{"id":3766,"date":"2026-09-04T00:00:00","date_gmt":"2026-09-03T22:00:00","guid":{"rendered":"https:\/\/humes.pl\/marketing-mix-modeling-how-econometrics-measures-the-impact-of-media-on-sales\/"},"modified":"2026-09-24T15:23:55","modified_gmt":"2026-09-24T13:23:55","slug":"marketing-mix-modeling-how-econometrics-measures-the-impact-of-media-on-sales","status":"publish","type":"post","link":"https:\/\/humes.pl\/en\/marketing-mix-modeling-how-econometrics-measures-the-impact-of-media-on-sales\/","title":{"rendered":"Marketing mix modeling: how econometrics measures the impact of media on sales"},"content":{"rendered":"<p>A media budget spread across six channels, sales growth of more than ten percent, and a question that no one in the organization can answer directly: how much of that growth was generated by advertising, and how much by price reductions, seasonality, and distribution? <strong>Marketing mix modeling<\/strong> answers this question using an econometric method &#8211; it estimates the sales components associated with individual media channels, price variables, and external factors. Below is an overview of how such a model is built, what data it requires, and where it most often breaks down.<\/p>\n<h2>How does marketing mix modeling separate the impact of media, price, and seasonality?<\/h2>\n<p><a href=\"https:\/\/humes.pl\/en\/glossary\/marketing-mix-modeling-mmm\/\">Marketing mix modeling<\/a> is based on a simple statistical logic: sales in a given period are a function of multiple simultaneous stimuli, and the model&#8217;s task is to estimate what share of the observed variation is associated with each of them. Technically, it is a regression model applied to a time series &#8211; most often weekly &#8211; in which the dependent variable is sales volume or value, while the explanatory variables include media spend, reach, own and competitor prices, promotional activity, weighted distribution, weather, holidays, and long-term trends.<\/p>\n<p>The key difference from simple correlation is that an <strong>econometric model<\/strong> accounts for multiple effects simultaneously. If a television campaign started in the same week as a price promotion in retail chains, a simple analysis would attribute the entire sales increase to whichever factor is currently being examined. A model that includes both variables can separate their effects based on historical variation in the data. However, this does not eliminate the problem of strong multicollinearity, nor does it in itself provide causal evidence. This is precisely why MMM quality depends not on the number of channels, but on whether the historical data contain sufficient variation to distinguish one stimulus from another.<\/p>\n<p>MMM often uses two mechanisms that go beyond simple linear regression. The first is <strong>adstock<\/strong>, meaning the delayed and decaying impact of advertising exposure &#8211; a campaign aired in week ten may also affect weeks eleven and twelve, with decreasing intensity. The second is <strong>saturation<\/strong>: the relationship between investment and effect may not be linear, as additional GRPs generate progressively smaller increases in sales. Without transformations that account for these phenomena, the model may incorrectly estimate the optimal level of investment, for example by assuming that doubling the budget doubles the effect.<\/p>\n<p>The third element is separating baseline sales from incremental sales. The baseline is the estimated level of sales that would have occurred without the marketing activity being analyzed, taking into account the other factors in the model, such as brand strength, distribution, purchasing habits, and seasonality. The increment is the estimated uplift associated with marketing activities. This decomposition is MMM&#8217;s main analytical output and the starting point for calculating the ROI of individual channels.<\/p>\n<h2>How do you build a marketing mix model step by step, and what data does it require?<\/h2>\n<p>An MMM project follows a repeatable sequence, and most of the time is spent not on modeling but on data preparation. A typical process includes several stages that are worth understanding before commissioning a study.<\/p>\n<ul>\n<li><strong>Defining the dependent variable and granularity.<\/strong> Sales in units or value terms, weekly or monthly, for the entire market or broken down by region and distribution channel. Weekly granularity provides more observations and better captures short campaigns.<\/li>\n<li><strong>Collecting media data.<\/strong> Net spend, GRPs, or reach, broken down by channel and preferably by campaign. Cost alone can be misleading because changes in media buying prices distort the investment-effect relationship; therefore, exposure measures are preferred whenever possible.