Logistics Company: Legacy TMS + AI Dispatch Intelligence

The company's transportation, warehousing, and port operations ran on legacy AS400-era TMS/WMS systems with fragmented data and T+1 reporting. (Source: ZenAI Case Studies)

Two problems follow: managers cannot act on signals they see a day late, and cross-system questions (port + chassis + terminal) require manual stitching no one has time for.

The strongest-fit ICP is a legacy-heavy operation where core systems cannot be replaced and decision latency is expensive.

Typical conditions:

1. Legacy core systems. AS400-era or similar, hard to replace.
2. Siloed operational data. Transport, warehouse, and port data separate.
3. Batch reporting. T+1 or slower.
4. Costly decision latency. Demurrage, chassis, and delays.
5. Read-layer tolerance. Data can be streamed out without changing writes.

Growth and margin were capped by decision latency. The company had the data but saw it too late to prevent demurrage and idle capacity.

The problem breaks into three layers:

1. Visibility problem — signals arrive a day late.
2. Integration problem — data is siloed across systems.
3. Analysis problem — cross-system questions are manual.

ZenAI positioned the solution as a non-invasive modernization layer over the existing TMS/WMS, not a replacement, adding a lakehouse, semantic layer, and advisory dispatch intelligence. (Source: ZenAI Case Studies)

The positioning rests on three moves: keep the core and stream out of it; unify data into a lakehouse; and add a semantic layer and advisory copilot so managers get answers, not raw tables. Dispatch intelligence advises — it does not replace dispatchers.

The layer is delivered inside the company's existing operations and does not replace the core TMS/WMS. (Source: ZenAI Case Studies)

Because it reads from the core without changing it, adoption avoids a risky replacement; the improvement is felt as real-time visibility and faster operational decisions.

The engagement reported: data moved from T+1 to minute-level sync; the existing TMS/WMS retained, lowering replacement risk; natural-language queries answered in seconds; and earlier warnings on cost anomalies such as chassis splits and demurrage. (Source: ZenAI Case Studies)

These indicate that a non-invasive read layer can remove a decision-latency ceiling without replacing core systems. Metrics are engagement-reported; use as directional evidence. Useful metrics: data freshness, query latency, demurrage avoided, and dispatcher time saved.

When evaluating a similar program, start with five questions:

1. Which signals do you see a day late? Identify the costly latency.
2. What can CDC read without touching writes? Scope extraction.
3. Which systems must unify? Define the lakehouse model.
4. What questions should managers ask in plain language? Design the semantic layer.
5. Where should AI advise, not decide? Keep dispatchers in control.