The question starts with the budget
When several platforms claim a booking, adding their conversions does not establish what each channel contributed. Attribution distributes credit across observed interactions according to its rules. A Marketing Mix Model, or MMM, uses aggregate data to estimate the relationship between marketing activity and a commercial outcome while considering other factors. These are different perspectives that can support different questions.
For a hotel, the relevant decision might be how to allocate the next investment, which channels appear close to saturation or how to interpret performance as rates and demand change. Before selecting a tool, define the decision and the outcome to be studied: bookings, revenue or another measure agreed with the business.
What a usable dataset requires
The documentation for Meridian, Google’s MMM framework, includes a KPI, channel spend and exposure, together with control variables. Their periods and dimensions need to be compatible. For a hotel application, we recommend reviewing how rates, promotions, inventory changes and other potentially relevant demand conditions were recorded. Each variable needs a justified role; collecting more columns does not automatically produce a better model.
It is also useful to reconstruct the decision calendar: campaign pauses, budget changes, openings, closures and commercial amendments. If the data mixes properties, currencies or accounting criteria, those differences need to be resolved or documented before interpreting results. An apparently complete series can conceal definition changes that alter its meaning.
More months do not automatically mean more certainty
Google describes its guidance on data volume as approximate. Sufficiency depends on the available information, variation and model complexity; evaluating credible intervals helps assess uncertainty. We therefore do not promise that a fixed number of months will make MMM viable for every hotel.
When all channels move together, separating their effects can be difficult. Trying to estimate too many components from limited information creates a similar problem. A feasibility review needs to be able to conclude that a simpler model, better information or additional experiments are required. Choosing not to model yet can be a useful outcome and avoids presenting precision unsupported by the evidence.
A checklist before commissioning a model
We recommend requesting a feasibility assessment that addresses these questions and records what remains unresolved.
- Which commercial decision will the model support, and what will its KPI be?
- Do spend, exposure and outcome data use consistent periods and definitions?
- Is there enough variation to distinguish the channels being studied?
- Which demand, pricing and operational factors need consideration?
- How will the model, its assumptions and its uncertainty be evaluated?
- What further information or experiments could help test the findings?
From a model to an investment decision
A budget scenario is an estimate conditional on the data and model, rather than a guarantee of future bookings. Its interpretation should include ranges, operational constraints and assumptions. Findings then need to be compared with commercial reality and updated as the business learns. DRS Audit can assess readiness and define next steps. Building, calibrating and operating an advanced model is agreed as a specific scope; it is not assumed to be included in every audit.
Further reading
These sources expand on the article’s concepts. Applying them to your hotel requires a review of your configuration, data and business objectives.
DRS Audit
Turn your questions into a working plan.
We define objectives, sources, responsibilities and deliverables around your hotel’s needs.
Discuss this solution