Travel marketing budgets, planned months before the trip.

The sections below cover the outcome choice, booking-season timing, shocks and the pandemic years, source markets, online travel agencies and metasearch, and the controls worth having.

Marketing moves bookings. Revenue arrives with the trip, often months later. A model on travel dates struggles to connect a summer departure to the January campaign that sold it; a model on bookings sees the link directly, but bookings can be cancelled and their value differs.

It is often worth taking both in and building the first model on both, then choosing per market. Some markets book early and steadily, others late and in bursts, and the best outcome can differ between them. Booking value, rather than booking count, is the better outcome when trips vary a lot in price.

The booking season for summer often starts in January. For winter it starts in autumn. A budget that follows the travel season spends the most when the decisions have already been made.

Within the booking season, the same spend does not buy the same result every week. When demand is high, marketing converts more easily. The model separates seasonal demand from the marketing effect, so a strong booking week is not read as a strong channel, and shows in which weeks the next part of the budget works hardest. Where the data supports it, the channels' response itself can change through the year. See time-varying saturation.

Travel has more shocks than most industries: strikes, ash clouds, storms, outages, and the pandemic years when demand all but stopped. None of these were caused by marketing, and none can be predicted.

Known shocks are flagged in the history, so the model does not learn from those weeks and does not forecast them forward. For a long disruption such as the pandemic, we try the model with and without the period and keep the version that fits better. Throwing away data has a cost, and so does teaching the model from a world that will not repeat.

A hotel chain or tour operator selling to travellers from several countries is running several marketing businesses at once. Each source market books at different times, through different channels, with different media costs. One model per source market, with the same setup in each, lets the budget be compared across countries on marginal return.

We build the largest market first. What it learns about each channel becomes the starting point for the next, scaled to its size, and that market's own data then corrects it. Destinations and properties usually stay dimensions in the data rather than separate models.

For hotels and many tour operators, online travel agencies bring a large share of bookings and take a commission on each. Because that cost follows the bookings, it moves with demand whatever the agency did, and a model that treats it as an ordinary media channel credits it with bookings it did not cause. Commission-based channels stay in the model at their historical cost, so the other channels are estimated correctly, and are locked in the recommendations.

Metasearch bought per click is different. You decide the bids and the spend, so it behaves like media and is modeled as a channel, with its own saturation. That makes it possible to ask the question that matters for the margin: how much direct-booking demand your own media creates, and what it costs compared with the commission you would otherwise pay.

Travel prices move constantly, and capacity runs out. We keep controls few, because every extra variable takes explanatory power away from the media and anything used to explain the past has to be known when the model forecasts. Large, planned price campaigns are marked. Periods when capacity was sold out are recorded on the timeline, so results are read with the context, rather than added as a variable the model can't know ahead.

Competitors' activity can be included when it clearly improves the model, but most of the time it is better left to the baseline and the trend. The model's job is to explain the effect of your own spend.

Odins is a managed marketing mix modeling platform. We connect the ad platforms and metasearch, set up the pipeline for TV and other offline media with you or your media agency, structure your bookings by source market, build and maintain the models, and review every recommendation before it reaches you. Data flows in continuously and the model is retrained once a month, while scenarios and forecasts are yours to run whenever you need them, in the platform or from your own AI assistant.

Moving budget toward channels with headroom typically gives 5 to 15% more effect from the budget you already have. See the managed service.