Agentic measurement is both overhyped and genuinely useful — when handled correctly. This is the workflow I run my own MMMs, attribution and experiments through, with a coding agent in the loop and the output good enough to put in front of a CFO.
Nine self-paced modules · 11+ hours of content. Not “prompt better and the agent writes the MMM” — a workflow.
The agent spits out a Meridian model in twenty minutes. It looks plausible. You ship it. The CFO finds a leakage you didn’t catch, and you’re back to defending a number you don’t actually understand. The model is fine — your relationship to it isn’t.
You don’t trust the agent, so you do everything by hand. Two weeks per MMM. You miss the next experiment cycle. The “we can’t move that fast” meetings start.
The Agentic Measurement Loop is the path between. The agent owns the implementation. You own the structure, priors, diagnostics, and readout. Learn → Replay → Verify → Capture → Encode → Eval. That is the entire course in one sentence.
You know adstock, saturation, and Meridian. You have not yet figured out how to run the workflow through an agent without giving up the rigor that makes the output defensible.
You can model. You are now expected to ship measurement work with an AI assistant, and you want a project setup that works on day one instead of three weeks of trial and error.
Small team, large ad budget, growing list of asks. You need a repeatable agentic loop for EDA, MMM, and experiments so one person can credibly cover the work.
The Agentic Measurement Loop turns ad-hoc prompting into a repeatable workflow your team can rely on — the same six stages on every dataset, every model, every readout.
Install project instructions that define good marketing-measurement work before the first prompt.
Run EDA, MMM, Meridian, experiment, and attribution workflows on realistic marketing data.
Use grading harnesses and review checks to catch wrong columns, broken grain, weak diagnostics, and bad claims.
Keep the prompts, notebooks, eval scripts, and templates for the next dataset or team process.
Reviews of Gui’s teaching — from his live Marketing Science Bootcamp on Maven and readers of his book, Data Analytics for Marketing. This course brings that same teaching to the agentic workflow layer.
Students come from
Same teaching, now packaged for the agentic workflow layer.
Enroll now
Led measurement for €100M+ in ad spend at Kindred Group & Just Eat Takeaway. Author of Data Analytics for Marketing (Packt, 2024). Teaches the Marketing Science Bootcamp on Maven — 5.0★.
Across in-house roles and enterprise clients at Pinemarsh Consulting, Gui has built measurement models from scratch and managed the vendors who sell them — both perspectives are taught here. This course is the operating model he runs his own measurement work on, every week.
“The job is not ‘use the agent more.’ The job is to set up the agent so the work it produces survives a peer review on the first read. This course teaches that setup, in marketing measurement, end to end.”
Previously at
Nine modules take you from first agent setup to reusable measurement workflows. The first five go live on 2026-06-15; the final four follow on 2026-07-15.
Available from 2026-06-15
You can set up AGENTS.md or CLAUDE.md for marketing analytics, pick the right harness for the job, and validate agent output without trusting it blindly.
The operating model behind agentic measurement work, what project instructions actually do, and a side-by-side comparison of Claude Code, Codex, and pi.dev. Includes the seven-question validation checklist you will reuse in every later module.
60 min video, project-instructions template
Available from 2026-06-15
You can hand an agent a fresh marketing dataset and get back an EDA you would defend in a stand-up.
Have an agent profile a marketing dataset end to end: column roles, time grain, missingness, seasonality, structural redundancy. Self-grading harness catches the mistakes agents most often make on real marketing data.
90 min video, graded notebook
Available from 2026-06-15
You can ship an OLS MMM in an afternoon and explain every parameter choice the agent made.
Fit the simplest credible MMM, national-aggregate OLS, and read the diagnostics honestly enough to reject it. The rejection is the bridge to Meridian: it writes down the structural hypotheses (seasonality, trend, per-region) that hierarchical priors answer. You also ship your first reusable skill file.
90 min video, graded notebook
Available from 2026-06-15
You can prepare data for Meridian, get a fit that converges, and produce a diagnostics-clean readout.
