Beta
With first users
Case study · Application
From n8n load tracking to a custom app that calls the driver when data goes dark
We shipped a Dispatcher + Worker loop on n8n, then transferred the full system into a custom application: self-scheduling load checks, AI voice check-ins, and a human exception queue when the automation cannot close the issue.
Zonik AI is a logistics product we co-founded.
- n8n to custom app
- AI voice agents
- logistics
- ops automation
Product
Zonik AI
Solution
Custom app (from n8n)
Impact
- 5 min
Tightest check interval
- 1 queue
Human exceptions
ProductApplication
CategoryLogistics
LocationSerbia
Year2026
Services used
- Application
- Automation
- n8n
- AI voice agents
Overview
n8n to find the contract, then a custom app to run it
Check intervals, risk rules, and call holds started as n8n knobs - easy to tune and replay. Then we moved the full loop into a custom application with the same contract.
Challenge
Someone has to watch every load
Someone has to watch GPS, spot late ETAs, and call drivers. Zonik AI is built so a person only handles what the system cannot settle.
Our side
- Dispatcher / Worker architecture
- n8n prototype + call result contract
- Full transfer from n8n into a custom application
- Monitoring for the system
Partner side
- Partner-led sales and operator relationships
- Fleet driver, truck, and load data
Solution
Dispatcher + Worker, self-scheduling each load
The Dispatcher wakes on a schedule and only pulls loads due for a check. The Worker evaluates one load - GPS, ETA, severity - then writes its own next check time. Quiet loads go quiet. At-risk loads get tight intervals and, when needed, an AI voice call.
Features
- 01
Self-scheduling triage
The system can check a load as often as every 5 minutes.
- 02
AI voice check-ins
When data cannot explain a delay, the system puts a voice agent on the phone and gets a structured reason back.
- 03
Call hold + force-call
No repeat calls after a clean answer. Dispatchers can force contact when they need it now.
- 04
Exception queue
Unresolved loads flip to needs-dispatcher. Humans work the queue - not every truck on the board.
Before → after
Before
I don't want people sitting on GPS all night.
After
Drivers get called when data breaks. Dispatchers handle what the system cannot settle. n8n was how we learned the loop.
“I don't want people sitting on GPS all night.”
Nenad Stevanovic
Co-founder, Zonik AI
Impact
Results
The loop now runs as a custom application in beta with Zonik's first users: triage, evaluation, voice contact, write-back, and escalation. It started as a full n8n Dispatcher + Worker system, then moved into owned code with the same job: the same contract, not a JSON dump.
What changed after launch
- Driver contact on the system's schedule
- Structured call results - not just audio to review later
- Dispatchers work a filtered exception queue
Technologies
- CA
Custom application
n8n
How we chose the models
Before switching models, we tested 14 AI models on real US broker rate confirmations in TestAImodels. The test found 7 reproducible extraction errors, all now fixed in the prompt, and showed that scanned and digital documents need different models. Since September 2026, Zonik runs two models on the rate confirmations its users upload, and any field where they disagree goes to the dispatcher to confirm.
Read the full test results →If your team is doing this by phone, we map the lightest version first
Watch state, contact a human when data stops making sense, extract structure, escalate the rest. We map the lightest version first. Book a free 30-minute strategy call and we will say whether that is a workflow or a product.
“I don't want people sitting on GPS all night. It watches the loads, calls the driver if something's wrong, and we only get what it can't finish.”
Nenad Stevanovic
Co-founder, Zonik AI
Free tool
Have an idea in mind? Try our product cost calculator
Instant results · Free · Takes about 3 minutes