Why it is not worth debating if MMM can measure Long-Term Brand Impact
“Can Marketing Mix Modeling (MMM) measure long-term brand impact?” In my opinion, this debate is largely misplaced.
📌 No consensus on what “Long Term” actually means
There is a joke in statistics – ask 3 statisticians for an opinion and you get 5.
Marketing is no different. Every marketer seems to have their own definition of what long term means.
Before asking whether MMM measures long-term impact, we should ask a more basic question:
What exactly is long term? 3 months? 6 months? 1 year?
Even marketing experiments such as Geo-Lift or Brand Lift measure impact only during the experimental window.
If a geo experiment runs for 6 weeks, it measures lift within those 6 weeks. No experiment can directly tell you what happens 18 months later.
In that sense, experiments are actually more short-term than MMM !!
📌 MMM already captures memory effects
MMM has mechanisms such as adstock that capture how much of your advertising is remembered weeks or months later.
When time-series dynamics such as ARMA are present, additional persistence effects also emerge.
Which means the impact of today’s marketing can extend into the future.
In that sense, MMM already captures ‘long term’ effects over time, which is what many people informally call long-term brand impact.
📌 Where things go wrong: extrapolating ROAS into the future
The real confusion starts when people attempt statements like:
“If TV ROAS today is 2, the true long-term ROAS is 4 after one year.”
Many frameworks justify this by:
– Benchmarking against hundreds of brands
– Estimating an average multiplier
– Applying that multiplier to MMM outputs.
This may sound scientific, but statistically it is extremely fragile.
Brand effects depend on things like:
– category maturity
– brand equity
– creative quality
– distribution strength
– price and promotion strategy
– competition
A multiplier derived from other brands in other markets cannot reliably describe your brand’s dynamics.
📌 Industry Benchmark Obsession.
We wrote earlier that industry benchmarks should not be a north star.
Every brand is trying to outperform the category. By definition, the average is not where anyone should aim.
This becomes even more distorted when applied to ROAS or MROAS benchmarks.
📌 MMM is not designed to predict distant futures
Just last week, Ridhima wrote in her excellent post:
“Good MMMs are not built to predict the future. They are built to change the future.”
If a model induces changes in budgets, channels or promotions, the system itself changes and future ROAS/ MROAS will change as well.
So why obsess over so-called long-term effect?
To me, this is simply the wrong yardstick to judge MMM.
MMM is not a fortune-telling machine. Its job is to help businesses make better decisions that reshape reality.
John Maynard Keynes “In the long run we are all dead” 🙂.
