Analytics, Attribution and Experiments

Introduction

"Which marketing is working?" is hard to answer. People see many things before they act, tools cannot see everything, and results fluctuate. This lesson helps you measure carefully and stay honest about uncertainty. It also builds on the Data Basics course.

A measurement plan

Start with questions, not tools.

1. Business goal: more bookings, more sales, more subscribers.
2. Marketing goal: a number you can influence, such as sign-ups per month.
3. Events to track: the specific actions that show progress, such as a form submission, a click on "book now" or a purchase.
4. Sources: how to tell where each visit came from.
5. Review rhythm: when you will look, and who decides what to change.

Free tools such as Google Analytics, Search Console and your email service's reports cover most needs. Choose one place as the "source of truth" for each number.

Tagged links

Add UTM parameters to links you control, such as newsletter links and posts:

?utm_source=newsletter&utm_medium=email&utm_campaign=autumn-guide

Use consistent, lowercase names and keep a naming sheet. Without tags, many visits appear as "direct" or "unknown".

The limits of attribution

Attribution assigns credit for a result to the touchpoints before it. Models include last click (all credit to the final step), first click, and various spread-out versions.

Reality is messy:

• People use several devices and may be counted twice.
• Privacy tools, consent choices and blockers hide some activity.
• Much influence is invisible: a recommendation from a friend, a podcast mention, seeing your name several times.
• Search, social and email interact. Email may close a sale that social started.

So no attribution model is "right". Use them as rough guides, compare several, and add simple questions such as "How did you hear about us?" on forms and in conversations.

Funnel numbers

Measure each stage so you can see where people drop out: visitors, sign-ups, enquiries, purchases. A change that doubles traffic but not sales may point to a problem later in the funnel. Compare rates, not just totals.

Experiments

An experiment compares two versions at the same time, changing one element.

1. Hypothesis: "A shorter form will increase sign-ups."
2. One change between version A and version B.
3. Split visitors randomly, or compare equal time periods when you cannot.
4. Decide in advance how long it will run and the minimum number of visitors.
5. Compare the rates.
6. Act and record what you learned.

Common mistakes:

• Stopping early when one version happens to look better
• Too few visitors, so chance explains the difference
• Changing several things at once
• Ignoring seasonality or other events
• Testing endlessly small changes instead of big ideas
• Reporting only wins

With low traffic, you may only be able to detect large effects. That is fine. Test bold changes, and combine results with judgement.

Statistics sense

• A difference can arise from chance. More data reduces that risk.
• Look at the size of the difference as well as whether it is "significant".
• Be cautious with percentages from tiny numbers.
• Correlation is not causation.

Privacy in measurement

Respect consent rules, minimise personal data, tell people what you track, and keep records secure. Where laws require consent for analytics cookies, configure your tools accordingly.

Try it yourself

Write a one-page measurement plan for a project: goal, three events, how you will tag links, and one experiment with a hypothesis, the single change and how you will decide the result.

Summary

• Plan measurement from goals, with defined events and tagged links.
• Attribution is a rough guide, so combine tools with direct questions to customers.
• Run controlled one-change experiments with enough data, and avoid stopping early or reporting only wins.