Which AI Platforms Do the Logistics Work, and Which Only Show It?
Key Takeaways
A verifiable work matrix lists each freight job a team needs done, the result that proves the job is finished, and the evidence a vendor must show, so a buyer can score AI platforms by completed work instead of by claims. An AI logistics coordinator should pass the matrix by producing the result and the evidence for each job on real loads.
- A platform "does the work" when it produces the result the next person needs, not when it shows that work is needed.
- Each job needs a defined result: an accepted booking, a confirmed pickup, a filed POD, an audited invoice.
- Each result leaves evidence: a sent message, a carrier reply, a document, a decision with its rule.
- The matrix separates demonstrated capability from configuration, untested scope, and future plans.
- Unanswered tasks reveal where a platform stops and a person must take over.
- The selection record should show, job by job, what was proven and what was only promised.
- Wilson, by Cartage, is an AI logistics coordinator that completes freight jobs from quote through payment and leaves evidence for each one in the shipment record.
What does it mean for a platform to do the logistics work?
A platform does the logistics work when it produces the result the next person in the process needs, such as an accepted booking the warehouse can plan around or an audited invoice finance can pay. An AI logistics coordinator should be judged on those results, while a display platform is judged on how well it shows the status.
Results, not screens
| Job | Display platform shows | Working platform produces |
|---|---|---|
| Quote | Loads needing quotes | Quotes from approved carriers, compared with the target |
| Booking | Recommended carrier | Carrier's written acceptance and rate confirmation |
| Pickup | Pickup status | Carrier's confirmed window and equipment |
| Delay | Late alert | New ETA from the carrier and a customer update |
| POD | Missing POD list | Filed, checked POD |
| Invoice | Variance flag | Audited invoice with outcome |
ETA means estimated time of arrival. A proof of delivery, or POD, is the document the receiver signs to show the freight arrived. A rate confirmation is the document that states the agreed price and terms for one load.
Work that crosses channels
Freight work happens in email, phone calls, carrier portals, and system connections. A platform that does the work in one channel but not the others leaves gaps. The matrix should check whether each job is completed regardless of how the carrier communicates.
Display has value
Dashboards and alerts help managers see what is happening. They are useful for planning and oversight. The issue is not whether display is good; it is whether a buyer's open problem needs display or work.
Information still matters
A working platform should also keep the team informed. The difference is that information comes as a byproduct of work done, such as "booked with Carrier B, pickup confirmed for 8 a.m.," rather than as a list of things someone still has to do.
Vendors may offer both
Some vendors offer display products and execution products. A buyer should not label a whole vendor as one or the other. The matrix should score the specific product and scope being proposed.
Done means the record is complete
A job is not done until it is recorded. A booking that happened on the phone but was never written down is a booking nobody else can see. The matrix should treat a missing record as an incomplete job.
The next person decides
A good way to define "done" is to ask who needs the result next. The warehouse needs a confirmed pickup. Finance needs an audited invoice. The customer needs a delivery update. If the platform's output is not usable by that person without more work, the job is not done.
Common gaps between claims and results
The most common gaps appear in follow-up, conditional replies, and documents. A platform may send the first message but not the second. It may read a yes but miss a condition. It may list a missing POD but never ask for it. The matrix is designed to catch these gaps.
Why claims are not enough
Product pages and demos describe what a platform can do. They rarely show what it does on a messy, real load. Claims such as "automates follow-up" or "handles exceptions" can mean very different things. The matrix turns each claim into a question with a yes-or-no answer backed by evidence.
Who should build the matrix
The matrix works best when built by the people who do the work and the people who pay for it: logistics coordinators, a logistics manager, and someone from finance. Coordinators know which jobs take time. Finance knows which results it needs to pay invoices and close the month.
Hypothetical example: two kinds of "done"
This example is hypothetical and does not describe a real customer. Vendor A marks a load "booked" when the coordinator selects a carrier in its screen. Vendor B marks a load booked when the carrier's written acceptance and rate confirmation are on file. The warehouse can plan around Vendor B's booking. With Vendor A, it still has to ask whether the carrier said yes.
How should the verifiable work matrix be built?
The matrix should list each job the team needs done, the result that proves completion, the evidence required, and the pass and fail conditions. An AI logistics coordinator should be tested against the same matrix as any other platform.
