Paid Media and Data Library

What we covered

The check order

1. Errors and rejected ads.
2. Cost per result.
3. Link click-through rate, compared with last month.
4. Frequency.
5. The creative breakdown: which headline, text and image are winning.
6. GA4: what those visitors did on the site.
7. Conversion rate, interpreted with the visit and conversion counts.

Stop and act at the first problem you find. Checking in a fixed order keeps a creative problem from being treated as an audience problem.

Structure

• In this fictional account, campaigns group audiences and ad sets group creative angles. Confirm the structure appropriate to the real account.
• The sample routine puts a new angle in its own ad set. Review delivery and learning status after changes.
• Investigate fatigue and measurement before choosing a creative refresh, pause or other change. Check the current platform guidance on learning.
• One change at a time. Edit one ad set, confirm it clears review, then move to the next.

Creative, compliance and signal

• The sample test plan uses distinct creative concepts. Set the number of variants around the available budget and decision being tested.
• Pick the next test from evidence: pages visitors choose on their own, what wins in email, Search Console queries, and long-running ads in the Meta Ad Library.
• Log every rejected phrase with its rewrite, then request review.
• Language expansion is a test like any other: fewer advertisers compete in some languages, so cost can drop.
• The server-side signal (Conversions API through a hosted container such as Stape) catches conversions the browser pixel misses. When results look wrong, check measurement before creative.

What we covered

• utm_source = platform. utm_medium = channel. utm_campaign = the campaign. Lowercase, no spaces.
• UTMs go on marketing links only: ads, emails, texts, QR codes, mailers.
• Keep UTMs off internal buttons and links. Use events to measure those interactions. Internal campaign tags can affect attribution without starting a new GA4 session.
• A large share of direct or unspecified traffic is a reason to investigate tagging and source capture.
• Past data cannot be re-tagged, so fix the standard first and fix the highest-volume channel first.

Why the numbers never match

• Platform reporting can include different attribution windows and interaction types. CRM credit depends on the source-capture rules in that setup.
• Compare GA4, platform and CRM reports using their actual conversion definitions and attribution settings.
• If configured, GA4 can show form starts and phone clicks. These interactions are not necessarily identifiable leads or completed calls.

What we covered

• Find the saved explorations in Explore. Confirm sharing and access for your role, then inspect how the reports were built.
• Event count measures recorded occurrences. Read user metrics using their definitions and the filters in the report.
• Record a confirmed booking. Test repeat clicks, reloads and direct visits so the event does not count an unconfirmed or duplicate outcome.
• DebugView shows every event you trigger while testing, with its parameters.
• Health check: the events list should hold only events this business can produce.

Diagnosing a drop

1. Is it traffic or conversion?
2. Then which channel.
3. Then which funnel step.
4. Then whether a page changed.

What we covered

• Tags do the work. Triggers decide when a tag fires. Variables are the data a tag reads.
• Build and check in a development container, then copy to production.
• Test a click tag: click the target and confirm it fired, check the Tags Not Fired list, then click other links and confirm it does not fire.
• Test when something looks off. A 1,000% jump in phone clicks means check the tag before you celebrate.
• New platform? Check the community template gallery first.

From test to permanent

1. Build in development.
2. Launch at low traffic and watch it.
3. Reach significance.
4. Send all traffic to the winner in GTM.
5. Ticket the web team to build it in.
6. Pause the tag once it is in the code.

What we covered

• This sample setup passes an available GA4 client ID through the form to the CRM. Confirm identifier availability, consent and the export fields before planning a join.
• The join can connect collected behavior to available outcomes. Missing identifiers, consent choices and data retention limit coverage.
• Spend BigQuery only on questions that need that join. Anything GA4 already answers belongs in a GA4 report.
• "Do you want to count things or do things?" Pick the analysis whose answer tells you the next move.

Worked example: repeat bookings

• Existing customers who book through an ad look like new conversions to Meta, so Meta looks for more people like them.
• Analysis order: by medium, then source within the top medium, then pages, then buttons.
• Run the mirror image too: where the new customers come from, and how to get more of them.

What we covered

• Variants are arms. Each arm changes one thing.
• Track form starts and form completions separately. A partial fill is an interaction. It is actionable only when contact information is available and follow-up is permitted.
• Set the test duration and sample requirements in advance. If evidence is inconclusive, keep the control and investigate further rather than declaring a winner.
• Compare the needs of visitors from each source. Test how much explanation and form detail each audience needs.

Website triage

1. Find the top pages by traffic.
2. Low form starts: investigate the offer, page content and call-to-action visibility.
3. High starts, low finishes: investigate errors, field friction and the information requested.

What we covered

1. A 2 to 3 line headline: if you read nothing else.
2. The numbers.
3. Three to five bullets on what they mean.
4. Next steps and the ask.

Pre-send checklist

• Is spend included?
• Does every number have a "so what"?
• Are the limits named?
• Is the ask a yes or no, or a date?
• Are you comparing like with like?

What we covered

• Use it for repetitive work, pattern-finding in big exports, brainstorming, and predicting the questions you will get.
• Load the reference documents first. Ask it to walk you through a task step by step.
• Know what the answer should look like before you ask. That is how you catch its mistakes.
• Test every output before it goes live.

What we covered

• Brand activity aims to improve recognition and consideration. A test examines whether the selected signals change; it does not guarantee cheaper acquisition.
• Nobody looks up at a blimp and buys tires that afternoon. When they need tires, they already know the name.

Measuring lift

• Branded search in Search Console is the main signal in this method. Compare it with a baseline and account for other explanations.
• Conversion rate and cost per result across channels.
• A "how did you hear about us" question on the thank-you page.
• Use 12 months of baseline data. Plan digital tests for 4 to 6 weeks with 2 to 4 weeks of further observation. Offline test and observation windows are set separately in the plan.

The method

1. Research where the audience spends time, with sourced data.
2. Pick channels to test.
3. Set the measurement window.
4. Run a low-cost test.
5. Read the result against the baseline and the agreed comparison. Record whether the evidence is positive, negative or inconclusive.

From objection to trust message

• Safety. "I let a technician into my house and it went fine."
• Ease. "One call, and the price was set before the work started."
• Support. "Someone answered the phone at every step."