Marketing analysis and one-page reports

0:00 Marketing analysis and reporting

How many marketing reports have you sent that nobody really read? Twelve slides of screenshots, a chart showing engagement is up, and no clear answer to the only question the reader cares about: so what should we do? AI can change that. It can turn messy exports into clear insights and a one page report in about thirty minutes, if you give it clean data, clear questions and a way to check. In this lecture you'll learn a four step analysis method, an important caution about attribution, and a monthly report template. You'll see two examples and watch me analyze a campaign export live.

0:45 Why it matters

Why does this matter? Because most marketers have more data than time. Ad platforms, email tools, social analytics, web analytics and your CRM all produce numbers, and pulling them into something meaningful by hand takes hours. So reports become screenshots with a comment. But screenshots aren't insight. A business owner or manager needs a clear story: what happened, why it probably happened, and what we should do next. AI is excellent at the reading and summarizing part. You bring the judgment and the checks.

1:22 The doctor analogy

Think of it like a good doctor. First, they check the vital signs, the basics, before jumping to conclusions. Then they ask focused questions, one at a time. Then they form a diagnosis, but a careful doctor treats it as a working hypothesis until a test confirms it. And they explain it in plain language to the patient. Your analysis should work the same way. Inspect the data first. Ask one question at a time. Explain what happened, and label the reasons as hypotheses. Then turn the most important hypothesis into a test.

2:02 The four steps

Here are the four steps. Step one, inspect. Before any analysis, ask for the date range, columns, totals for spend and conversions, missing values, duplicates, and anything odd, like conversions with zero spend. Step two, one question at a time: which campaigns had the lowest cost per conversion in the last thirty days, compared with the previous thirty? Ask it to show the calculation for one campaign. Step three, explain for a human: the three most important changes, what happened, the likely reasons labeled as hypotheses, and what to test next. Step four, check: ask for three totals you can verify against the dashboard, and recompute at least one yourself.

2:50 Attribution caution

Now an important caution. Different platforms count conversions differently. They use different attribution windows, some include people who only saw an ad, and some use modeled conversions. So the conversions reported by Google Ads, Meta and your email tool usually overlap, and adding them together will often give you more conversions than you actually had sales. When you ask AI to combine exports, tell it to keep platforms separate and flag possible double counting, and compare against your own sales data. For questions about which channel really caused the sales, you need incrementality testing, which the performance marketing course covers.

3:33 Example 1: the email export

A simple example. A marketer exports ninety days of email campaign data and asks for the open rate trend. Before analyzing, she runs the inspection step. It reports that one week appears twice, probably from exporting overlapping date ranges. Without that step, the duplicate would have inflated the totals and made one campaign look like a star. She removes it, reruns the question, and the story changes: the real winner is a different subject line style. Thirty seconds of inspection, and a wrong conclusion avoided.

4:10 Example 2: Sara's monthly report

Now a realistic business case, with illustrative details. Sara is a freelance marketer in Abu Dhabi who manages ads and email for a skincare retailer. Her monthly report used to take half a day of exporting, charting and writing. Now she runs the four steps on the exports, verifies three totals against the dashboards, and drafts with the report template, using last month's report as the format example. One month, the explain step suggests that a drop in ad performance lines up with the end of a promotion. That's a hypothesis, so she checks the promotion calendar, and it's confirmed. The report says so, and proposes a test for next month. Her client says these are the first reports she actually reads.

5:03 Watch me do it

Let me analyze an export. I've attached ninety days of ad data by campaign and week, with no customer level data. Step one, inspect. It reports the date range, totals for spend and conversions, and one oddity: a campaign with conversions but zero spend in one week. That's probably a reporting delay. I note it. Step two: which campaigns had the lowest cost per conversion in the last thirty days versus the previous thirty? Show the calculation for one. There's the table and a chart, and the calculation for the top campaign looks right. Step three: explain the three biggest changes for a business owner, with hypotheses and a test for each. Step four: three totals to verify. I open the dashboard and check total spend. It matches. Now I'd trust this enough to write the report.

6:03 The one-page monthly report

Here's the report template. A headline: one sentence on what happened and why it matters. A KPI table against last month and the target, with numbers exactly as in the data. What worked and what didn't, three bullets each, with evidence. Hypotheses and next month's tests. And decisions needed from the client or manager. Ask the AI to write check next to any number it had to calculate, and to avoid filler sentences. That last section, decisions needed, is what turns a report from a record into a tool. If a report doesn't lead to a decision or a test within a week, it probably didn't need to be written.

6:51 Common mistakes

The common mistakes. Analyzing before inspecting, which lets duplicates and oddities distort everything. Adding up conversions across platforms as if they were one source of truth. Reports that describe numbers without saying what to do. Sharing charts nobody has checked. And uploading customer level exports with names or emails when aggregated campaign data would do. Use aggregated data, keep your raw exports so every number can be traced, and treat every explanation as a hypothesis until it's tested.

7:25 Recap and try this now

Let's recap. Inspect exports first, ask one question at a time, explain the findings for a human with hypotheses labeled, and verify three totals. Don't add up conversions across platforms. And write one page reports that end with decisions and tests. Here's your try this now. Export sixty to ninety days of data from one platform this week, remove any customer level personal data, and run the four steps in the lesson text. Verify three totals. Then draft your next monthly report with the template, send it to one stakeholder, and ask them: what would you decide based on this?