A forward deployed pod of 3–5 tkxel engineers builds AI agents that read freight and customs documents, work shipment exceptions and support maintenance, connected to your ERP, TMS, WMS and MES, with people approving every filing and work order.
Book discovery callAgents extract data from bills of lading, commercial invoices, packing lists and certificates of origin, cross-check them against each other and the booking, suggest HS codes for unclassified lines and stage the entry in your customs system for a licensed broker to file.
Agents watch carrier EDI, telematics and TMS milestones for missed pickups, dwell and delays, find the affected orders, draft the customer notification and propose a rebooking. Dispatchers approve the action, and each exception is tagged by cause for carrier scorecards.
Agents compare carrier invoices with rate confirmations, contracts and proof of delivery, flag accessorial charges such as detention or lumper fees that lack support, and queue disputes for your freight audit team with the evidence attached.
Agents read sensor alerts, operator logs and equipment manuals, suggest a likely cause and procedure, and draft a work order in SAP PM, Maximo or Fiix with parts availability checked. Planners approve, and agents never send commands to control systems.
The pod builds the integrations agents and people both need: order, inventory and production data moving between SAP, Oracle or Dynamics, your MES and your WMS, with reconciliation reports so planners stop rekeying data between screens.
Agents read supplier acknowledgements, ASNs and email replies, compare promised dates with MRP requirements, chase late confirmations and warn buyers of shortages before they stop a line. Buyers keep authority over expedites, substitutions and price changes.
We’ll map the documents, systems and approvals for one workflow in 30 minutes.
Book a scoping callScanned, stamped and carrier-specific paperwork defeats template OCR, so pilots tuned on clean samples fail at real volume.
ERP, TMS, WMS and MES each hold part of the picture, and an agent without access to all of them guesses.
Plant networks are segmented for good reason, and any connection near control systems needs review against standards such as IEC 62443.
A dispatcher or planner will not wait on a slow tool, so latency and clear escalation matter as much as accuracy.
We sit with brokers, dispatchers or maintenance planners to trace the documents, systems and approvals in one workflow, and collect a sample of real paperwork. Your IT and OT security teams agree which networks and systems the pod may touch.
Deliverables: workflow map, document sample set, integration plan, network access approval
The pod builds the agent and its integrations to your ERP, TMS, WMS or CMMS in your environment, reading plant data only through approved historians or gateways. We test against historical shipments or work orders your team has already resolved.
Deliverables: working agent, system integrations, evaluation set, exception taxonomy
The agent runs on one lane, warehouse or production line with people approving every filing, booking or work order. Their corrections become test cases, and we measure accuracy by document type and exception class.
Deliverables: pilot results, error analysis, approval workflow, monitoring dashboard
Your team takes ownership with runbooks and a playbook for adding sites, carriers or document types. Code and configuration stay in your repositories, with no per-document licence tied to tkxel.
Deliverables: runbooks, rollout playbook, source repositories, training sessions
We build in your repositories and environments from day one, so there is no separate rebuild between proof of concept and release.
Each pod is led by an engineer who has shipped AI systems before, drawn from tkxel’s team of around 150 AI professionals. The same people stay through delivery.
Autonomy matrices, audit trails, and human review are designed with your risk and security teams before release, backed by ISO 27001 controls.
Every engagement starts with agreed business metrics and evaluation sets. We report against them throughout, and the pod exits when your team can run the system.
Mostly on the paperwork and coordination between systems: reading freight and customs documents, working shipment exceptions, auditing carrier invoices, following up with suppliers and turning maintenance alerts into draft work orders. Agents pull data from your ERP, TMS, WMS, MES and email, prepare the next step and ask a person to approve it. They work alongside your planning and control systems rather than replacing them.
Current models handle varied layouts, stamps and handwriting far better than template OCR, but accuracy still depends on scan quality and how unusual a carrier’s format is. That is why we cross-check fields across the bill of lading, commercial invoice and packing list and route mismatches to a person. For customs, the agent prepares and suggests; a licensed customs broker reviews and files, because classification errors carry penalties the broker answers for.
No, and we advise against it. Our agents read operational data through historians, MES or approved gateways and write only to business systems such as your CMMS or ERP. Sending commands to PLCs or SCADA would put a probabilistic model inside a safety-critical loop, which IEC 62443-style network segmentation exists to prevent. If you want closed-loop process optimization, that is a controls engineering project, and an agent build is the wrong tool for it.
We integrate with common platforms such as SAP, Oracle, Microsoft Dynamics, NetSuite, Manhattan, Blue Yonder and Maximo, using whatever APIs, EDI feeds, database views or file drops your systems expose. Where an older system has no API, we use your approved integration middleware or, as a last resort, controlled screen automation. We confirm the integration path for each system during scoping, before committing to a timeline.
We start with one lane, warehouse or plant so the agent is tested on real volume, then add sites with a rollout playbook covering new document types, carrier formats, languages and local approval rules. Each addition gets its own test set before going live. Agents can read documents in the major trade languages, but customs and labour rules differ by country, so each new country needs a review with your compliance team.
About 12–16 weeks for the first workflow at one site or lane, including a pilot with human approval on every action. The main variables are integration access, particularly to older ERP or WMS instances, and OT security review when plant data is involved. Expanding to further sites is usually faster because integrations, test sets and approval workflows are reused rather than rebuilt.
Strategy, Agents
8 October 2026 · 8 min read
Read article
Agents, Integration
1 October 2026 · 8 min read
Read article
Governance, Agents
1 October 2026 · 8 min read
Read article