Notes from the account

All articles

• 29 July 2026/9 min read

  Google Ads bid strategies, ranked by how much conversion data they need

  Google's own help pages document far lower requirements than the industry repeats. Target CPA can be started with no conversion history at all, and Target ROAS on Search and Shopping asks for 15 conversions in the past 30 days, not 50. The thresholds below are quoted from Google's documentation and ordered from the strategy that needs the least conversion data to the one that needs the most, so you can find the highest rung your account can actually stand on.

• 29 July 2026/8 min read

  Conversion tracking mistakes, ranked by how hard the damage is to undo

  Every other list of tracking mistakes is ranked by how often they happen, which tells you what to expect rather than what to fix first. Ranked by recoverability instead, the order changes completely: the faults that quietly corrupt months of historical data outrank the ones that break loudly, because a broken tag gets noticed and fixed while a wrong number gets reported to the board.

• 29 July 2026/8 min read

  Google Ads default settings, ranked by how much budget they expose

  A new Google Ads campaign ships with settings already chosen for you, and several of them widen where your money can go. Ranked by how much of the budget each one can quietly redirect, the order is not the one most checklists give: the network settings outrank the match type everyone argues about, because network spend is hardest to see in the reporting.

• 29 July 2026/8 min read

  Conversion actions, ranked by how well they predict revenue

  Most accounts optimise toward the conversion action that was easiest to install, which is almost never the one that predicts revenue best. Ranked by predictive value instead, the order runs from a qualified sales outcome down to a page view, and the gap between the top and the bottom is the difference between buying customers and buying form fills.

• 28 July 2026/8 min read

  How to audit a Google Ads account

  Audit a Google Ads account in four passes, in this order: verify that conversion tracking measures what you think it measures, find spend that reached the wrong queries, check whether the structure allows good bidding, and only then look at the bidding itself. The order matters because every later finding is meaningless if the conversion data underneath it is wrong.

• 28 July 2026/9 min read

  Conversion rate optimization, a practical guide

  Conversion rate optimization is a four-step loop: quantify where people leave, research why they leave, prioritise the fixes by expected impact and evidence quality, then either test the change or ship it and measure honestly. The single most common mistake is skipping the research step and testing ideas taken from best-practice lists, which produces inconclusive results and a team that stops believing in testing.

• 28 July 2026/8 min read

  GA4 conversion tracking that does not break

  Set up GA4 conversion tracking by designing the event taxonomy before tagging anything, marking only real business outcomes as key events, validating every event against a real interaction on a real device, and reconciling the totals against your CRM. Most GA4 setups that break were never validated in the first place, and the failure is usually silent.

• 28 July 2026/7 min read

  How to decide a Google Ads budget

  Your Google Ads budget comes from your own numbers, not from a benchmark. Work out what a customer is worth, decide what you can afford to pay to acquire one, check whether the available search volume can absorb that spend at that price, and then let marginal return decide whether to scale. Any recommended figure that arrives before those four steps is a guess wearing a suit.

• 28 July 2026/7 min read

  Where AI actually belongs in a marketing stack

  Use plain code for anything rule-based, use a language model only where the input is unstructured language and judgement genuinely beats rules, and keep a human on any decision that spends money. Most of what is sold as AI marketing automation is ordinary scheduled automation, which is fine, because that is where most of the value is.

• 28 July 2026/9 min read

  A technical SEO audit checklist that finds real problems

  Run a technical SEO audit in four passes: compare the raw HTML against the rendered DOM, verify indexing and canonical logic, check status codes and redirects, then look at architecture and speed. Do it in that order, because a page whose content only exists after JavaScript runs makes every later finding meaningless. Group the results by cause, not by symptom.

• 28 July 2026/8 min read

  How to test paid social creative and get a decision

  Test paid social creative by consolidating budget into few enough ad sets to exit the learning phase, testing distinct concepts rather than small variations, giving each variant enough spend to be readable, and agreeing a stopping rule before launch. On paid social the creative is the targeting, so a testing system that produces decisions matters more than any single winning ad.

• 28 July 2026/7 min read

  How to write a content brief that makes the piece inevitable

  Write a content brief by starting from what already ranks and what those pages fail to answer, then stating one primary intent, the questions the piece must answer, the specific claims it will support, the internal links it must earn, and how you will judge it. A good brief makes a good piece nearly inevitable, and a bad brief cannot be rescued by good writing.

• 28 July 2026/7 min read

  Google App Campaigns, and the few levers you actually get

  Google App Campaigns remove keyword, placement, and audience targeting, so the levers you have left are the asset mix, the in-app event you optimise toward, the bid and budget, and the SDK setup that reports events back. The single most consequential decision is which event you optimise toward, because everything the system learns is downstream of it.

• 28 July 2026/8 min read

  AI agent use cases, department by department

  AI agents earn their place wherever a documented, repetitive process requires reading unstructured language and then doing something with it. That means support triage, sales research, finance exception handling, operations scheduling, and internal knowledge retrieval. Anything numeric or rule-based should stay as plain code, and anything that spends money or contacts a customer should require a human approval step.