_1777711377206.png)
5 prioritised actions, not 50 alerts: rethinking the control tower
Most TMS dashboards drown ops teams in red. We rebuilt the control tower around the five decisions a dispatcher actually makes before lunch.
Detention overrun · Invoice mismatch ·
Yard congestion · Carrier offline · ...
Walk into any logistics control tower at 9 AM. The wall has six monitors. Each monitor has a dashboard. Each dashboard has between 30 and 80 things flashing red. In the middle of the room, two dispatchers are on phone calls. Neither of them is looking at the wall.
That's not a criticism — it's a tell. The wall is built for the people visitingthe control tower (auditors, COOs, the occasional VP). The phone calls are how the actual work gets done. Somewhere in the middle, a great TMS got sold and a real workflow got worse.
What we found when we shadowed dispatchers
We sat behind 11 dispatchers across 7 enterprise shippers for a full day each. We counted, with a clicker, every distinct decision they made. Across 11 days of observation, the median dispatcher made 37 decisions per shift — not 800, not 50. Of those 37, exactly 5 were the kind that couldn't be delegated, automated, or postponed.
The other 32 decisions had a default-correct answer 80%+ of the time. The other 200+ interactions were clerical — looking things up, copy-pasting, replying to "where is shipment X" messages. None of that needs a dispatcher.
The five decisions, in order
1. Indent allocation under capacity stress
When demand exceeds confirmed carrier capacity, someone has to decide which shipments move now and which slip a day. The model can rank options. The dispatcher commits.
2. Exception triage on in-transit shipments
The "truck has been stopped 4 hours, here's the driver's reason, here's the customer SLA, what do we do" call. Time-sensitive, financially material, relational. Always human.
3. Disputed delivery resolution
POD shows 24 cartons, customer's GRN shows 22. The dispatcher decides whether to credit, reship, or escalate. The data is on screen. The judgement is theirs.
4. Carrier performance escalations
A carrier slips below threshold for the third week. The dispatcher decides whether to cut volume, call the GM, or trigger a formal review. This one we can route, but not decide.
5. End-of-shift billing release
Which trips close today, which roll over. Cash flow downstream depends on this. It's a five-minute decision when the data is clean. It's a forty-five-minute decision when it isn't. Most TMSes optimise the wrong end.
"If your TMS is showing me 50 things, it doesn't know me. The whole job is filtering. Don't outsource the filter to me."
How we rebuilt the dashboard around five things
The new control tower has exactly five cards above the fold. Each card represents one of the five decisions. Each card shows: the decision, the recommended action with confidence, the data behind it, and one button to commit. That's it.
What we deliberately didn't do
We didn't add a "see all" button. We didn't add filters. We didn't add a "configure your dashboard" mode. Every one of those was requested. Every one of them, we politely declined. The whole point of the rebuild is that there is one right view of the morning, and we know what it is. Letting the user re-add the chaos defeats the rebuild.
We do have an "Everything" tab — exactly one click away — for the auditors and the COOs. They are not the user. They are visitors.
- 1State dashboards and action dashboards are different products. Don't ship one and call it the other.
- 2If a screen doesn't unlock a decision, it's a report. Reports go in a different tab.
- 3The hardest part of building an opinionated cockpit is saying no to power-user requests that re-introduce the chaos.
Writes about how the world's largest shippers actually run freight — the real workflows, the stuff vendors don't put in slides.
More from the team
After 14 customer interviews, every transporter chose WhatsApp over a slick web portal. Here's the data — and what it taught us about adoption in emerging-market logistics.
Our OCR pipeline turns smudged, curled, sometimes wet paper PODs into structured JSON in 30 seconds. Here's the architecture — and the failure modes nobody warns you about.
A teardown of the 90-day rollout — what changed in the indent process, the auction floor, and the settlement workflow.