At a glance
| Client | Formula 1 team, via a UK security supplier |
|---|---|
| Sector | Professional sport, international travel |
| Service | Agentic AI Workflow Design |
| Standard | ISO 31030, travel risk management |
| Tools | Claude, ChatGPT and Perplexity for multi-agent research and verification; Canva for the designed document |
| Delivered | More than 30 advisories across the 2025 and 2026 seasons |
| Turnaround | Hours, against a two-to-three analyst-day manual estimate |
Why did the team need a faster way to produce travel advisories?
A Formula 1 team lands in a new country about every two weeks. Each trip needs a current advisory before anyone travels. The advisory must cover security, local law, health, transport, weather and emergency contacts.
Writing one by hand takes two to three analyst days. The race calendar does not wait for that. A late advisory is the same as no advisory.
The team also needed a document it could trust. A single wrong fact in a travel advisory can put people at risk. Speed on its own was not enough.
What did Insightful AI build?
We built a two-stage AI reporting system aligned to ISO 31030. ISO 31030 is the international standard for travel risk management. It sets out how organisations should assess and manage risk to people who travel for work.
Stage one drafts. An AI research process works through an agreed hierarchy of official sources. It drafts each section of the advisory under a human analyst's direction. The analyst decides what goes in and what stays out.
Stage two checks. A separate verification audit reads the finished document. It tests every claim against primary sources. Anything it cannot confirm is labelled, never removed silently. It then issues a usability verdict.
The drafting process never marks its own homework. That single rule is what makes the output safe to send.
How does the governance work?
Governance is built into the system, not added at the end. Four controls apply to every advisory:
- A human analyst signs off before delivery. AI never publishes on its own.
- Every claim carries a source. The source log ships with the document.
- Unconfirmed information is labelled as unconfirmed in the advisory itself.
- The verification audit runs as a separate process with no access to the drafting notes.
These controls map to the third stage of our Adoption Roadmap: establish governance. They sit alongside the People, Process, Principles model that shapes every build.
What were the results?
| Measure | Before | After |
|---|---|---|
| Time to produce one advisory | 2 to 3 analyst days | Hours |
| Advisories delivered | Manual, one at a time | More than 30 across 2025 and 2026 |
| Claim checking | Manual review | Every claim, every advisory, against primary sources |
| Unconfirmed information | Mixed in with confirmed facts | Labelled and separated |
Delivery continues into the 2026 season.
What we learned
Two design choices did most of the work.
Separate drafting from checking. An AI process that reviews its own output finds fewer errors. A second, independent process finds more. This costs a little time and saves a lot of trust.
Label uncertainty, do not hide it. A travel advisory that says "unconfirmed" beside a claim is more useful than one that reads smoothly. The analyst knows where to look.
Who is this for?
This approach suits any organisation that sends people abroad on a schedule:
- Sports teams and touring productions
- Charities and NGOs with staff in the field
- Security suppliers who want to scale their reporting without adding analysts
- Corporate travel and duty-of-care teams