Resources
How marketing data actually works.
Warehouses, tracking, attribution, reporting, budgets, AI, and the checks that keep a number right. Written to be useful whether or not you ever buy anything.
Resources
- 6 min read
How to automate client reporting.
Not the sales version. The five pieces of work it actually takes, and what each one costs.
Read it - 7 min read
What it takes to onboard a new client's data.
The job that happens once per client, and the part no tool can do for you.
Read it - 6 min read
What goes in a monthly client report.
The structure clients read, and the parts that only make it longer.
Read it - 5 min read
Year-over-year reports when the definitions have changed.
The comparison every client asks for at the end of the year is the one most likely to compare two different things.
Read it - 7 min read
What a marketing data warehouse is, and when an agency needs one.
A plain explanation, and an honest answer on whether your agency needs one yet.
Read it - 7 min read
What a marketing data warehouse costs to run.
Two bills, and only one of them is decided by how your queries are written.
Read it - 6 min read
What the GA4 export gives you, and what it leaves to you.
Raw events rather than reports. Useful, and further from a dashboard than it looks.
Read it - 5 min read
Why AI gets your clients' marketing data wrong.
Connecting a model to a client's data does not make it right. Here is what breaks.
Read it - 6 min read
Can AI write your client reports?
The writing is the easy part. The numbers underneath decide whether it is right.
Read it - 7 min read
Why ad platform numbers never match on the same campaign.
They are not measuring the same thing. Once you know what each one counts, the gap stops being a mystery.
Read it - 7 min read
Reporting across currencies and time zones.
Both are the same failure. A number that means something different depending on where it was read, and no error either way.
Read it - 5 min read
Meta's engage-through attribution, explained for client reports.
Fewer conversions in Ads Manager does not always mean fewer sales. Sometimes it means a narrower definition.
Read it - 7 min read
A UTM naming convention that holds across clients.
UTMs are the join between what a client spent and what happened. Get them wrong and no amount of tooling recovers it.
Read it - 7 min read
A GA4 and tracking audit checklist for new clients.
What to check, in the order that stops the most bad data, and how to tell whether Consent Mode is actually working.
Read it - 5 min read
ETL, ELT and reverse ETL, in plain English.
Three terms for moving data around, and the only difference that matters to an agency.
Read it - 7 min read
What happens when a platform changes its API.
A pipeline that stops gets fixed today. A pipeline that keeps returning rows can be wrong for a month.
Read it - 6 min read
How to tell if client marketing data is wrong.
A dashboard draws whatever it is given. These are the checks that tell you it was given the wrong thing.
Read it - 7 min read
What a consent choice does to your reporting data.
It decides what is collected, what gets estimated on your behalf, and why two figures for the same week are both correct.
Read it - 6 min read
Server-side vs client-side tracking, and what changes in client reports.
More conversions, a break in your trend line, and a new way to count one sale twice.
Read it - 6 min read
How AI agents actually read client marketing data.
Giving an agent access is the easy half. Giving it something it can answer from is the work.
Read it - 7 min read
A dashboard filter is not an access control.
Every rule you enforced inside a reporting tool stops existing the moment something queries the tables underneath it.
Read it - 6 min read
How to measure a client's traffic from AI search.
A lot of it lands in Direct and gets credited to nothing. Here is how to pull it apart.
Read it - 7 min read
How AI assistants choose what to cite.
Mostly the boring things. The text is there without running anything, and the page answers the question it claims to.
Read it - 5 min read
Search Console's AI report, and what to tell clients about it.
It answers one question well: how often a client's pages were shown in an AI answer. Do not ask it for the rest.
Read it - 7 min read
What a marketing data pipeline is, and what breaks in it.
Four stages, and each one fails in its own way. Knowing which is which is most of the job.
Read it - 7 min read
How far back a new client's marketing data goes.
Further than you expect in places, and gone for good in others. Worth knowing before the day you need it.
Read it - 6 min read
Attribution and marketing mix modelling, compared.
They answer different questions, and using one for the other is where most budget arguments with clients come from.
Read it - 6 min read
MER vs ROAS, and which one to report to clients.
One number every platform claims for itself, and one nobody can claim at all.
Read it - 6 min read
How to test whether a client's ads caused the sales.
Reported conversions tell you what was near a sale. A holdout test tells you what caused one.
Read it - 6 min read
What data you need for marketing mix modelling.
The model is the quick part. The data it learns from is the project.
Read it - 7 min read
What it takes to build a unified customer view for a client.
Four systems each hold part of a person. Joining them up is a keys problem, not a tooling problem.
Read it - 7 min read
Sending warehouse data back out to the ad platforms.
Reporting is data coming in. Activation is the same data going back out, and it fails in different ways.
Read it - 6 min read
Who owns client data: the agency, the client or the vendor.
The question is not who collected it. It is what each party still has the day a contract ends.
Read it - 6 min read
What a first-party data strategy actually involves.
Less a strategy than an inventory, a home and a set of permissions.
Read it - 5 min read
How to choose client reporting software.
Most tools make a good-looking report. The differences are in what they charge for and where the data lives.
Read it - 6 min read
How to report on fifty clients without fifty rebuilds.
The work should not grow with the roster. If it does, the problem is the shape of the data, not the number of clients.
Read it - 7 min read
Keeping one client's data out of another's report.
The failure is never dramatic. It is a copied dashboard with one filter somebody forgot to change.
Read it - 5 min read
What white-label reporting actually means.
Your name on the report is the easy part. Your name on its mistakes is the part to plan for.
Read it - 5 min read
Data Studio for agency client reporting.
It is free, familiar and good at charts. The trouble starts when the same report has to exist forty times.
Read it - 6 min read
How to write metric definitions that hold across clients.
Most arguments about a number are arguments about what the number means. Write it down once.
Read it - 6 min read
Budget pacing: the formula, and the report.
Spend against plan, while there is still time to do something about it.
Read it - 5 min read
How to report AI Max campaigns to clients.
When Google picks the query, the headline and the page, the report has to show what it picked.
Read it - 5 min read
How often marketing data should refresh.
Almost always daily. Here is why more often is usually worse, not better.
Read it - 6 min read
What MCP changes about asking AI for marketing numbers.
It gives an assistant a way in to each tool. It does not give it a way to compare them.
Read it - 6 min read
What a semantic layer is, and whether you need one.
One definition, read by every tool that asks, for every client. Worth the service at a certain size, overhead below it.
Read it - 7 min read
How to report a client's visibility inside AI answers.
Most of it never becomes a visit. What is left is a sample, and a sample is only worth reporting if it has a method.
Read it - 6 min read
How to set up alerts across client accounts, and keep them worth reading.
An alert that fires every day becomes a filter rule. Four things are worth interrupting somebody for.
Read it - 6 min read
Why reporting breaks as the roster grows.
The setup that was fine for five clients fails at thirty, and it fails in three specific ways.
Read it
Get started
Want this running on your own data?
Book a call with someone who can answer the technical questions.