Before you check market, then tech
Seven hypotheses, each with a stop criterion, tested before we build more
- 8 interviews by 13 November, against seven hypotheses with stop criteria
- 2 offices share last month’s loads; we replay what zhol would have flagged
- 2 live tests on real restriction days, with the dispatcher’s verdict
The prototype exists so that dispatchers have something to react to. Nothing new is automated until the interviews and a supervised pilot confirm it.
interviews: the last disruption, what it cost, what they pay for software, four price questions
offices share last month’s loads; zhol replays what it would have flagged
live tests on real restriction days, with the dispatcher’s verdict
Targets for Workshop 2, 13 November 2026.
Breadth as well as depth: seven questions, one minute, one tap per answer: time spent checking, cost of the last incident, a fair price, whether they would use the evening alert, and whether they would share what their drivers see. It feeds four of the thresholds above, so the eight interviews are not the only numbers. Target: 30 answers from offices with lanes abroad. Open the survey
What an office gets back in two days: its past month run through zhol, every trip it would have flagged, the hours and euros, and a column for the dispatcher to mark each flag right or wrong. See the example report
| Hypothesis | Confirmed if | Otherwise |
|---|---|---|
| Offices re-plan loads because of restrictions every week | 6 of 8 report 2 or more re-plans a month | Move to the South–North blockade segment |
| They can put a cost on it | Median €300 or more per incident | Sell time savings at a lower price |
| The owner will pay | Median acceptable price €8 per vehicle or more | Sell through partners and associations |
| The evening before is when they decide | Most re-plans happen between 16:00 and 21:00 | Move the alert to the moment they do decide |
| Offices share tomorrow’s loads every day | 2 share history; 80% of pilot days on time | Build the Excel and TMS import first |
| Matching is precise enough | Precision 80% or more; at most 1 missed problem a month | Narrow to official sources and fewer corridors |
| Review cost stays small | Median 90 seconds per event | Automate trusted sources before adding offices |