When an AI Concierge Should Stay Quiet
An in-trip assistant is judged as much by the messages it doesn't send as by the ones it does.
Field Notes · Operations & Automation
Most disruptions announce themselves hours in advance. The operators who handle them well have already decided what they'll do.

New Mexico’s summer monsoon brings afternoon thunderstorms. Anyone who has run trips in the high country in July knows the rhythm: clear mornings, clouds building over the peaks after midday, and lightning on exposed ridges by mid-afternoon. It’s about as predictable as disruption gets, and it still catches operators out every season.
What follows is a composite drawn from corporate offsite work near Santa Fe. The times and numbers are typical of what we see. The people are left anonymous.
Thirty-eight people are on a three-day offsite in late July. Day two looks like this:
Day three has a free morning, a closing lunch at noon and departures from 2:00 PM.
6:15 PM, the day before. The forecast gives a 40% chance of thunderstorms over high terrain after 1:00 PM tomorrow. The playbook for afternoon storms on exposed terrain moves to watch. An exposure check flags two items: the ridge hike and the open-air dinner. Travelers hear nothing. Two suppliers get pre-alerts. One is the caterer, whose contract includes a covered courtyard alternate on six hours’ notice. The other is the shuttle company, which needs three hours’ notice for changes.
6:00 AM. The updated forecast is 70%, with cells building from about 1:40 PM. That crosses the act threshold. The routing engine produces four options, each with its cost, the exposure that remains, and the time the option stops being available:
| Option | What changes | Added cost | Exposure left | Available until |
|---|---|---|---|---|
| A. Go early | Hike 11:00–1:00, workshop cut short | Shuttle change fee | Ends 40 minutes before forecast onset, no margin | 7:10 AM |
| B. Swap days | Hike moves to tomorrow’s free morning, 7:30–10:00; this afternoon becomes a clay studio session in two groups; dinner moves to the covered courtyard | Courtyard and studio fees | None | 9:30 AM |
| C. Replace | Hike cancelled; studio sessions and courtyard dinner | Studio fee, forfeited guide fee | None | 9:30 AM |
| D. Hold | Keep the plan; the guide decides at the trailhead | None yet | High | Always |
6:20 AM. The operations lead reads the options on her phone. B is clearly best. It leaves no exposure at a modest cost, and the only thing lost is a free morning that half the group would have spent on email. She chooses it at 6:24. At 6:40 she calls the client’s event lead, who agrees within two minutes because the options arrive already costed.
6:45 AM. Change requests go out. The playbook drafted them and the operations lead approved them in one step. The caterer moves dinner to the courtyard. The shuttle company cancels the 1:15 ridge run and adds a 6:50 AM pickup tomorrow. The studio confirms sessions at 2:00 for 24 people and 3:45 for 14. The guide moves to tomorrow at 7:30. All four confirm by 8:05.
8:15 AM. One message goes to all 38 attendees before the workshop starts. The client’s event lead gets it ten minutes earlier.
1:15 PM. The shuttles don’t leave for the ridge.
1:50 PM. Thunder rolls over the high country.
7:30 AM the next day. The group is on the ridge under a clear sky.
None of that took heroics. It took decisions made in advance: what each forecast number means, which alternates are pre-contracted, who decides, and by when. That’s what a playbook is.
Every playbook we write has the same seven parts, and each fits on one page:
These are starting points, not standards. Tune them each season using your closeout logs.
| Disruption | Signal | Watch | Act | Typical first move |
|---|---|---|---|---|
| Afternoon thunderstorms | Hourly forecast for the activity’s location and window | 30% or more inside the window | 60% or more, or lightning nearby | Move exposed activities to the morning, or swap days |
| Wildfire smoke | Air-quality forecast | AQI above 100 | AQI above 150 during planned exertion | Lower the exertion, move indoors, or change areas |
| Road or trail closure | Agency notices, supplier messages | Closure announced for the date | Closure confirmed | Re-route and brief drivers |
| Arrival flight delay | Flight status | Delay over 45 minutes | Delay over 2 hours, or a cancellation | Re-time the transfer and protect the first-night dinner |
| Supplier failure | Missing confirmation | Unconfirmed 48 hours out | Unconfirmed 24 hours out | Bring in the backup supplier |
| Extreme heat | Forecast high | 95°F or higher | 100°F or higher | Move exertion before 10:00 AM |
Every option has an expiry, set by the notice its suppliers need. Your decision deadline is the earliest expiry among the parts of the option you’d most like to keep, minus a buffer. For option B:
| Part of option B | Notice needed | Latest change |
|---|---|---|
| Courtyard dinner | Six hours before the 3:30 PM setup | 9:30 AM |
| Clay studio sessions | Tentative block released by the studio | 10:00 AM |
| Cancel the 1:15 PM shuttle run | Three hours | 10:15 AM |
| Guide moves to tomorrow | Same-day agreement | Flexible |
With a 15-minute buffer, the deadline was 9:15 AM. So why decide at 6:24?
