Skip to content
The loopWorking togetherSecurityWorkSuiteJournal About Trust Center
Start a conversation NL

Home / Work

AI-Powered Daily Workforce Logistics

An event staffing company was spending significant time every day on manual transportation coordination. We built an AI agent that does it with a single Telegram command.

AI AgentProcess AutomationLogistics
Dailycoordination now automatic
Minutesof planning, from hours
Approvalbefore anything reaches a crew

The Problem

Crew4You is a Dutch event staffing company that provides construction crews for festivals, events, and pop-up installations across the Netherlands. With roughly 30 employees and around 10 dispatched daily, the logistics were a daily headache.

Every morning, the manager had to manually compose a detailed transportation plan: figuring out who has a car, which employees need a ride, calculating optimal carpool groupings, determining pickup times and routes, and finding public transit alternatives when carpooling was not possible.

This was not a five-minute job. It involved cross-referencing shift assignments, employee locations, vehicle availability, and travel times, all before the team even left for work. Mistakes meant someone arriving late to a job site, directly impacting client satisfaction.

What We Built

A fully automated logistics agent triggered by a single Telegram command. The manager types /planning and receives a complete, optimized transportation plan within seconds, ready to forward to the team.

How the System Works

1. Live data from the workforce platform. The agent connects to the company's shift management system and pulls real-time assignments: who is working where, when they need to arrive, and their contact details.

2. Smart carpool optimization. Using Google Maps, the system identifies which employees have vehicles, calculates driving times between all employee homes and job sites, and assigns passengers to drivers based on minimum detour time. It works backwards from required arrival times to calculate exact pickup moments.

3. Public transit fallback. When no carpool option works, no drivers at a location or detours that are too long, the system finds public transit routes. It specifies arrival time (not departure), returns specific bus and train connections, and flags "last mile" problems when the walk from a transit stop to the job site exceeds 1 km.

4. Complete Dutch planning message. The output matches the exact formatting the team is used to: date headers, per-location blocks with job details, carpool instructions with pickup times and addresses, public transit directions with line numbers, and standard reminders.

5. Interactive problem resolution. When the system detects an impossible transport scenario, it does not just fail. It sends a message with inline buttons: accept a late arrival or manually reschedule. Edge cases become quick decisions instead of blockers.

Challenges We Solved

The carpool math is not trivial. With multiple drivers, multiple passengers, and multiple job sites, finding the optimal grouping requires testing permutations. We built deterministic route optimization that handles this reliably, no AI guesswork for calculations that need to be exact.

Public transit quirks. Google Maps sometimes suggests leaving the night before for very early shifts, or proposes four-hour journeys for domestic travel. We built validation layers: same-day departure checks, duration caps, and honest messaging when no viable transit route exists.

Real-time accuracy. Shifts change frequently. The system fetches fresh data on every request rather than relying on cached schedules, slightly slower, but the planning is always based on the latest assignments.

The Result

What used to consume significant daily time now takes seconds. The manager types one command and gets a complete, optimized plan. The system handles last-minute shift changes, employees calling in sick, and venue changes, all automatically.

More importantly, it removed a scaling bottleneck. Before, growing the team meant linearly growing the coordination overhead. Now, whether Crew4You manages 10 or 50 daily assignments, the logistics coordination takes the same effort: one Telegram tap.

"I used to dread the morning planning. Now I type one command and it's done. The system even handles weird edge cases better than I did manually."

Technology

  • Python for the core automation engine
  • Google Maps API for route optimization and transit routing
  • Telegram Bot API for delivery and interactive problem resolution
  • Modal for serverless deployment, zero infrastructure to maintain

Honestly

The part that was genuinely difficult

The real constraint was people

Route optimisation is a solved problem. Who has a car, who can drive, who lives near whom and who cannot work together is not — and that is what actually determines a workable plan.

It had to be trusted before it was automatic

The output goes to real crews at six in the morning. Nothing sends without a human approving the draft, and that was a deliberate design decision rather than a limitation.

Why it worked

Because we automated the arithmetic and left the judgement where it belonged. The planner still decides; they just stopped doing the maths by hand at midnight.

The product

What it actually looks like.

Rebuilt in code for this page rather than screenshotted, so it stays legible and translatable. The real interface carries the client's own branding and data.

Tell us what your business runs on.

If it is the kind of system that cannot have a bad day, we should talk. One conversation, no deck, with the people who would build it.