AI Watchdog
Scans fleet data on a schedule for anomalies in fuel, compliance deadlines, driving behaviour, bookings, maintenance, cost and utilisation, and explains each one in plain English with suggested actions.
Fleet AI is only useful if you can check it. Every insight Vectis produces names the trips, vehicles and records it came from, and stays inside your organisation's data.
LTC556 fuel cost per 100 km up 34% on its trend
Distance flat, idle time +61%. Two stops now exceed 25 minutes with the engine running.
3 trips cited →14 journeys unclassified with the period closing
Held by four drivers. One reminder clears the period.
Send one reminder →KMR421 is 900 km from its service interval
At its current weekly distance the service falls due before the next free slot on the board.
Service schedule cited →The pool is one vehicle larger than demand requires
Peak concurrent demand reached 11 of 14 twice in 13 weeks, on the same Thursday slot.
See the demand curve →Scans fleet data on a schedule for anomalies in fuel, compliance deadlines, driving behaviour, bookings, maintenance, cost and utilisation, and explains each one in plain English with suggested actions.
Threshold, trend, missed-deadline and statistical-outlier rules, each with its own severity and cool-down. System defaults to start with; your own rules on top.
A daily or weekly summary of fleet health: highlights, anomalies, deadlines approaching, cost trends and recommended actions.
Natural-language questions answered from your organisation's data with the sources cited, on the web and in the mobile app. Follow-up questions keep their context.
“Explain this report” on a report page reads the numbers for you: trends, outliers and what to do about them, next to the report rather than in a separate dashboard.
Guides a new organisation through settings, pools, vehicles, users and its first booking, and answers setup questions in context rather than sending you to a manual.
The constraints matter more than the capabilities. These are enforced in the service, not asked for in a prompt.
Every query is scoped to your organisation. Another customer's fleet cannot inform an answer about yours.
Answers cite the records they were built from, so a figure can be checked against its source.
The assistant proposes; a person confirms. Bookings, classifications and records are not changed without an explicit action by someone with the permission to take it.
Vectis does not train models on your fleet records, and the AI provider's API terms exclude training by default.
Which vehicles cost us the most per kilometre last quarter, and why?
Three vehicles sit above the fleet median of $0.41/km for 1 Jan – 31 Mar:
LTC556 is the outlier: fuel is 61% of its cost per kilometre against a fleet average of 38%, driven by idle time on the stores route rather than distance.
Knowledge sources are versioned and their ingestion is observable, so an answer can be traced to the source version that produced it.
Long-running work runs as durable workflows with retries. A failed step is recoverable, not silently lost.
AI usage is metered and quotas are set per organisation and enforced server-side, so cost is bounded.
AI usage, proposed changes and who confirmed them are recorded per organisation.
Ask Claude, ChatGPT or Microsoft Copilot Studio about your fleet data, signed in as yourself.
We will load a sample of your data into a sandbox organisation and let you interrogate it.