Conversion, funnels, cohorts and retention

Conversion is the share who did the wanted action out of those who could. A funnel shows at which step people are lost, and cohorts show whether they come back.

Not how many came, but what came of it

In the last lessons you installed a counter and studied bounces. But the number of visitors and page views tells you how many people passed by, not what they did. The project exists for the main action from the brief, so the main question is: what share of those who came did that action, and where were the rest lost? The answer comes from conversion, the funnel and, for those who return, retention.

Conversion

Conversion is the share of people who did the wanted action out of those who could:

conversion = those who did it ÷ those who could

An example: in a month 1000 people came to the page with the form, 30 sent a request, conversion 3%. Everything important hides in the words “out of those who could” (the denominator). It can be the visitors to the whole site, the visitors to this page, the visits or the people from one source. If you don’t say who it was counted out of, the number is unclear. So write it down: “3% of visitors to the page with the form”.

Where the word comes from. “Conversion” in marketing means “turning”: a visitor is turned into a client, a subscriber, an inquirer. The idea is old, from direct marketing, where response rates were counted long before the internet.

A few rules:

1. Count on the main action from the brief. A “conversion” of page views into page views is no use.
2. Look at conversion by source separately. People from search, from a newsletter and from ads behave differently, and an overall figure wipes out the difference.
3. Don’t confuse a growth in number with a growth in share. Advertising can bring people you don’t need: there will be more requests and conversion will fall. What is good in that is decided by the cost of a lead (the lesson on paid traffic).
4. On small numbers a share jumps about (three people out of fifty-six is 5.4%, and four people is 7.1%). Don’t draw conclusions from single digits.

Micro-conversions. The main action can be rare: an expert’s service gets few requests. Then micro-conversions help: small steps that bring a person nearer to the main one: tapped the phone or the email, reached the prices, downloaded a document, subscribed to a channel. There are many, they pile up faster, and they show where people are lost. But don’t make them an end in themselves: a micro-conversion shows the path, not the result (more in the next lesson).

The funnel

You know the funnel as a sketch from the lesson on the funnel. Here it becomes a table of numbers: what steps a person goes through and how many get through each.

A typical funnel for a page with a form: came, saw the offer, pressed the button, started filling in the form, sent it. GA4 and Yandex Metrica can build such funnels themselves: in GA4 it is a “funnel exploration” in the Explore section, and in Metrica a compound goal of successive steps.

You count the share that goes on at each step: the next step divided by the previous. The weakest step is where this percent is lowest, not where most people dropped out by number: at the start of a funnel many who just looked in always drop out.

Cautions:

• Fix the steps in advance. Otherwise you will pick steps to suit a convenient picture.
• A funnel shows the place of loss, not the cause. Look for the cause through the person’s eyes (heatmaps, recordings, asking people; the lesson on session recording is about that).
• People don’t go strictly by the steps. So the order of steps in funnels is sometimes made “open”, but that is a subtlety.

Open the diagram. At the top is the funnel of an invented site: move the sliders and see how the numbers and the result change. These are the same numbers as in the practice.

1. Came to the site 2000
   Go on: 40%
2. Saw the offer 800
   Go on: 30%
3. Pressed the button 240
   Go on: 50%
4. Started filling in the form 120
   Go on: 50%
5. Sent the form 60

Overall conversion: 3.0% (60 of 2000). The weakest step is “pressed the button”: only 30% of those at the previous step get there.

Cohorts by week

Cohort (week of the first action)	People	Week 1	Week 2	Week 3
6 October	200	70 (35%)	50 (25%)	40 (20%)
13 October	160	48 (30%)	32 (20%)	not reached yet
20 October	120	48 (40%)	not reached yet	not reached yet

Read the table by rows: each has its own cohort of people. The percentages are the share who came back out of the cohort’s size. The empty cells on the right are not zeros but “the time has not come”. The data is invented.

Cohorts and retention

Conversion and the funnel answer the question “what do people do in one visit.” Retention answers the question “do they come back.” That needs a new tool: the cohort.

A cohort is a group of people with a shared start: for example everyone who first sent a request or subscribed to a channel in the same week. For each cohort you count what share came back after one week, two, three. The result is a retention table:

• Rows: cohorts by the week of their start.
• Columns: “after 1 week”, “after 2 weeks” and so on.
• In a cell: the share of the cohort who came back by that time (did some specified action).
• At the lower right the table is “cut off”: for fresh cohorts the later weeks haven’t come yet. Those aren’t zeros but emptiness.

Why this matters: an overall “came back” percentage mixes old and new and hides change. Cohorts show whether those who came after your change are retained better than those who came before. If a new cohort is retained worse than earlier ones, you will see it at once.

What “came back” means for your project. For an app it is “uses it again.” For the site of an expert with a service bought rarely it doesn’t work like that (you saw this at the Retention stage in the lesson on AARRR): a person doesn’t visit daily. So “retention” there means something else: stayed in touch (didn’t unsubscribe, opens the emails, reads the channel), or came back before a request (almost all decisions are made over two or three visits), or got in touch again. Choosing the definition is part of the decision.

An example row for a cohort: 200 people first subscribed in the week of 6 October. The next week 70 came back: that is 35%. After two weeks, 50: 25%. After three, 40: 20%. The curve falls fast at the start and then levels off: a typical shape.

To count it yourself, once the data is in a table, you need the query language SQL. Here is what a query that counts how many people of a cohort came back in each week looks like (you ran such queries yourself in the lessons on SQL; here is only the idea):

SELECT cohort_week, weeks_after, COUNT(DISTINCT user_id) AS returned
FROM visits_by_user_week
GROUP BY cohort_week, weeks_after
ORDER BY cohort_week, weeks_after;

The query reads almost like a sentence: take from the table by weeks, group by cohort and week number, count the distinct people, sort.

Check yourself: what does an empty cell at the lower right of a retention table mean?

That the time hasn’t come yet: the cohort is too fresh to measure the later weeks. It is not a zero. A cell is a zero only if the period has passed and no one came back.

Practice

Work with the sample numbers.

1. The funnel. Name the steps in your own words: the first names the source, the last the main action from the brief in the same words. Enter the share that goes on at each step.
2. The totals. The overall conversion and the weakest step with its share.
3. Micro-conversions. At least two; at least one tied to a section of the site from the brief.
4. Cohorts. Enter the percentages of those who came back for each cell of the table.

In this practice the numbers are sample ones, and the check compares your calculations with the right ones. You will count your own retention and conversion when you have gathered data on your own site. Think which of the micro-conversions is nearest to your main action, and why.