basics

technicals

MMM issues

Category: Timeline Stories

  • Adstock functions in MMM

    Two main groups of adstock functions: geometric: adstocki = adsi + 𝞭 * adstocki-1 convolution: adstock = 𝜮conv ⨂ ads check the example

  • MMM vs. linear model

    MMM is not just a linear model; we must make a few adjustments adstock saturation possible lag read more

  • Data quality and availability

    without data, you cannot do MMM, but there are so many issues; most common are: data are not available data are not structured in any way data are confidential

  • What’s wrong?

    nothing; we have to manage our expectations; unrealistic expectation is the main issue of MMM failure: we want to know too many details rather than a general overview data are not available, so we cannot estimate the effectiveness external factors we are not aware of also may have an impact

  • How does the statistical model work and what is it the benefit?

    Model for weekly based on historical data:  KPI sales influencing factors: online ads spend newsletter It can be applied to the media plan. check the example

  • What data do we need?

    for MMM, we need historical data over time: 2-year weekly data – gives the best result 3-year monthly data read more

  • What came first, the chicken or the egg?

    MMM tends to add too much weight to media activity however, during a heavy promotion period, there may be other activities that impact KPI (e.g., increased distribution, visibility) other activity during promotion period may be price discounts or promo bundles

  • Competitors influence

    competitors may influence us both positively and negatively, depending on a category and message we are not sure what will be the competitors’ actions in the future

  • Real reasons to do MMM

    proof that you use sophisticated tools while spending money on marketing activities proof that we spent money wisely show off at a board meeting

  • Issues with MMM

    time-consuming does not invent anything, only optimizing existing spending for media channels result may vary based on assumptions may give unrealistic recommendations