exception investigation, notification coordination; anything affecting contract commitment or compensation requires human checkpoint.

Vibe CodingAgentic WorkflowOps-signed

// LAYER L2

AI layer

LLM Gateway + policy / contract / FAQ / routing-rules RAG + predictive models (ETA, exception risk) + eval pipeline + guardrails. Common model mix: Llama / Mistral / Gemma open-source models alongside commercial LLMs. Customer contract floors and tariffs never reach commercial LLMs.

LLM GatewayLlama / Mistral / GemmaEval · Guardrails

// LAYER L1

Data layer

Vector DB (policies, contracts, FAQs, SOPs, destination manuals embedded) + structured data (CRM customers, contracts, cases, tracking-event history) + full audit log. Contract floors, tariffs, ops-sensitive data stay in private deployment; PII cross-border transfer is strictly tiered.

Vector DBCRM + TrackingAudit log

// Pre-launch · 6 operational disciplines

Six things to resolve before any transport AI launch.

Commercial-sensitive data classification

Contract rate floors, margins, key-account structures — which may go to commercial LLMs vs only private deployment? Without clarity, do not start; this red line costs more than PII.

AI does not autonomously change ops commitments

AI can suggest ETAs, suggest reroutes, suggest compensation tiers — but the “committed ETA,” the “actual reroute decision,” and the “final compensation amount” are always dispatcher / ops / service-lead signed. AI suggestions must be tagged “recommendation” and routed through human approval.

ETA prediction confidence labelling

Every AI-predicted ETA carries a confidence interval. Low-confidence scenarios surface only a range externally (“3–5 days”), not a point estimate (“Tuesday 14:30”); only high-confidence cases get point estimates. Wrong ETA commitments directly damage customer trust.

Proactive notification frequency & content discipline

Proactive notification is good, but over-notification becomes spam. Rules: (1) every notification must be actionable or explanatory — never just a status update; (2) one shipment / parcel / container gets at most three notifications per 24 hours (unless truly urgent); (3) opt-out takes effect immediately.

Partner & customer data boundaries

If the partner portal has an AI assistant, design strictly so each logistics partner / agent accesses only their own shipments and customers. AI prompts must carry and verify partner_id — leaking another partner’s / customer’s data is contract breach + regulatory issue.

Eval & transport context dataset

Before launch, run 100+ real scenarios (multilingual service, cross-leg handovers, exceptions, contract deviation, extreme-weather / strike / port-congestion disruptions). Re-run on every model version. No pass, no ship.

// FAQ

Five questions that come up most in transport AI discussions.

01Which transport AI use case has the best ROI?

Three consistently high-ROI use cases: (1) **ETA prediction + proactive notification** — tracking events + historical routes + seasonality / port-congestion / weather signals feed AI to produce confidence-tagged ETAs and “proactive notify” triggers, sharply reducing inbound service queue; (2) **multilingual service triage + draft replies** — cross-language / timezone inquiries and complaints route to the right queue in the first second + pre-generated drafts for review; (3) **complaint document structuring** — cargo damage / insurance claim photos / documents / reports get AI-extracted into structured fields for staff confirmation. All three are augmentation; ops decisions (reroute, change ETA commitment, compensation) are always human-signed.

02Why not use off-the-shelf logistics AI products?

If a vertical logistics SaaS fits, buy it — engineer’s judgement. Custom is right when: (1) your operating network is unusual (multi-business mix: postal + international freight + warehousing; or non-standard route / line structure), and SaaS schema does not match; (2) deep integration with proprietary TMS / WMS / port systems is required (not just webhooks but event-driven real-time sync); (3) commercial-sensitive data (contract floors, margins, key-account structures) cannot live on a commercial multi-tenant platform. If the client already runs Salesforce, our stronger recommendation is **Salesforce + Agentforce + Service Cloud + Field Service** — service / sales / dispatch AI in one platform.

03Tracking event volume is huge — how does AI handle it?

Rule one: do not run an LLM on every event. Two-layer architecture: (1) **rules / statistics layer** — simple rules or statistical models filter first; 95%+ of events are “normal progress,” archived directly without bothering AI; (2) **LLM layer** — only “anomalous / needs proactive communication / needs explanation” events go to the LLM for summaries and notification copy. LLM usage, cost, and latency stay controllable. Postal + maritime event/sec typically runs thousands to tens of thousands — this architecture is necessary, not nice-to-have.

04Where does agentic workflow fit in transport?

Good fits: (1) cross-system status checks (TMS shipments, WMS inventory, port berths, partner portal handover status); (2) initial exception investigation (why is this shipment delayed, possible causes, which customers affected); (3) proactive notification coordination (which customers to notify, in what language, at what cadence). Bad fits: (1) autonomous changes to externally committed ETAs; (2) autonomous reroute decisions (touches cost, contract SLA, fuel); (3) autonomous compensation issuance. Principle: agents handle “gather / look up / propose,” dispatchers / ops / service leads decide “commit / change / communicate externally.”

05What is EKel’s custom AI experience in transport?

Honest answer: we have no publicly referenceable transport custom-AI case (our postal + maritime delivery experience is in the Salesforce platform layer). But the methodology (LLM Gateway design, RAG pipeline, eval framework, guardrails, Vibe Coding custom apps, agentic workflow design + limits, commercial-sensitive data tiering) is industry-neutral; we have applied this discipline in financial-services, aviation, retail, government, and education AI work — event-driven integration, high-volume tracking events, multilingual / multi-timezone service, ops-vs-commercial AI boundaries all transfer. Transport-specific domain (route conventions, port practices, cross-leg handovers, contract SLAs) we will learn alongside your team from week one.

Transport AI is most stable with “ETA + anomaly + multilingual service + draft.”

In 30 minutes we can map your compliance constraints, tracking integration state, and acceptable risk — then decide which AI use cases truly belong in production.

Book Free Consult→

We use cookies

We use strictly necessary cookies to run this site, plus optional analytics cookies (Google Analytics) to understand how visitors use it. See our Cookie Policy and Privacy Policy.