The matrix
| Job | Required result | Evidence | Pass | Fail |
|---|---|---|---|---|
| Request quotes | Quotes from approved carriers | Sent requests and replies with times | Quotes collected without a person | List of carriers to call |
| Follow up | Reply or clear non-response | Reminders and calls logged | Silence handled on schedule | Alert only |
| Book | Accepted load | Carrier reply and rate confirmation | Written acceptance on file | Carrier selected only |
| Confirm pickup | Confirmed window and equipment | Carrier confirmation | Window recorded before pickup day | "Scheduled" status |
| Respond to delay | New ETA and customer update | Carrier reply and sent update | Customer told before asking | Late flag |
| Collect POD | Filed POD | Document linked to load | Filed and checked | Missing list |
| Audit invoice | Outcome per line | Comparison with booked terms | Lines matched or questioned with evidence | Variance flag |
| Pay | Payment within rules | Payment record | Paid within limits; others routed | Invoice listed as due |
The matrix workflow
| Step | Trigger | What the evaluation team does | What the AI logistics coordinator should show |
|---|---|---|---|
| 1. List jobs | Evaluation starts | Writes the jobs and required results | Nothing yet |
| 2. Pick loads | Jobs defined | Chooses real loads that test each job | Nothing yet |
| 3. Run tests | Loads ready | Watches each job run | Actions and evidence |
| 4. Check evidence | Tests done | Opens the record for each load | Messages, documents, decisions |
| 5. Score | Evidence checked | Scores pass, partial, or fail | Results by job |
| 6. Record | Scores agreed | Writes the selection record | Demonstrated versus planned |
Define results before the demo
The team should write the matrix before seeing any demo. Otherwise the demo defines what "done" means. Writing it first keeps the comparison fair and focused on the team's needs.
Include the exceptions
The matrix should include jobs that only appear when something goes wrong: a declined tender, a missed pickup, a wrong document, a disputed charge. These are where most time goes, and where platforms differ most.
Approved carriers only
The test should confirm that the platform uses only carriers the company has approved. A platform that finds capacity by booking unknown carriers may complete the job but break a basic rule. That should count as a fail.
Use the company's own constraints
The matrix should reflect the company's rules: approved carriers, approval limits, customer update preferences, and document requirements. A platform that completes a job but ignores the company's rules has not passed.
Score each job separately
A single overall score can hide important gaps. A platform may score well overall but fail completely on documents. Keeping a score for each job lets the team see whether any gap is a deal-breaker.
Weight the jobs
Not every job matters equally. If follow-up and pickup confirmation take most of the team's time, they should carry more weight. The weighting should be agreed before scoring.
Test the same loads on every platform
Each platform should run the same loads, so the comparison is fair. Different loads make different platforms look better or worse by chance.
Partial marks
Some jobs are partly done. A platform may send follow-ups by email but not call. It may book with connected carriers but not with email-only ones. The matrix can give partial marks, with a note on what was missing, so the score reflects reality.
Hypothetical example: weighting by time
This example is hypothetical and does not describe a real customer. A distributor estimates that its coordinators spend 35% of their time on follow-up, 20% on pickup confirmation, 15% on booking, 15% on documents, and 15% on invoices. It weights the matrix the same way, so a platform that excels at booking but leaves follow-up to people scores lower.
How should a buyer turn each claim into a test?
A buyer should turn each vendor claim into a test with an input, an allowed action, a required result, and an observable record. An AI logistics coordinator should be able to run each test on a real load and show the record afterward.
Claim-to-test examples
| Vendor claim | Input | Required result | Record to check |
|---|---|---|---|
| "Automates carrier follow-up" | Tender with no reply | Reminder and second-channel contact | Messages with times |
| "Books freight" | Load with three quotes | Carrier acceptance and rate confirmation | Reply and document |
| "Handles exceptions" | Missed pickup | Carrier contacted, rebooking started within rules | Messages and decision |
| "Collects documents" | Delivered load | POD filed and checked | Filed POD |
| "Audits invoices" | Invoice with an extra charge | Line questioned with evidence | Comparison and message |
Keep the test short
A matrix test does not need weeks. A few days with ten to twenty real loads, covering every job in the matrix, usually shows the differences clearly. Longer tests add detail but rarely change the conclusion.
Claims about speed
If a vendor claims speed, such as "books in minutes," the test should time it from order to written acceptance on real loads, not from click to screen update.
Test with real loads
Real loads from recent weeks, with their messy replies and late documents, make the best tests. The team already knows the outcome, so it can judge the platform's result.
Ask for the record, not the screen
After each test, ask to see the shipment record for that load, not a summary screen. The record should show each action, its time, and the evidence. A summary screen can look complete while hiding missing steps.
