The folklore problem
Best practices nobody has tested. Attribution models mistaken for ground truth. Dashboards that measure activity instead of effect. Claims pass between agencies and conference stages until repetition makes them feel true.
media science a working notebook
Media is a science.
Practise it like one.
Digital marketing produces enormous amounts of data and surprisingly little evidence. I’m Rob - I work in digital media, and this notebook is where the industry’s received wisdom gets tested against what the numbers actually show.
Written and edited by a person. Nothing publishes without sign-off - here’s exactly how AI helps.
entry 00 why this exists
An industry running on folklore
The scientific alternative
Treat every campaign, tag and ranking theory as a hypothesis. Instrument it properly, test it where you can, and be honest about uncertainty where you can’t. Some cherished tactics survive that treatment. A surprising number don’t.
the beats four programmes, one discipline
What gets tested here
Analytics & measurement Does the dashboard measure activity, or effect? GA4, attribution, incrementality - and the difference between what the numbers say and what actually happened. Tagging & data infrastructure Can this data support the conclusions built on it? GTM, consent mode, server-side tagging: the unglamorous plumbing that decides what you can know before analysis starts. Search & AI discovery What does ranking mean when the reader is a model? SEO as it collides with AI answers - what still ranks, what gets cited, and how discovery is being rebuilt in front of us. Media strategy Would this budget survive an experiment? Channel and planning decisions treated as hypotheses with costs and error bars, not articles of faith.
protocol applied to every post
The method
- Ask A falsifiable question - one the data could actually answer no to.
- Instrument Tagging, consent and collection choices decide what you can know. Plumbing first, or the conclusions are decoration.
- Test Experiments where possible, quasi-experiments where not, clearly-labelled judgement where neither is available.
- Decide Write the result down - especially when it is inconvenient - and update the playbook, not the story.
“When you can measure what you are speaking about, and express it in numbers, you know something about it; when you cannot measure it … your knowledge is of a meagre and unsatisfactory kind.”
William Thomson, Lord Kelvin - Popular Lectures and Addresses, 1883
How this notebook is written
An AI research assistant watches the industry’s news feeds and drafts background briefs; a human decides what any of it means. Every post is reviewed and signed off before it publishes - the publishing system enforces it, not just policy.
The full AI use disclosurestart here the archive