Demand forecasts
Bookings and inquiries by product, date and channel, with booking-pace curves against the same point last year. You see a slow departure eight weeks out, not three days before.
Platform · Signals
Forecasts that come with a next step
Signals predicts demand, cancellations and price sensitivity from your own booking history, then turns each prediction into something a person can act on this week: a departure to promote, a guest to call, a quiet week to package.
What it does
Most tourism forecasting is a chart that says next October will look like last October. It’s rarely wrong enough to notice and rarely useful enough to change a decision.
Signals is built backwards from decisions. Which departures need a push, and when? Which bookings are likely to cancel, and is there a date we could move them to instead? Which shoulder weeks have demand hiding in them? Every signal arrives with three things: a range rather than a single number, the drivers behind it, and a suggested action a person can approve or ignore.
Four kinds of signal
Cancellation risk
A risk score for each booking, built from lead time, deposit status, party shape, channel and communication patterns. It comes with the top reasons, so the call you make is the right call.
Price sensitivity
How demand for each product responds to price by season and lead time, so you can see where a discount buys bookings and where it only gives away margin.
Shoulder-season shaping
Who already books your quiet weeks, what they value, and which packages would bring more of them. Results feed Itinerary Studio and your campaigns.
How it works
How a signal gets made
- 01
Gather what you already have
Bookings, cancellations, lead times, inquiries, deposits and channel data, plus on-site and assistant search demand where you have it. Event calendars and weather history fill in the context.
- 02
Model each question on its own terms
Hierarchical forecasts for demand, so thin products borrow strength from their category. A calibrated classifier for cancellation risk. Elasticity estimates for price. No single model pretends to answer everything.
- 03
Backtest on your seasons
Before anything goes live, we replay held-out seasons and show you the error. You decide whether a signal is good enough to trust.
- 04
Turn it into an action
Each signal carries a suggested move: release an allotment, send a reassurance note, offer a date change, build a package. A person approves it or doesn't.
- 05
Watch for drift
Outcomes flow back in. When accuracy slips because of a new competitor, a closed road or a shift in who travels, we see it and retrain. That's part of Managed AI Operations.
Shoulder-season shaping
Grow the quiet season instead of discounting it
Quiet weeks are rarely empty of demand. They're short on reasons. Signals looks at who already books them, how far ahead, from where and in what party shape, and what those travelers rated highly. Then it suggests packages built on what the season genuinely offers: cooler hiking, emptier trails, softer lodge rates, clear high-desert nights.
- Finds the segments that already prefer your shoulder dates, and the channels that reach more of them.
- Suggests packages and routes that suit the season, checked by Route Engine for daylight hours and seasonal closures.
- Measures the effect against matched weeks, so you know whether the package worked or the weather did.
- Helps destinations spread visitors away from the busiest weekends. See shoulder-season demand shaping.
Measured, not promised
From the desert lodge pilot
A 10-week pilot with a small collection of desert lodges, measured against the same weeks the prior year. Read the case study.
- Late cancellations flagged high-risk at least 14 days before they happened
- 7 in 10
- Pilot weeks 5–10
- Late-cancelled room-nights resold before arrival
- 24% → 57%
- Versus the same weeks the prior year
- Midweek occupancy in the pilot's shoulder weeks
- +7 pts
- Versus the same weeks the prior year; not adjusted for market demand
Pilot measurements from one engagement and one season. Forecast quality depends on the depth and cleanliness of your history, which is why we show backtest error before go-live.
Checklist
Guardrails, on by default
Prediction in travel touches sensitive ground. These defaults come from our Responsible AI & Traveler Data Governance practice, and they aren't optional extras.
0 of 8 ready
We used to discount the summer midweeks because they felt empty. Signals showed us who was already booking them, and why. We built two packages around those guests and stopped giving the rooms away.
Questions
Frequently asked questions
How much history do we need?
Two full seasons of bookings and cancellations is a comfortable start, and one can be enough for cancellation risk. Products with thin history borrow from similar products. After the backtest we tell you plainly what's reliable and what isn't yet.
Does Signals set our prices?
No. It shows how demand responds to price and suggests where to test. Price changes go through a person, and Signals never personalizes a price to an individual traveler.
What about events that have never happened before?
Models learn from the past, so a new festival or a long road closure will surprise them. You can add known future events by hand, and drift monitoring flags it when reality leaves the forecast's range.
Can it read demand from search and AI assistants?
Where you have the data, yes: on-site search, planner queries, and requests that arrive through Agent Gateway. Questions asked before anyone books are often the earliest signal you'll get.
Is our data pooled with other customers'?
No. Your models train on your data, for your use. We add general public context such as holiday calendars and weather history, but your bookings are never pooled with anyone else's.
Related
Keep exploring
- Use casesShoulder-Season Demand ShapingShoulder-season demand shaping: predictive signals spot soft weeks early and suggest packages worth building, long before a discount looks like the only option.
- WorkDesert lodge collectionHow a small collection of desert lodges flagged 7 in 10 late cancellations two weeks early and filled quiet shoulder-season midweeks with packages.
- SolutionsHotels, Resorts & LodgesAI for boutique hotels and lodges: cancellation risk flagged early, packages built from real guest behavior, and a concierge that knows today's conditions.
- Field Notes · Data & AnalyticsPredicting Cancellations Before They HappenA practical walk-through of cancellation prediction for lodges and tour operators: defining the label, choosing features, calibrating scores and acting on them.
- Field Notes · Data & AnalyticsShaping Shoulder-Season Demand With Predictive SignalsA countdown playbook for shoulder-season demand: forecast with an interval at 120 days, stage packages and targeting before price, and measure true lift.
Backtest it on your own history
Send two seasons of booking and cancellation exports. We'll replay a past season and show you what Signals would have flagged, and how often it would have been wrong.