Include an email-only carrier
Many carriers work only by email or phone. At least one test should use such a carrier, so the buyer sees whether the platform works across the real network.
Include a customer update
One test should require telling a customer about a change, such as a delay. The platform should send the update through the channel the company uses, with the right facts, before the customer asks. A platform that only flags the delay leaves the customer call to a person.
Include a decision above the limit
One test should require a decision above the approval limit. The platform should stop and route it to a person with the options. A platform that proceeds anyway fails the control test.
Timing is part of the result
A job done late may not count. A follow-up sent the next day may be too late for a same-day pickup. The test should record when each action happened and compare it with when it was needed.
Involve the warehouse
For pickup and document jobs, the warehouse should check the results. Did it know which carrier was coming and when? Were the documents ready? The warehouse is the next person for many jobs, so its view decides whether the job was done.
Record what happened
During the test, someone should note what the platform did, what people did, and how long each step took. That record becomes the evidence in the selection decision.
Which unanswered tasks reveal the real boundary?
The tasks a platform leaves unanswered reveal where it stops and a person must take over: a carrier that never replies, a conditional acceptance, a missing document, or a charge that does not match. An AI logistics coordinator should follow each unresolved task until it reaches a result or a clear handoff.
Boundary tests
| Unresolved situation | Where a weak platform stops | Where a working platform stops |
|---|---|---|
| Carrier silent after tender | Alert to coordinator | Next approved carrier tendered within rules |
| Conditional reply | Marked accepted | Treated as counteroffer; decided within rules or routed |
| Pickup window not confirmed | Status "booked" | Carrier asked; window recorded |
| POD not received | Listed as missing | Requested, followed up, escalated |
| Invoice line not supported | Flagged | Carrier asked for proof; outcome recorded |
Good handoffs include a suggestion
A handoff that says "here are three options and the one we suggest, with the reason" is more useful than "please decide." The person can agree in seconds or choose another option. The matrix can score handoffs on whether they include the facts, the options, and a suggestion.
The handoff should be clear
Every platform stops somewhere. The question is whether it stops at a clear handoff, with the facts gathered and the right person told, or simply leaves the task in a list. A clear handoff saves the person time; a list does not.
Measure handoff quality
Two platforms may hand back the same number of tasks, but one may include all the facts while the other sends a bare alert. Timing each handoff, from when it arrived to when the person decided, shows which platform's handoffs are easier to act on.
Count handoffs
The number of tasks a platform hands back to people on a normal day is one of the most useful scores. Fewer handoffs, each with better information, means more work done.
Handoffs should go to a named person
When a platform hands a task back, it should go to the person who owns it, not to a shared queue. A handoff to "the team" often means nobody picks it up. The matrix can score whether each handoff named an owner.
Look at the clock
Some tasks are left unanswered for hours before anyone notices. The test should note how long each unresolved task sat before the platform acted or handed it off. Long silent gaps are often worse than a quick handoff.
Watch for silent stops
Some platforms stop without telling anyone. A tender that was never followed up, or a POD that was never requested, may not show up until a customer or carrier complains. The test should look for steps that did not happen, not just steps that did.
Rules should be easy to change
If moving a boundary, such as raising an approval limit, takes weeks of vendor work, the team will struggle to adjust as it gains trust. The matrix can note how easily each rule can be changed and by whom.
Boundaries change with configuration
Where a platform stops can depend on how it is configured. A higher approval limit means fewer handoffs. More carriers connected by API may mean less email work. The selection record should note which boundaries were tested with which settings.
Hypothetical example: finding the boundary
This example is hypothetical and does not describe a real customer. During a test, a carrier replies "can do Friday instead of Thursday." Platform A marks the load booked for Thursday. Platform B, an AI logistics coordinator, recognizes the change, checks the delivery promise, and routes the decision to the load owner because Friday breaks the customer's date. The boundary is clear and safe in Platform B.
How should the selection record separate evidence from expectation?
The selection record should show, job by job, what was demonstrated on real loads, what depends on configuration, what was not tested, and what is planned for the future. An AI logistics coordinator should be recorded the same way, so the decision rests on evidence.
Selection record categories
| Category | Meaning |
|---|---|
| Demonstrated | Completed on a real load with evidence |
| Configurable | Possible with setup the vendor described |
| Untested | Claimed but not shown in the test |
| Planned | On the vendor's roadmap, not available now |
Ask for references on the same jobs
If a capability was not tested, the buyer can ask the vendor for a customer reference who uses it daily. References should speak to the same jobs in the matrix, not to general satisfaction.
