{"id":15879,"date":"2026-06-23T13:17:00","date_gmt":"2026-06-23T12:17:00","guid":{"rendered":"https:\/\/coreanalytics.stageweb.builders\/2026\/06\/23\/the-5-data-errors-to-avoid-when-conducting-an-mmm\/"},"modified":"2026-10-07T11:53:18","modified_gmt":"2026-10-07T10:53:18","slug":"the-5-data-errors-to-avoid-when-conducting-an-mmm","status":"publish","type":"post","link":"https:\/\/coreanalytics.it\/en\/2026\/06\/23\/the-5-data-errors-to-avoid-when-conducting-an-mmm\/","title":{"rendered":"The 5 “Data” Errors to Avoid When Conducting an MMM"},"content":{"rendered":"\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

An econometric model is only as good as the data that feeds it. After more than 15 years of modeling and hundreds of completed projects, we have identified five recurring errors in data preparation that, if not corrected, compromise the quality of the analysis even before the model is estimated. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tError #1: Inconsistent Temporal Granularity\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Sales data arrives daily, TV data weekly, and digital spend monthly. If granularity is not harmonized before building the model, systematic biases are introduced. The rule: the maximum granularity of the model is that of the coarsest data point. If you only have weekly TV data, the model will operate on a weekly basis. It is preferable to have consistent data at a lower frequency rather than mixed data at a higher apparent frequency. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tError #2: Confusing Investment with Media Pressure\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

The cost of a GRP or a CPM varies over time. If you input average spending in dollars into the model instead of pressure units (GRPs, impressions, clicks), you are mixing two things: how much you spent and how much you were seen. An increase in spending might reflect rising costs, not an increase in visibility. The model must work on exposure metrics, not cost. Cost should only be used to calculate the final ROAS derived from the incremental lift calculated by the MMM. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tError #3: Ignoring Exogenous Factors\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Ice cream sales rise in the summer. Travel bookings drop during an economic crisis or conflict. If the model does not include these exogenous variables (weather, seasonality, macro indicators, special events), it will attribute effects to the media that actually belong to the context. At Core Analytics, we use a proprietary library of exogenous variables calibrated for the Italian market. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tError #4: Digital Data from a Single Source\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Relying on data exported from a single platform (Google Ads only or Meta only) creates silos. Conversions attributed by each platform overlap: the same sale can be counted two or three times. The model needs a single source of truth for conversions (typically a proprietary tracking system or CRM) and investment and pressure data for each channel separately. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tError #5: Historical Series That Are Too Short\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

A robust model needs sufficient variance in the data. If you have always invested the same proportions across the same channels, the model struggles to distinguish their effects. The rule of thumb: at least 2 years of data, ideally 3. If the series is shorter, Bayesian approaches that incorporate informative priors from previous studies or industry benchmarks become essential to compensate for limited sample variance. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t

\n\t\t\t\t\tThe Bottom Line\t\t\t<\/h2>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Data integration is not a technical step to be delegated and forgotten. It is the foundation upon which the credibility of every analysis rests. Investing time in data quality before building the model is not a cost: it is the single action with the highest return on the quality of results. <\/span><\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t

\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\n\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t

\n\t\tDo You Want to Learn More About Different Attribution Models?\t<\/h3>\n\t\t\t\t\t\t

Fill out the form below to download <\/span>our guide to exploring the strengths and weaknesses of the various models available on the market.<\/p><\/div>\n\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t
\n\t\t\t\t\n
\n

<\/p>