When we first walk into a freight forwarder, we ask the same question: how many minutes per shipment go into documents versus into actual movement decisions? The answer is almost always more than forty percent, and almost always a surprise to the operations director. Document work is the silent tax on planner capacity. It does not show up on the dispatch board, it does not get measured against KPIs, and it is the single largest reason planners report being too busy to think.
The bottleneck is not data entry
Modern systems extract fields from CMR, Bill of Lading, and customs documents with reasonable accuracy. Optical character recognition is a solved problem for the structured 80 percent of fields on a typical waybill. The bottleneck is exception management: missing fields, ambiguous addresses, customs mismatches, reference numbers that do not line up between booking and physical paper. McKinsey notes that the logistics technology landscape remains extremely fragmented, with 37 percent of providers running five or more separate solutions in warehousing alone, and 34 percent running eight or nine different transportation tech systems [1]. Exception handling sits in the gaps between those systems, which is why it does not get automated in the first place.
Document automation does not save planners from documents. It saves the operations director from being the person who has to track them down.
What to automate first, in order
Three checks, in sequence: completeness, internal consistency, cross-system consistency. Completeness is a structured-extraction problem, well-served by intelligent document processing. The global IDP market is forecast to grow from USD 2.4 billion in 2024 to USD 37 billion by 2033, a 35 percent compound rate driven mostly by exactly this layer [2]. Internal consistency is a rules problem: weight totals match line items, addresses validate, dates are plausible. Cross-system consistency is where AI starts to earn its place, because the same shipment is referenced in four different ways across TMS, WMS, customs broker, and client portal. Reconciling those references is what saves planner time. Skipping the first two layers to jump straight to the third is the single most common mistake in logistics AI projects.
The four signals that tell you it is working
First, planner-minutes-per-shipment drops by at least 25 percent within eight weeks of deployment. If the number does not move, the automation is not touching the binding constraint. Second, exception escalation queue depth stops growing. Most forwarders see queue depth grow linearly with volume; with proper automation, growth becomes sub-linear. Third, customs hold rate drops, because cross-system consistency catches mismatches before the broker does. Fourth, planner job satisfaction scores improve, which sounds soft but predicts retention more reliably than salary.
Why most freight automation projects stall
Two patterns. The first is tool selection driven by IT rather than operations: a platform gets selected for its API surface or integration story, then operations is asked to fit. The second is a too-broad scope: automating CMR, customs, invoicing, settlement, and tracking in one initiative, which creates a multi-quarter project that ships nothing useful for the first six months. Both patterns produce the same result, which is automation that lives next to the workflow rather than inside it. McKinsey research finds that while 85 percent of logistics companies say their digital projects added value, most still report significant integration challenges [3]. The fix is to start narrow: one document type, one exception class, one team, four weeks. Measure planner-minutes saved. If the number is real, expand. If not, the diagnosis was wrong, and a bigger project would have failed bigger.
Start narrow. One document type. One exception class. One team. Four weeks. If the planner-minutes number does not move, the diagnosis was wrong.
The Belgian context
For Belgian forwarders specifically, the CMR is the right starting document. It is the most volume-dense, the most error-prone, and the most consequential for downstream customs work given the port of Antwerp's role in EU trade flows. Manual CMR processing typically costs a mid-market forwarder between EUR 8 and EUR 14 per shipment in fully-loaded labor. Automating the completeness and internal-consistency layers brings that to under EUR 3 within a quarter for a forwarder handling 50,000 shipments per year. That is EUR 250,000 in recovered planner capacity, available for higher-value work. The cross-system consistency layer is where the next million in capacity hides, but only after the first two layers are stable.
What to measure on day one
Three baseline numbers, captured before any automation goes live: planner-minutes-per-shipment broken down by activity, exception escalation rate by document type, and customs hold rate by client segment. These are the only numbers that matter for the first quarter. Vanity metrics like documents processed per hour or extraction accuracy are tracked but not optimized for. The goal is operational impact, not technical impressiveness. The forwarders who confuse the two end up with very fast document processing and exactly the same planner workload they started with.
A note on the AI versus RPA question
Document automation in logistics is rarely pure AI. Most working deployments are 70 percent rules-based with 30 percent machine learning at the edges where rules fail. This is fine. The framing of 'AI-powered' matters for marketing but not for outcomes. The right question is not which technology stack is in play but which exception class each automated step is handling and how the workflow degrades when the step fails. Designs that fail gracefully outperform designs that try to do too much, every time.