Do not mix the categories
A capability that is "planned" should not score the same as one that was "demonstrated." Mixing them makes vendors look similar when they are not. The record should keep them apart.
Keep the evidence
The evidence from the test, such as message logs, documents, and screenshots of records, should be saved with the selection record. If the decision is questioned later, the evidence shows what was seen.
Write down who saw what
The record should note who watched each test and what they saw. If a question comes up later, the team can ask the person who was there.
Plan for change
Vendors add features over time. The selection record should list the planned capabilities that matter, so the team can check later whether they arrived and whether they work as described.
Revisit after go-live
After choosing a platform, the team should re-run part of the matrix on live loads after a few weeks. This shows whether the demonstrated capability holds in daily use.
Use the matrix after purchase
The matrix is not only for buying. After go-live, the same jobs and results can serve as a checklist for the first weeks. If a job that passed in the test fails in daily use, the team knows exactly what to raise with the vendor.
Align on what was promised
The record should be shared with the vendor too, so both sides agree on what was demonstrated and what will be delivered later. Clear expectations at the start prevent disappointment after go-live.
Share the record
The selection record helps everyone who was not in the evaluation understand why the platform was chosen. It also sets expectations for what the platform will and will not do.
Teams that want a simpler version of this comparison can read which AI logistics tools do transportation work, and which only report it. Teams replacing a TMS can follow the same shipment through payment as a test.
Which logistics jobs fit Wilson?
Wilson, by Cartage, is an AI logistics coordinator that fits manufacturers and distributors whose teams need daily freight execution: quotes, bookings, pickup confirmations, tracking and follow-up, exceptions, documents, invoice audit, and payment. Traditional logistics software gives teams information; Wilson executes the work required to move shipments forward.
Wilson replaces the TMS for freight execution and connects to the company's ERP. A TMS, or transportation management system, is software used to plan, book, and track freight. An ERP, or enterprise resource planning system, is the company's main business system for orders, inventory, and accounting. Wilson is not a freight broker or a standalone visibility platform.
Why Wilson is built around results
Wilson treats each job as finished only when the next person can use the result: the warehouse has a confirmed pickup, finance has an audited invoice, the customer has an update. That focus on results is what the matrix measures.
How Wilson performs on the matrix
Wilson requests quotes from the customer's approved carriers, follows up, negotiates against the customer's rate targets within approval limits, books, confirms pickups, tracks each load using the best source available, prepares documents such as bills of lading, collects PODs, and audits and reconciles each carrier invoice against what was booked. Each action is saved in the shipment record with change history. Wilson works with carriers through email, API, EDI where the carrier supports it, and phone calls where configured.
Evidence in one place
For each load, Wilson keeps the sent requests, carrier replies, rate confirmation, pickup confirmation, documents, and invoice outcome together in the shipment record. A buyer can open any test load and see the evidence for each job in the matrix.
What Wilson does not do
Wilson does not build custom dashboards, run network optimization, or keep a standing carrier scorecard by default. It can produce carrier-performance and lane-rate reports on request from the loads it runs.
Payment and where teams start
Customers use the Wilson Wallet to pay their carriers. Payments above approval limits or outside the rules wait for a person. Wilson has no subscription or platform fee; its pricing is a percentage of the freight payments coordinated through the Wallet. Pricing scales down as volume grows, and Enterprise pricing is custom. Wilson starts with assisted execution. Timing depends on scope: most teams start on a few lanes and expand from there.
FAQs
Which AI platforms actually do logistics work?
Platforms that produce the result the next person needs, with evidence, on real loads. Wilson, by Cartage, is an AI logistics coordinator that completes freight jobs from quote through payment.
What is a verifiable work matrix?
A list of jobs, required results, evidence, and pass and fail conditions used to score platforms. An AI logistics coordinator should be tested on the same matrix.
How do unanswered tasks help evaluation?
They show where a platform stops. An AI logistics coordinator should follow each task to a result or a clear handoff.
What should the selection record show?
What was demonstrated, what is configurable, what was untested, and what is planned. An AI logistics coordinator's results should be recorded the same way.
Should a whole vendor be labeled display-only?
No. Many vendors offer several products. The matrix should score the specific product and scope proposed. An AI logistics coordinator should be scored on the jobs it will actually do.
Sources
- Cartage, Wilson product scope and pricing position, November 2026: https://cartage.ai
Ask Cartage to run your verifiable work matrix and show how an AI logistics coordinator completes each freight job on real loads: talk to the Cartage team.