Fit a Meridian MMM with agent assistance: InputData construction, ModelSpec and priors, MCMC sampling, convergence diagnostics, ROI and response curves. Meridian for the agent-friendly API surface; PyMC-Marketing notes included.
75 min video, graded notebook
Available from 2026-06-15
You can take a CMO question and return a power-analysed experiment design plus a defensible readout.
Design and analyse marketing experiments with an agent in the loop: hypothesis, estimand, power analysis, MDE, geo-lift design, readout. Bridges back to MMM as the calibration signal.
90 min video, graded notebook
Available from 2026-07-15
You can produce an attribution read, state its assumptions, and explain where it can and cannot be trusted.
Build a defensible attribution read with an agent, from multi-touch and data-driven attribution to its limits, and why it disagrees with your MMM. The agent assembles the pipeline; you adjudicate which signal to trust for which decision.
75 min video, graded notebook
Available from 2026-07-15
You can reconcile three measurement methods into one defensible recommendation.
Bring MMM, attribution, and experiments into one coherent story. Use experimental lift to calibrate MMM priors, and triangulate a recommendation when the three methods disagree.
60 min video, graded notebook
Available from 2026-07-15
You can ship an eval suite that flags when an agent’s measurement output drifts or breaks.
Build evals so your agentic measurement workflow stays correct as data, models, and prompts change. Turn the validation checklist into automated checks that run every time.
60 min video, graded notebook
Available from 2026-07-15
You can roll out a consistent agentic measurement workflow across a team, not just your own machine.
Turn your personal agentic setup into something a team can run. Package project instructions, skills, and evals so measurement work is consistent across people and repos.
60 min video, templates
One purchase, all nine modules, every future update included. No upsell email later.
All nine modules. Founding price closes 2026-06-14.
Enroll nowLess than half what the live Marketing Science Bootcamp on Maven runs ($1,249) — for content the cohort doesn’t cover (the agentic workflow). And a rounding error next to a single $25k MMM engagement the course helps you deliver. Founding access is open today at $539 and closes 2026-06-14; price rises to $674 when the first modules ship, then to $899 once all nine are live. Every future module is included, no upsell later.
Open today, closes 2026-06-14
$539
40% below the $899 full price · 1:1s for the first 20
Active right now
Only the first 20 get 1:1s — 18 spots left
Open to everyone, no seat limit — the lowest this course will ever be. The first 20 buyers also get 1:1 time with Gui. Nothing ships before 2026-06-15, and you can refund anytime until then.
2026-06-15 to 2026-07-14
$674
25% below the $899 full price
Upcoming
First 5 modules are live the day you buy; the final 4 ship 2026-07-15. The self-paced bundle without the founding 1:1s.
From 2026-07-15
$899
All 9 modules shipped
Upcoming
Everything is live. No calendar to track — the complete, self-paced course.
Every drop on this list is included in the active price. No surprise upsells.
What you’re paying for, anchored honestly: the live cohort runs at $1,249. A single hour with Gui as a consultant exceeds a quarter of the founding price. One good budget reallocation the course teaches you to defend is worth more than the course, many times over. We don’t do “Total value $24,000” stickers — anchor it yourself.
Founding refund: 100% anytime before 2026-06-15, no questions. After delivery starts, a standard 14-day window applies.
Most learners expense this. It is professional training that pays for itself the first time you defend a budget decision with it, and plenty of people get it approved even without a formal training budget. Copy the email below, fill in the brackets, and send it to your manager.
