Travel brands book holidays months in advance and compete on price every day. MMM separates the effect of media from pricing, seasonality, and external shocks so you can make smarter investment decisions.
Travel and hospitality marketing is one of the most complex measurement problems in any sector. Booking windows stretch from same-day to twelve months ahead. Pricing changes daily or even hourly. External factors like school holiday dates, airline fuel costs, and political events can swing demand by 20 percent or more. And yet most travel brands still rely on last-click digital attribution to decide where to spend their media budget. That approach is not fit for purpose.
Why last-click fails in travel
Consider the typical family holiday booking journey. It starts with inspiration, often triggered by TV, social media, or a recommendation. The family researches destinations and accommodation over several weeks. They compare prices on metasearch sites. Eventually they book, usually via a branded search or direct on the website. Last-click attributes the entire booking value to branded search, which cost almost nothing in media. The TV and social media that created the desire get zero credit.
MMM solves this by measuring the statistical relationship between media spend and bookings over time, controlling for price, seasonality, and other factors. It reveals the true contribution of each channel across the full booking window, including the delayed effects of brand campaigns that run months before the peak booking period.
The typical travel and hospitality media mix
- TV and connected TV: brand-building, particularly effective in the autumn and January peaks when consumers start planning holidays
- Out-of-home and cinema: high visual impact, supports destination and lifestyle positioning
- Paid search: brand and non-brand terms, high spend, particularly efficient for capturing late-stage bookers
- Metasearch (Google Hotels, TripAdvisor, Kayak): significant spend for accommodation brands, often treated separately from media
- Paid social: video formats perform well for travel inspiration, particularly on Instagram and TikTok
- Email: highly effective for past customers and loyalty members, low cost, often under-modelled
- Affiliate and cashback: volume driver but low-margin and prone to attribution overlap
Seasonality and booking windows
Travel has two layers of seasonality that overlap in ways that make measurement tricky. The first is media seasonality: when brands advertise. January is the biggest media investment month for most travel brands (the January sale), followed by autumn. The second is booking seasonality: when customers actually book. School holidays create predictable booking spikes, but the booking dates can be months before the travel date. A model built on booking date rather than travel date will look very different from one built on travel date, and neither is inherently wrong. Choose the metric that best matches your planning horizon.
Define your outcome metric carefully before you start. Bookings made, travel dates, revenue recognised, and deposits taken are all different numbers. Pick the one that best reflects how you make budget decisions and stick with it.
Key data sources for travel MMM
For travel MMM, you need weekly booking volumes (or revenue), media spend by channel and week, average selling price or revenue per booking, promotional activity (price reductions, free extras, early booking discounts), school holiday calendar for your target markets, competitor pricing index where available, and external demand indicators such as consumer confidence indices.
Capacity constraints are often forgotten. If you sell out a hotel or a flight route, your bookings plateau regardless of how well media is working. The model needs a capacity or availability variable to prevent it from wrongly concluding that media became less effective when the product simply sold out.
Common measurement mistakes in travel
Modelling without controlling for price changes is the most common error. Travel pricing is dynamic, and a 10 percent price cut will drive more bookings than almost any media campaign at equivalent cost. If the model does not include a price variable, it will confuse the effect of price reductions with the effect of media, making channels look far more or less effective than they are.
A second mistake is not accounting for the impact of earned media and PR. Travel brands often generate significant coverage through editorial press, influencer trips, and award wins. These create measurable spikes in bookings that have nothing to do with paid media. If the model does not include a proxy for earned media activity (a monthly PR coverage score, for example), those spikes get attributed to whatever paid channels happened to be running at the same time.
Planning with MMM in a volatile market
The travel sector is more exposed to external shocks than most. A pandemic, a geopolitical event, or an airline collapse can wipe out demand overnight. MMM models built on pre-2020 data were largely invalidated by the pandemic. The lesson is not that MMM does not work for travel, but that models need to be refreshed regularly and that your planning scenarios should include stress tests for demand shock scenarios. A well-built model with frequent refreshes is still far more robust than gut instinct or last-click data.
How do we model the January sale period given that it is so different from the rest of the year?
Flag it as a promotional event variable in the model. If you run price promotions in January every year, the model learns that pattern and separates the promotional effect from the media effect. This gives you a cleaner read on what media is actually contributing on top of the promotional uplift.
Can MMM work for a hotel group with properties in different markets?
Yes, but it works best when you model each market separately or use a hierarchical model that allows for market-level differences. A hotel in London and a hotel in the Cotswolds have different customer profiles, different booking windows, and different competitive landscapes. A single global model will average out those differences and give you results that are accurate for no individual property.
We shifted heavily to digital during the pandemic and have not run TV since. Does that affect our ability to model TV effectiveness?
Yes, it does. You need variation in spend levels for the model to estimate effectiveness reliably. If you have not run TV in two or more years, the model cannot estimate current TV effectiveness from your data alone. The solution is either to use a small test market to reintroduce TV and measure the impact, or to incorporate industry benchmarks from comparable brands as a prior in a Bayesian model.