<\/li>\n<li><strong>Control variables.<\/strong> Shelf price, a price index relative to competitors, the share of promotional sales, weighted distribution, competitors&#8217; media activity, holiday calendars, temperature, and changes in the product portfolio.<\/li>\n<li><strong>Cleaning the time series.<\/strong> Aligning calendars, handling missing data, identifying outliers and one-off events that need to be included in the model as dummy variables rather than ignored.<\/li>\n<li><strong>Specification and estimation.<\/strong> Selecting the functional form, adstock and saturation parameters, testing collinearity between channels, and checking the signs of coefficients for economic plausibility.<\/li>\n<li><strong>Validation.<\/strong> Comparing model fit on training and test data, analyzing residuals, testing coefficient stability across subsamples, and checking whether the model reproduces known market events.<\/li>\n<li><strong>Decomposition and simulations.<\/strong> Breaking sales down into components and calculating response curves and allocation scenarios.<\/li>\n<\/ul>\n<p><\/br><\/p>\n<p>The data horizon required for MMM depends on data granularity, the number of variables, the length of the purchase cycle, and variation in marketing activity. In practice, at least two to three years of weekly observations are often used; a shorter series may mean too few observations relative to the number of variables and may make it difficult to separate seasonality from trend. If a brand always runs campaigns in the same months, the model will face a fundamental problem distinguishing the effect of advertising from the effect of seasonality, because both variables move in similar ways.<\/p>\n<p>As Hume&#8217;s Institute experts point out, a marketing mix model is only as good as its input data &#8211; without reliable control of seasonality, prices, and promotional activity, estimates of media effects may be biased because the model may attribute variation actually generated by a discount or a seasonal peak to advertising. In practice, this means that obtaining price and promotional data from the sales department is just as critical as access to media plans.<\/p>\n<p>It is worth distinguishing between two functions of MMM. The first is historical diagnosis: assessing how past campaigns performed and what their estimated return was. The second is simulation: using response curves to model alternative budget allocations. The second function is most reliable within investment levels close to historical ones &#8211; extrapolation far beyond the observed range involves high uncertainty and does not constitute reliable <strong>advertising effectiveness research<\/strong>.<\/p>\n<h2>How does MMM differ from digital attribution, and where does the method fall short?<\/h2>\n<p>The most common misunderstanding concerns the relationship between marketing mix modeling and attribution in digital channels. These are two different measurement perspectives, not competing versions of the same analysis. Typical digital attribution operates at the individual user and conversion-path level and primarily covers channels for which identifiers and event data are available. It usually does not directly measure the impact of television, OOH, radio, or changes in shelf price. MMM operates at an aggregated level and can include all available channels and control variables, but it typically does not go down to the level of individual creatives or audience segments.<\/p>\n<p>The practical consequence is that adding conversions from digital attribution and the increment from MMM for the same business objective may lead to double counting. Performance channels in last-click attribution usually show high effectiveness, while their contribution in an econometric model may be lower because some of those conversions are attributed to the baseline or to demand-building campaigns. This discrepancy does not necessarily indicate an error in either method, but may result from differences in the definition of the measured phenomenon and the assumptions adopted.<\/p>\n<p>Below is a list of pitfalls that most often undermine the credibility of <strong>MMM<\/strong> and that are worth discussing with the provider before starting a project.<\/p>\n<ul>\n<li><strong>Channel collinearity.<\/strong> When television and online campaigns always start together and in the same proportion, the model cannot reliably separate their contributions. The coefficients become unstable, and a minor change in specification can reverse the result.<\/li>\n<li><strong>Omitted variables.<\/strong> Missing data on distribution, competitor activity, or packaging changes may cause the effects of these factors to be attributed to variables included in the model, often media.<\/li>\n<li><strong>Confusing trend with seasonality.