@noahzweben@bcherny@noahzweben@bcherny for which plans it is supported? Initially it was working for me in terminal at least but shows now unsupported for me as max user?
Traditional MMMs treat intercept as constant. But real businesses grow or shrink over time.
Modern tools like Meridian allow a time-varying intercept baseline that shifts with your brand's health.
Small detail, big impact.
Day 15 of learning Marketing Mix Modeling
Back after a break! Today's concept: The Intercept 📉
It's the unsung hero of your MMM equation.
🧵 #LearningMMM
The intercept = your baseline sales when all marketing spend is zero.
It captures:
• Brand loyalty
• Organic demand
• Repeat customers
Basically, what your business earns without ads running.
Advanced MMMs try to capture these interaction effects.
Ignoring synergy means you might cut a channel that's quietly helping others perform.
The whole is greater than the sum.
Example: Someone sees your TV ad, then searches your brand on Google.
TV drove awareness. Search captured intent.
Together they convert. Alone? Maybe not.
MMM gives you the big picture. Incrementality tests give you ground truth.
Use both together:
• MMM for planning
• Tests for validation
Trust, but verify.
Incrementality tests = real-world experiments.
Turn off ads in one region, keep them on in another. Measure the difference.
If your MMM predicted a 10% lift and the test shows 9% you're on track.
As a Python person, I'm starting with PyMC-Marketing.
But the concepts - adstock, saturation, priors - transfer across all of them.
Pick one. Start experimenting. Learn by doing.
Day 11 of learning Marketing Mix Modeling
Time to get hands-on 🛠️
The best part? Top MMM tools are free and open-source.
No expensive consultants needed.
🧵 #LearningMMM
The big three:
• PyMC-Marketing - Python, Bayesian, flexible
• Google Meridian - open-sourced Jan 2025
• Meta Robyn - R-based, automated
Each has trade-offs in customization vs ease of use.
MMM:
• Uses aggregate data
• Works across online + offline
• Privacy-friendly
Better for strategic budget allocation. No user tracking needed. Different tools, different jobs. Use both if you can.
MTA (Multi-Touch Attribution):
• Tracks individual user journeys
• Digital-only
• Needs cookies/tracking
Great for tactical, real-time optimization. But blind to offline.
Why it matters:
If 70% of sales are base, your marketing isn't driving as much as you think.
If incremental is high, you know your spend is working.
MMM gives you the truth, not vanity metrics.
Day 9 of learning Marketing Mix Modelling
Today's concept: Base vs Incremental Sales 📊
Not all sales come from marketing. Some would happen anyway.
MMM helps you separate the two.
🧵 #LearningMMM
Base sales = what you'd sell with zero marketing. Brand loyalty, repeat customers, organic demand.
Incremental sales = the lift your marketing actually created.
This split is the whole point of MMM.
As a data engineer, this feels familiar.
MMM is only as good as the data feeding it. Garbage in, garbage out.
Spend time on data prep. Your model will thank you later.
Day 8 of learning MMM
Real talk: your data won't be perfect 🧹
Missing spend values. Inconsistent channel names. Gaps in history.
So how do you handle messy marketing data?
🧵 #LearningMMM"
Common fixes:
• Missing values → interpolate or use zero (know the trade-off)
• Inconsistent naming → standardize early
• Short history → Bayesian priors help fill gaps
The goal: clean enough, not perfect.