We tracked 1.2M PTL shipments. Here's what surprised us.

We pulled every PTL (part-truckload) shipment that touched the Traqo network in the last 18 months. 1.2 million shipments, 14 carriers, four sortation models, and the whole spread of customer SLAs. Then we asked the data a handful of questions we'd been arguing about internally. Most of the answers were different from what we expected.

Surprise 1 — Pickup is a 4-hour event, not a 30-minute one

Carriers quote pickup windows like "10 AM – 12 PM". The actual median pickup happens at 12:14 PM, and the 90th percentile is 4:08 PM. The "window" is closer to a "promise band with a strong rightward skew". Once you internalise that, three things change in your ops planning: dock booking, driver detention, and outbound transfer planning at the origin hub.

Surprise 2 — The variance is predictable

We expected carrier-level variance to be the dominant factor. It isn't. About 60% of the variance in pickup time is explained by three predictable inputs: the day of the week (Mondays are 38 min later than Wednesdays), the originating PIN code's hub distance, and the customer's average dock turnaround time as observed historically.

The implication is operational, not analytical: if your control tower can predict the actual pickup time within ±45 minutes, you can plan the rest of your day around it. Most TMSes don't, because they show you the carrier's quoted window — which is the wrong number.

Surprise 3 — Hub dwell is the silent killer

For shipments that miss SLA, we expected the failure to be on the road. It isn't. The single largest contributor to late delivery is dwell time at the originating hub — specifically, the gap between hub-in and hub-out at the first sort. Across the dataset, the median dwell is 6.2 hours. The shipments that miss SLA have a median first-hub dwell of 14.8 hours.

Surprise 4 — Damage rate is bimodal

Carrier damage rates aren't normally distributed. They're bimodal — most carriers cluster between 0.4–0.9% damage, but there's a small group sitting at 2.1–3.4%. The cluster gap is striking once you see it on a histogram. The bad cluster shares one characteristic: a sortation model with manual cross-dock at intermediate hubs. The good cluster uses sealed bay-to-bay transfers.

Surprise 5 — Billing accuracy correlates with delivery accuracy

We didn't expect this one. Carriers with above-median on-time delivery also had above-median billing accuracy (correctly weighted, correctly rated, no spurious detention charges). The correlation isn't subtle — it's r = 0.71 across 14 carriers. The story is probably mundane: carriers with disciplined operations have disciplined back-offices. But the implication for shippers is concrete — if you're choosing between two carriers on price alone, the cheaper one's invoice is statistically more likely to be wrong.

What we changed in the product

We made four changes to Traqo's PTL module on the back of this analysis:

1. Pickup ETAs are now learned, not quoted. Old field is still there but secondary.
2. Hub dwell appears as a first-class signal in the at-risk-shipment view, ahead of road delays.
3. Carrier scorecards now report damage rate alongside its sortation model, so the conversation is structural.
4. Billing accuracy is now part of the carrier health score, not a separate finance metric.

What we'd want to look at next

Two open questions we didn't answer here: (1) does same-day pickup correlate with higher delivery damage (because of rushed loading)?, and (2) is there a measurable lift from posting expected hub-out times to the customer in real time, or is it just theatre? If you have data and want to compare notes, we're at hello@traqo.ai.