<\/strong> Inappropriate time-series decomposition can cause long-term category growth to be recorded as a campaign effect. It is worth understanding how <a href=\"https:\/\/humes.pl\/en\/seasonality-and-time-series-how-to-seasonally-adjust-data-to-distinguish-trends-from-fluctuations\/\">seasonal data are cleaned<\/a> before modeling.<\/li>\n<li><strong>Model overfitting.<\/strong> High fit achieved by adding more variables and dummy variables may look good in a report, but it does not carry over to future periods. Out-of-sample validation is one of the key tests for this risk.<\/li>\n<li><strong>Arbitrary adstock parameters.<\/strong> Adopting a decay rate &#8220;from the literature&#8221; without checking whether it is consistent with the data and knowledge of the channel can materially change ROI results, especially for channels with long-lasting effects.<\/li>\n<li><strong>Treating the result as a measure of creative quality.<\/strong> MMM can account for differences between campaigns, but it usually does not reliably isolate the strength of the message itself from reach, budget, and other factors. Evaluating the advertisement itself requires a different methodology &#8211; <a href=\"https:\/\/humes.pl\/en\/advertising-pre-testing-and-post-testing-how-to-measure-communication-effectiveness-before-media-spend-begins\/\">ad pre-testing and post-testing<\/a> serve this purpose.<\/li>\n<\/ul>\n<p><\/br><\/p>\n<p>A separate limitation concerns causality. A marketing mix model often relies on observational data, so its results depend on the assumptions adopted, the quality of control variables, and the identification of effects. Stronger causal conclusions require an appropriate research design or additional validation. This is why MMM is increasingly combined with geographic experiments or media holdout tests &#8211; the results of an experiment can then serve as a calibration point for the model&#8217;s coefficients.<\/p>\n<h2>When does marketing mix modeling make sense, and when is it worth choosing another tool?<\/h2>\n<p>Not every organization has the conditions in which <strong>marketing mix modeling<\/strong> will deliver stable results. Below are criteria worth checking before deciding to undertake a project.<\/p>\n<ul>\n<li><strong>Data history.<\/strong> Usually at least two years of consistent sales and media data at the same level of granularity; however, the required period depends on data frequency and model complexity.<\/li>\n<li><strong>Variation in investment.<\/strong> Budgets and channel allocation must differ between periods. A fixed allocation that has remained unchanged for years gives the model no basis for estimation.<\/li>\n<li><strong>Budget scale.<\/strong> The cost of an MMM project should remain reasonably proportionate to the spending covered by the analysis.<\/li>\n<li><strong>Access to price and promotional data.<\/strong> Without it, the assessment of <strong>advertising effectiveness<\/strong> may be distorted and misleading.<\/li>\n<li><strong>Portfolio and distribution stability.<\/strong> Major changes in the product range or entry into new retail chains require additional control variables; otherwise, they will distort the decomposition.<\/li>\n<\/ul>\n<p><\/br><\/p>\n<p>When one of these conditions is not met, alternatives may include a field experiment in selected regions, incrementality analysis in a chosen channel, or a simpler descriptive model focused on a single variable. In Hume&#8217;s Institute projects, it has been observed that for brands with a short media history, designing a controlled test is more useful than attempting to extract conclusions from an insufficient time series.<\/p>\n<h2>Frequently asked questions<\/h2>\n<h3>What data does marketing mix modeling require?<\/h3>\n<p>The foundation is sales data, often on a weekly basis, and media data broken down by channel &#8211; preferably in the form of exposure measures such as GRPs or reach, supplemented by spending data. In practice, a period of at least two years is desirable, although the required horizon depends on the granularity and complexity of the model. Available control variables are also important: own price and, where relevant to the category, competitor price, the share of promotional sales, weighted distribution, holiday calendars, and competitors&#8217; media activity. The more complete the set of relevant control variables, the lower the risk of attributing effects from other sources to media.<\/p>\n<h3>How does MMM differ from attribution in digital channels?<\/h3>\n<p>Digital attribution usually operates at the level of individual users and conversion paths, primarily covering channels measurable through identifiers. MMM works with aggregated data and can include all channels, including television and OOH, as well as non-media variables such as price or seasonality. The results of both methods should not be automatically added together, because they may concern the same effect while being calculated according to different definitions and assumptions.