AI in SAP TM — what's real

Every TM conversation now has an AI slide in it. As a practitioner, I care about a narrower question: which of it changes what a planner, a dispatcher or a freight-cost accountant actually does this quarter — and what does each use case quietly assume about the system it lands in?

Where AI already earns its keep in transportation

• Arrival prediction: predicted ETAs from telematics and historical execution data are the most mature AI in logistics — useful exactly in proportion to how connected your carriers are. An ETA model on top of unconnected carriers predicts nothing.
• Document drudgery: extracting structured data from freight invoices, delivery documents and rate confirmations. Unglamorous, real and measurable — this is where settlement teams feel AI first.
• Optimisation's quiet expansion: TM's planning and carrier-selection optimizers are not new AI — but their inputs (cost models, constraint sets) increasingly are. The optimizer was always willing; the data rarely was.
• Assistants and agents: SAP positions its assistant and agent layer across the suite, transportation included. The honest practitioner posture: treat every announced capability as a hypothesis to verify against the current roadmap and your own release — not as a plan input.

What every use case assumes — the unglamorous prerequisites

• Execution truth: AI on top of freight orders that don't reflect physical reality (see execution & tracking) automates a fiction.
• Clean charges: prediction and audit use cases stand on disciplined charge calculation — rate structures a model can learn are rate structures a human can explain.
• Connected partners: most valuable signals originate outside your system — carriers, terminals, telematics. The network question (SAP Business Network for Logistics) precedes the AI question.
• Somewhere for context to live: cross-system AI assumes unified, governed data. That architecture conversation is the executive track's territory — see shared business context & unified data.

How to read the roadmap without the gloss

Three questions filter ninety percent of the noise: Does this use case run on data we already produce reliably? Does it remove a task someone does weekly, or add a dashboard nobody asked for? And if the model is wrong, who notices, and how fast? A use case with good answers to all three is worth a pilot. A use case with none is a slide.

The bridge to the WHY

Whether AI in transportation pays is a practitioner question. Whether your organisation is structured for it — data, governance, operating model — is an executive one. That side of the argument lives in the Autonomous Enterprise and AI agents in ERP on Executive Clarity — the WHY to this page's HOW.

Next step

Which AI use case has actually survived contact with your operation? The SAP TM Community on LinkedIn is collecting real answers — insight over noise, facts over fog.