How marketing mix modelling works
MMM looks at a long run of history — typically weekly totals — and tries to explain the movement in sales or leads. The inputs are aggregate: spend by channel, plus everything else that moves demand, such as price changes, promotions, public holidays, festival seasons, competitor activity and weather where it matters. A regression then estimates how much of the outcome each input accounts for.
Because it works on totals, no cookie, click identifier or logged-in user is involved. It can include offline advertising, print, radio, sponsorship and even distribution changes, which no click-based method can touch. Good models also estimate saturation, showing where extra spend in a channel stops producing a matching increase, and carryover, the tail of response that continues after a campaign stops.
Why MMM matters
It answers the planning question rather than the reporting one: given a fixed budget for next quarter, roughly where should it go? Click attribution cannot answer that, because it only ever sees the channels it can tag, and it is blind to the brand advertising, word of mouth and shelf presence that create the demand those clicks capture.
Its second virtue is durability. Consent refusal, cookie expiry and closed platform reporting all degrade user-level tracking; none of them affects a model built on weekly totals. That is why large advertisers have returned to a technique that predates the web.
Where MMM goes wrong
The honest limits are severe. A model needs years of consistent weekly history, real variation in spend, and inputs recorded accurately — if channel budgets always moved together, the model cannot separate them. Most small and mid-sized advertisers simply do not have that history, and a model built on thin data will still produce confident-looking output.
Correlation is the deeper trap. A model observes that spend and sales moved together; it does not prove one caused the other. Spend that follows demand — extra advertising during a festival period, for example — will be credited with the sales that season would have produced anyway. Vendors compound this by presenting results without any range around them, so a rough estimate is read as a precise fact.
How to act on it
Treat MMM as a compass for budget planning, not a measurement of last week. Use it to set the shape of a plan and to spot channels running past the point of diminishing return, then use tagged reporting for day-to-day optimisation and an incrementality test to check any conclusion that would move serious money.
Ask any provider three things: what data went in, what uncertainty sits around each estimate, and which results have been validated against a live experiment. A model that cannot answer those is a story with equations attached. For most businesses in Nepal the practical starting point is much simpler — clean, consistent tracking and reporting and a tidy record of monthly spend and revenue — because a model is only ever as good as the history feeding it.