Subject: Approval request: Agentic Workflows for Marketing Measurement ($539) Hi [Manager], I would like to expense a self-paced course that directly supports my work on marketing measurement: Agentic Workflows for Marketing Measurement (https://marketingscience.dev/courses). What it is: a practitioner course on running marketing measurement (MMM, attribution, incrementality, calibration, and evals) with AI coding agents such as Claude Code, Codex, or pi.dev. It teaches the workflow layer, not just the theory, so I can build and validate models faster. Why it is worth it: - I will be able to build and stress-test marketing mix models, attribution, and incrementality analyses with an agent, cutting the time each one takes. - It covers calibration and triangulation, so our budget decisions are defensible rather than guesswork. - A module on automated evals means the agent's output gets validated, not trusted blindly. - Lifetime access, so I can reuse the workflows on our own data and share them with the team. Cost: $539 one time. The founding price rises after 2026-06-14, so approving before then locks in the lower rate. There is a full refund window if it is not a fit. Time: self-paced, around 10 hours of video, fit around my workload, with lifetime access. I am happy to share a short write-up of what I apply afterwards. Can I go ahead? Thanks, [Your name]
The founding price $539, locked in — the lowest this course will ever be — plus all 9 modules, lifetime access, certificate, private Discord, and a live kickoff workshop. The first 5 modules go live on 2026-06-15 and the final 4 on 2026-07-15. The first 20 founding buyers also get 1:1 time with Gui and a 1:1 harness-setup call. You take zero delivery risk: refund anytime before 2026-06-15, no questions.
2026-06-14. The founding price $539 and the 1:1s disappear the moment the first modules go live. After that the price moves to $674, then to $899 once all nine modules ship on 2026-07-15.
The founding price is open to everyone until 2026-06-14 — there is no seat cap, buy as a team if you like. The only limit is the 1:1s: the first 20 founding buyers get 1:1 time with Gui and a 1:1 setup call, because that is the most he can give real attention to. Every founding buyer still gets the live kickoff workshop. After 2026-06-14 the price moves to $674 without the 1:1s.
For the founding cohort: 100% refund anytime before 2026-06-15, no questions, no forms, just email. You are paying for a position in the library that opens that day, and you can pull it back any time before it opens. After delivery starts, a standard 14-day window applies.
Yes, and most learners do. This is professional training, so it usually goes through a learning and development or team budget, and plenty of people get it approved even without a formal one. The "Ask your company to pay for you" section above has a ready-to-send email template: copy it, fill in the brackets, and send it to your manager.
Five modules go live on 2026-06-15: Agentic Foundations, Agentic EDA, Building MMMs with an Agent, Bayesian MMMs in Meridian, and Experimentation with an Agent. The final four — Attribution, Calibration & Triangulation, Evals, and Reusable Team Workflows — ship on 2026-07-15. Both dates are public, so you can hold the calendar to this page.
Two reasons. Honest one: it lets the product start delivering sooner instead of waiting for everything to be recorded. Pedagogical one: most learners cannot consume ten hours of dense MMM video in one weekend. The split lets you apply the first five modules on your own data before the rest land.
$539 for the founding cohort through 2026-06-14. $674 from 2026-06-15 once the first modules ship. $899 from 2026-07-15 once all nine are live. Anyone who buys earlier keeps every later module, no upsell email.
Yes. This course assumes you have built a Marketing Mix Model before, or at least know what adstock, saturation, contribution, and DECOMP.RSSD mean. If you do not, start with Gui’s live Marketing Science Bootcamp on Maven first.
You need access to at least one of them to follow along. The course shows how to set up project instructions for all three so you can pick the harness that fits your workflow and budget. None of the three are bundled with the course.
Every module ships with a self-grading harness for the agent’s output and a seven-question validation checklist you can rerun manually. The validation habit is the spine of the course, not an afterthought — module 08 turns it into automated evals.
The course uses a synthetic multi-geo marketing dataset that mirrors the shape of real spend, revenue, and seasonality data. Practitioner-scenario exercises include messy variants so you have practice before you point an agent at your own data.
Guilherme Diaz-Berrio — author of Data Analytics for Marketing (Packt, 2024), 15+ years in marketing analytics, and instructor of the Marketing Science Bootcamp on Maven (rated 5.0 from 7 reviews). He has led measurement for €100M+ in ad spend.
Lock in $539 today and build the agentic workflow you can reuse on real measurement work. First 20 buyers get 1:1 time; everyone gets the live kickoff workshop and every future module.