For
Logistics & freight
Logistics, freight & supply-chain shared services
For logistics and freight teams, the automatable AI workflow is the EDIFACT exception queue, not the already-automated happy path. Chokmah brings production EDI experience, classifies exceptions by failure mode, and builds a classifier-and-proposal step with a human checkpoint: leaving code the operations team owns.
- The exception queue is the automatable workflow; the EDIFACT happy path is usually already automated.
- Chokmah's EDI/EDIFACT experience is real production experience: the domain wedge no generalist can copy.
- Exception queues collapse into six or seven recurring failure modes; the distribution decides the build.
- Segments carrying commercial terms stay under human sign-off permanently: a wrong correction is a wrong booking.
- Because the queue is already logged and owned, a baseline exists before the build starts.
Where it hurts
The exception queue eats a small team alive
The happy path is already automated. What is not is the queue of EDIFACT messages that fail validation, carry an unknown partner code, reference a record that does not exist yet, or arrive out of sequence. Each one is opened by hand, read segment by segment, corrected and resubmitted.
Partner-specific mappings live in two people's heads
The knowledge of why partner X sends a non-standard code and how to fix it sits with two or three operations staff and in years of ticket history: undocumented, unversioned, and a single point of failure when they take leave.
Volume rises with partners, not revenue
Every new trading partner adds mapping variations and exceptions. The workload grows with the partner book rather than the top line, so the queue quietly gets worse as the business grows.
A wrong correction is a wrong booking
EDIFACT messages are legally significant trading-partner documents. An incorrect automated edit to a rate, quantity, party or date segment is not a bad answer. It is a wrong booking with commercial and audit consequences.
The market, with sources
MIT found more than half of enterprise GenAI budgets went to sales and marketing, despite better returns in back-office automation.
MIT found 95% of enterprise generative AI pilots delivered no measurable P&L return, with the cause an organisational workflow gap.
Only 4% of Indian firms have embedded frameworks to manage AI risk, though 83% of executives call governance essential.
This is the ground we can defend
Most of what a consultancy tells you about AI in logistics is generic. This page is not, because the EDI and EDIFACT experience behind it is real production experience: parsing IFTMBF, IFTMIN and IFTSTA messages, handling partner-specific mappings, and working the exception queue that everyone else automates around. That is the differentiator no course catalogue and no generalist competitor can copy.
A freight or logistics shared-services team runs inbound EDI for a book of trading partners. Bookings, shipping instructions and status messages arrive as EDIFACT and are mapped into an operations or booking platform. The happy path is usually solved. The exception queue is where the manual work (and the automatable workflow) actually lives.
UNH+1+IFTMIN:D:99B:UN'
BGM+340+REF-88213+9'
DTM+137:20260724:102'
NAD+CZ+PARTY123::87'
RFF+CN:CONTAINER-NOT-YET-CREATED'
LOC+9+INMAA:139:6'
LOC+11+NLRTM:139:6'
MEA+WT+G+KGM:19850'
UNT+9+1'Why the exception queue, not the happy path
The happy path is already automated by the mapping layer: building an agent to redo it adds nothing. The exceptions are different: they are bounded, already logged, and already owned by a named team, which means a baseline exists before you start. A workflow you cannot baseline is a workflow you cannot prove anything about later.
Most exception queues collapse into six or seven recurring shapes: an unknown partner code, a missing master-data reference, a segment-order violation, a date-format variance, a duplicate submission, a genuinely malformed message. The distribution decides the whole engagement. If one shape dominates the volume, that shape is the build and everything else waits. The EDI exception triage scenario walks the full method.
The three workflows we see most often
Across freight and supply-chain shared services, the same automatable workflows recur.
- •EDI ops
EDIFACT exception triage
Classify the failure, retrieve the partner rule and prior resolutions, and propose a segment-level correction for a person to accept. Never submit unattended.
- •Documents
Shipping-document retrieval and reconciliation
RAG over bills of lading, manifests and partner agreements, where retrieval quality decides accuracy and versioned documents defeat naive chunking.
- •Master data
Party and container reference resolution
Matching inbound references against master data that may not exist yet: a bounded, high-frequency lookup that is ideal first-agent territory.
What we would refuse to automate
We would not build unattended correction of any segment carrying commercial terms (rate, quantity, party, date) because an automatic edit alters a legally significant document. We would not paper over a partner that produces repeated structural failures, because the fix there is a conversation with that partner, not a model that learns to hide the problem. And we would not introduce a new step during peak season in a first release, when a wrong correction costs the most. This refusal list is non-negotiable #1 made concrete.
Where we would start
With an Adoption Diagnostic to classify twelve months of exception history, then a Workflow Sprint that builds a classifier-and-proposal step with an evaluation harness on frozen historical cases. You own the repository, the harness and the runbook.
Who this is not for
The domain expertise is real; the engagement is a scenario. If you are a large IT services firm with your own EDI practice, or one of the Fortune 500 MNC GCCs under a global master-vendor agreement, we are not the right fit.
Put the exception queue to work
Start with a diagnostic that classifies a year of exception history by failure mode, then a sprint that ships a classifier-and-proposal step your operations team owns. Real EDI experience, honest method.
Relevant scenarios
Frequently asked questions
Not the whole of it, and that is the wrong target. In a mature freight operation the happy path is already automated by the mapping layer. What remains manual is the exception queue, and that is where an agentic system earns its keep: classifying the failure, retrieving the partner-specific rule and prior resolutions, and proposing a segment-level correction for a person to accept. Unattended correction of commercially binding segments is not something Chokmah would build.
Because exceptions are already logged, already timed and already owned by a named team, so a baseline exists before you start, and a workflow you cannot baseline is one you cannot prove anything about later. The exception queue also already contains a human checkpoint, so adding an agent proposal step changes the work without changing the control model.
Real production experience: parsing forwarding and status messages such as IFTMIN, IFTMBF and IFTSTA, handling partner-specific mapping variations, and working the exception queue rather than automating around it. This is the one genuinely defensible differentiator on the site, which is why the logistics and freight page is written from it and says so plainly.
Three things, permanently: unattended correction of any segment carrying commercial terms such as rate, quantity, party or date; structurally malformed messages from a single partner, because the fix there is a conversation with that partner; and a first release during peak season, when a wrong correction is most expensive. Naming these before starting is a diagnostic obligation, not a limitation.
See what a two-week diagnostic finds
We interview your people, shadow two workflows, and score which three to automate, and which to leave alone.