<\/p>\n<h3>How often should an MMM model be updated?<\/h3>\n<p>A full re-estimation is often performed once a year, with quarterly data updates and verification of whether the coefficients remain stable. An accelerated update is justified after a significant change in the media mix, product portfolio, pricing policy, or following the entry of a new market player. Re-estimating too frequently based on small additions of data increases the risk of random coefficient fluctuations without a genuine improvement in model quality.<\/p>\n<h2>Ask about a marketing mix model for your brand<\/h2>\n<p>If you are considering measuring the impact of media on sales using hard data, <strong><a href=\"https:\/\/humes.pl\/en\/contact\/\">ask about a marketing mix model for your brand<\/a><\/strong> &#8211; Hume&#8217;s Institute experts will assess the availability and quality of input data and propose a scope of analysis tailored to the specifics of your category.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>How much of your sales does advertising really drive? Marketing mix modeling separates the impact of media, price and seasonality. We explain what data the model needs.<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":""},"categories":[912],"tags":[],"slowa_kluczowe":[],"class_list":["post-3766","post","type-post","status-publish","format-standard","hentry","category-badania-i-analizy"],"acf":[],"_wp_attached_file":null,"_wp_attachment_metadata":null,"wpml_media_processed":null,"_wpml_media_usage_in_posts":null,"_wp_attachment_context":null,"_oembed_35c905c64c03156f243b94f18c4eb80f":null,"_wp_attachment_image_alt":null,"rank_math_description":"Marketing mix modeling shows how much of your sales comes from media, price and seasonality. Learn what data the model needs and where it falls short.","rank_math_focus_keyword":"marketing mix modeling","rank_math_contentai_score":null,"_wpml_post_translation_editor_native":null,"_menu_item_type":null,"_menu_item_menu_item_parent":null,"_menu_item_object_id":null,"_menu_item_object":null,"_menu_item_target":null,"_menu_item_classes":null,"_menu_item_xfn":null,"_menu_item_url":null,"_wp_page_template":null,"rank_math_og_content_image":null,"_wp_trash_meta_status":null,"_wp_trash_meta_time":null,"_wp_desired_post_slug":null,"rank_math_primary_category":null,"_acf_changed":null,"wp_pattern_sync_status":null,"_form":null,"_mail":null,"_mail_2":null,"_messages":null,"_additional_settings":null,"_locale":null,"_hash":null,"_config_validation":null,"_wp_old_slug":null,"rank_math_internal_links_processed":"1","_top_nav_excluded":null,"_cms_nav_minihome":null,"_thumbnail_id":null,"_last_translation_edit_mode":"native-editor","_wpml_word_count":"2336","_dp_original":null,"_edit_last":null,"_edit_lock":null,"rank_math_seo_score":null,"_wpml_location_migration_done":null,"_wpml_media_duplicate":null,"_wpml_media_featured":null,"_wp_old_date":"2026-09-24","copied_media_ids":[],"referenced_media_ids":[],"rank_math_title":"Marketing mix modeling: how MMM works | Hume's Institute","job_department":null,"_job_department":null,"job_location":null,"_job_location":null,"job_offer_external_link":null,"_job_offer_external_link":null,"footnotes":null,"inline_featured_image":null,"blog_podtytul":null,"_blog_podtytul":null,"blog_czas_czytania":null,"_blog_czas_czytania":null,"blog_dalsza_lektura":null,"_blog_dalsza_lektura":null,"slownik_krotka_definicja":null,"_slownik_krotka_definicja":null,"slownik_cytat":null,"_slownik_cytat":null,"slownik_na_stronie_glownej":"1","_slownik_na_stronie_glownej":null,"slownik_slowa_kluczowe":null,"_slownik_slowa_kluczowe":null,"slownik_w_praktyce":null,"_slownik_w_praktyce":null,"slownik_powiazane":null,"_slownik_powiazane":null,"slownik_kluczowe_punkty":null,"_slownik_kluczowe_punkty":null,"lang":"en","translations":{"en":3766},"pll_sync_post":{},"_links":{"self":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3766","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/comments?post=3766"}],"version-history":[{"count":1,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3766\/revisions"}],"predecessor-version":[{"id":3767,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/posts\/3766\/revisions\/3767"}],"wp:attachment":[{"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/media?parent=3766"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/categories?post=3766"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/tags?post=3766"},{"taxonomy":"slowa_kluczowe","embeddable":true,"href":"https:\/\/humes.pl\/en\/wp-json\/wp\/v2\/slowa_kluczowe?post=3766"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}