Monitoring 2,000+ Insured Shipments a Day Without a Monitoring Team
A client who insures high-value FedEx shipments asked for a team to watch the tracking. We built an automation on n8n instead, live in about two weeks: any shipment quiet for four hours is flagged, at 24 hours the client is alerted, and one person spends about ten minutes a day on it.
- Systems Shipped
- 50+
- Client Retention
- 98%
- Uptime
- 99%
- Engineers & Designers
- 40+
2026-08-01/TechCentera TeamThe Obstacle
The client's customers ship high-value goods with FedEx through the client's platform and insure them through it. Many shipments are worth $50,000 to $200,000, and any one that goes quiet is a potential claim. A shipment that stops updating might be delayed, misrouted, held, or gone, and the difference between those is the difference between a phone call and a payout.
At the time, around 200 shipments were moving on any given day, and the plan was to hire a team of people to watch the tracking: open each shipment, check the last scan, note the ones that had stopped, every day, around the clock. The volume was going to grow, and so would the team.
In discovery we asked what that team would actually be looking for. The answer was a shipment with no new status. A tracking page can't show a problem, but it does show silence, and silence can be written as a rule.
The Design Decision: A Rule, With AI Only Where a Rule Can't Reach
The decision that matters is whether a shipment has gone quiet, and that is arithmetic: the time since the last scan against a threshold. No judgment, no model. So the core of the system is a rule, run by a workflow on n8n, not an AI agent.
Two things are not arithmetic. A scan's timestamp means nothing until you know what time it was where the scan happened, and that changes with every hub a package passes through. And a run of scans has to be read as activity, not just counted. Google Vertex AI does those two jobs: it works out the local time zone of each scan and reads the shipment activity, so the four-hour clock is measured correctly wherever the package is.
What We Built
- The intake. Shipment reports arrive by email. The workflow picks each one up as it lands, extracts every shipment on it and checks it against the log, so nothing is recorded twice.
- The tracking. For every shipment, the FedEx Tracking API returns the current status, the latest scan location, the timestamp and the tracking details. Vertex AI works out the local time zone of each scan and reads the activity, so "four hours since the last scan" means the same thing in every place a package passes through.
- The flags. Green means moving. Four hours without a new status and the shipment is flagged: the automation keeps watching it, and so does the team. Twenty-four hours and it turns red: high alert, and the client is notified so they can raise it with their FedEx account manager while there is still time to act.
- The log. Every shipment is recorded in Google Sheets: a complete tracking log, and a separate list of flagged shipments with duplicate checks, so the flagged list never carries the same shipment twice.
- It runs 24 hours a day, 365 days a year, and it went live in about two weeks.
What the System Looks Like Day to Day
Reports land by email and the workflow takes it from there. Through the day the tracking log fills with scans; shipments quiet for four hours move to the flagged list; anything past 24 hours turns red and the client is notified. At the end of the day, one person on the existing team opens the sheet for about ten minutes, goes through the flagged list, and follows up with the client on the reds so they can take them up with FedEx. Nobody was hired. Nobody watches a screen.
What Changed
- A planned monitoring team replaced by one rule, two AI steps and about ten minutes of one person's day
- 2,000+ shipments a day watched, up from around 200 at the start, with no change in headcount
- A quiet shipment is flagged at four hours and escalated at 24, every day of the year
- The client hears about a problem shipment in time to raise it with their FedEx account manager, not when a customer calls
- Every shipment, scan and flag logged in a sheet, so there is a record of what was seen and when
What We'd Tell Another Founder
The request was "we need people to monitor this." The need was "tell me when a shipment goes quiet." Nine times out of ten the second version has a cheaper answer than the first. Ask what the people would actually be looking for, and you usually find the rule.
Then be honest about where the rule stops. Here it stopped at telling the time across time zones and reading what a run of scans means, so that is exactly where AI went, and nowhere else. One rule, two AI steps, one person for ten minutes a day: that proportion is what most AI automation should look like, and it is why the volume could grow tenfold without the headcount growing with it.
Tell Us How Your Operation Runs.
We'll Tell You What to Build.
A free 30-minute consultation with our founder. You leave with a clear view of what to build, what to automate and what to leave alone. No pitch deck. It is the shortest way to find out whether we are the right custom software development company for your operation, and we'll say so either way.
Book a Free ConsultationFive initial discovery calls a week, one a day, 11 AM to 3 PM Eastern.