Workshop notes,
logged.

An anonymised Google Meridian build for a four-market DTC brand. The first fits credited PMax with double Google's own figure; the ad account showed why, and brand-search clicks plus Google Trends halved it. What held, what moved, and what it means for the media plan.

A redirect can drop a click ID while the page still loads. Click Journey, a free Chrome extension, follows a click through every hop and says which one lost its click IDs and UTMs, without sending anything out of the browser.

Bubble's October creative refresh went from a Meta Ads MCP read to an approved 33-ad pack in about an hour, and 274 assets have been approved since July: the timings, and a modelled hand build to compare.

Creative automation you own: Claude Design templates with every testable axis as a parameter, YAML batch specs, headless renders traced to a commit, and a filename grammar that joins each ad back to its results.

Impressions and CTR sit at the top of the funnel, but neither sets your CPA. Put in today's numbers, move any one or work back from a target, and CPA, ROAS and profit follow in your browser.

Paste modifiers, services and places and get every useful combination in exact, phrase or broad match, checked against Google's limits, ready to paste or as a Google Ads Editor CSV.

ValueTrack sends campaign IDs, never names, so UTMs and call tracking carry numbers nobody can read. A free Google Ads Script keeps campaign and ad group names in your tracking URLs through launches and renames, Performance Max included.

The knowledge system behind this publication: plain markdown in git, maintained by AI agents through named rituals, a mechanical leak-check between client work and anything shareable, and a one-way pipeline that turns its methods into published articles and tools.

The question that keeps coming up needs less than a deep crawl: is this site technically sound? A free Python CLI answers it in about a minute, with a verdict and the top three fixes.

An in-house Google and Meta operation had grown to 80+ campaigns, with bidding not yet tied to HubSpot signups. Rebuilt on one naming convention with HubSpot as the bidding signal: signups doubled, cost per signup more than halved, spend roughly flat.

Structured data helps an AI answer engine tell who you are, but it doesn't make the engine prefer you. What I shipped: one @id per entity, schema matching the page, copy it can quote, and retrieval bots let in.

Three Lighthouse categories lock at 100; Performance drifts with every change. The measure, fix, re-measure loop in Chrome DevTools, the layout shifts a faster paint exposed, why mobile Performance stopped at 96, and the second pass that took it to 100.

Google's Conversational Analytics answers plain-English questions from a BigQuery table, and a wrong answer looks equally confident. How I set one up read-only by IAM, accurate on the trap questions, reproducible from YAML in git.

When 70-80% of revenue arrives by phone days after the click, the default set-up optimises against on-site proxies. A five-tier funnel, sales with revenue as the bidding goals, and a four-phase rollout.

Drop in a Google or Meta ad and pick three rival angles. An AI-simulated panel scores all four and suggests the test to run. On Google, any usable ad expands into RSA or Performance Max assets.

More than reporting infrastructure: the foundation under attribution, MMM and audience automation, so each one starts from clean, joined data. How I build one for a multi-brand operator running approximately fifteen DTC stores across Europe.

A ten-step Google Meridian tutorial with a public toolkit: one weekly table, a demand control, defensible priors, three holdouts, a sensitivity grid, budget scenarios and the Scenario Planner, refreshed every quarter.

Running one measurement layer well doesn't cover the other: Enhanced Conversions and CAPI feeding the networks, and MMM guiding next quarter's budget. Why both matter, what a disagreement between them signals, and what good looks like.

Google Enhanced Conversions and Meta CAPI recover meaningful ground against consent denial, blockers and cross-device loss. A realistic read on what they fix, what they don't, and where the ceiling sits.

IM review, management interview prompts, systems access questions, attribution red flags, and the three numbers that test whether a target's growth is real or model-driven. From multi-million pound deals advised since 2020.

Catch the Consent Mode v2 failures that can break tracking for weeks without an error. Seven 60-second staging checks before any production deployment that touches measurement.