Because the last responsible moment is a limit, not a goal. Waiting only pays if the information is likely to improve. At 6:00 AM, with a 70% forecast and a clear onset time, the 9:00 update was unlikely to change the choice, and every hour of waiting cut into the attendees’ notice. Decide when the information stops improving or the deadline arrives, whichever comes first.
Option A teaches a second lesson. It was gone by 7:10 AM. Some options expire early, so if you want early-start alternatives, the watch phase has to begin the evening before.
Option B worked because the alternates were agreed in April, not improvised in July. Useful clauses for the monsoon season include:
Store each clause with the supplier record, so that a change request cites the clause it relies on. That is what lets automated supplier requests go out in minutes instead of turning into phone negotiations.
During the watch phase, travelers hear nothing. When the plan changes, they hear it once: the new plan first, the reason in a sentence, and what they need to do.
Change to this afternoon. Your new plan is below.
The ridge hike moves to tomorrow at 7:30 AM (shuttle from the lobby at 6:50). Thunderstorms are forecast over the high trails from about 1:40 PM today, and we don’t take groups onto exposed terrain in that window.
This afternoon: clay studio sessions at 2:00 PM (Group A) or 3:45 PM (Group B). Your group is printed on your badge. Dinner is at 6:00 PM in the covered courtyard.
Nothing to do unless you’d rather skip tomorrow’s hike. Reply “skip” and we’ll arrange a late breakfast instead.
Then come the follow-up questions: Can I switch to Group B? Is the courtyard far? This is where an in-trip concierge earns its keep. It answers from the new plan and hands anything unusual to a person. Three attendees replied “skip,” and each got a late breakfast without anyone picking up a phone.
| Automate | Keep human |
|---|---|
| Watching forecasts and feeds against every active itinerary | Choosing the option |
| Finding exposed activities and the travelers affected | Safety calls on the ground; the guide can always turn back |
| Generating and costing options with a routing engine | The conversation with the client |
| Drafting messages and supplier requests | Exceptions: medical needs, anyone upset |
| Tracking confirmations and chasing silence | Changing the thresholds |
The left column is where the minutes go. The right column is where judgment goes. Keeping them apart is what makes re-planning, rather than rebooking, realistic at 6:00 in the morning. That division of labor is the core of Nova’s Disruption Response Automation work. It matters most for corporate travel and events, where one change affects dozens of people and a client is watching. The offsite planning case study shows the same pattern across four live events.
The aim isn’t to stop the weather getting a vote. It’s to make sure that when it does, the shuttle sits in the lot and your group already knows where it’s going instead.
Questions
Use two thresholds. A watch threshold, typically 18 to 24 hours ahead for weather, starts internal preparation and supplier pre-alerts. An act threshold, tied to the deadlines of your options, triggers the re-plan. The right lead times come from your suppliers' notice periods, not from the forecast.
Usually not. Keep the watch phase internal and tell travelers once, when the plan has actually changed, with the new plan first. The exception is anything travelers must prepare for themselves, such as packing rain gear or setting an earlier alarm.
Keep reading
An in-trip assistant is judged as much by the messages it doesn't send as by the ones it does.
A group doesn't have a preference. It has several, plus a memory of who gave in last time. Plan for both.
Every itinerary that falls apart on day two was already infeasible on day zero. Here is how to see the math hiding inside an ordinary trip plan.
Bring the problem this article describes to a 45-minute discovery call. We will tell you honestly whether AI is the right tool for it yet.