Marketing Engineering with AI · Coming soon
Every ad platform claims the same sale, and finance wants one answer
For marketers and analysts asked which channel works, whose platform reports add up to more sales than the store recorded. You build rule-based and data-driven attribution, then a small marketing mix model, on public GA4 and Olist data, and write a one-page readout that says what each method can prove and where it is uncertain.
The course: Attribution and marketing mix modelling you can explain to a CFO
$25 $49 early-bird price
This course is being built now. Join the early-access list to reserve the early-bird price and shape what it covers.
Joining the list is free. You only pay if you buy at launch, at the early-bird price.
If this sounds like your week
- Meta, Google and GA4 each take credit for the same order, so the channel totals add up to more than your revenue.
- Last click makes prospecting and video look worthless, and you suspect it is wrong but cannot show why.
- Finance asks which spend to cut, and "the platform says it works" is not an answer they accept.
- You tried an open-source mix model, it produced numbers, and you could not explain them to anyone.
What you do now, and where it stops working
Most teams read each platform’s own dashboard, or GA4 on last click, and reconcile the gap in a spreadsheet before the budget meeting. Each platform counts conversions under its own rules, so the reconciliation never closes, and the meeting argues about whose number to believe instead of what to fund.
What changes after the course
- One table shows each channel’s credit under last click, first click, linear and a data-driven rule, side by side.
- A mix model separates baseline sales, seasonality and paid spend, so you can see what would have sold anyway.
- A one-page readout for finance includes a section that states what the model cannot prove.
- A planned holdout or geo test checks the model’s largest claim against what happens.
How it works: Three views, one readout
- 1Reconcile: rebuild orders from raw data so every channel is credited against the same revenue total.
- 2Attribute: compare rule-based and data-driven attribution in SQL and Python, and find where they disagree.
- 3Model: fit a small mix model with carryover, saturation, seasonality and baseline, and read its uncertainty ranges.
- 4Test: design an incrementality test for the channel where the methods disagree most.
- 5Report: write the CFO readout — what each method shows, how confident it is, and what the test will settle.
What you get
- SQL and Python notebooks for attribution and the mix model
- Sample dataset prepared from public GA4 and Olist data
- CFO readout template with an uncertainty section
- Incrementality test planning worksheet
- Video walkthrough of the model from raw data to readout
This is for you if
- You have to justify the marketing budget by channel and want a method finance can audit.
- You write SQL and will run Python notebooks with a walkthrough beside you.
- You have read about mix modelling, or tried an open-source tool, and need to understand its output.
Skip it if
- You want software that reports a "true ROAS" automatically. This course teaches the method.
- You already run calibrated mix models with regular lift tests.
- You read dashboards and do not want to open SQL or a notebook.
Before you join
- Is mix modelling only for big brands with big budgets?
- With modest spend or a short history, a model’s ranges get wide. The course shows you how wide, puts that in the readout, and covers when a simple holdout test answers your question better than a model does.
- Can I explain this to a CFO, or will it look like a black box?
- The readout is what the course builds toward. You write one page in plain terms: what each method shows, the range of uncertainty, and the test that would settle the biggest open question. A template with an uncertainty section is included.
- Is attribution worth doing now that tracking is broken?
- Click-path attribution sees less every year, which is why the course does not stop at it. You use attribution for what it still shows, the mix model for what it cannot see, and an incrementality test to check both.
- My business is nothing like the course data. Will it transfer?
- The course runs on public GA4 and Olist e-commerce data so everyone can follow the same steps, and no client data is involved. Each notebook marks the inputs you replace — your orders, and your spend by channel and week — to run it on your own business.
- Who is teaching this?
- Jay West leads DevOps for the marketing organisation of a multi-location specialty retailer. He is building this course in public as he builds the project, and the working notebooks are included. The course is plain about what these models cannot do, because the readout depends on it.
Rather have it built for you? Describe the project and the same work can be done for your team.