Viral coefficient (k-factor): how to measure it

Theo

Every Halves group needs members, so every user is a little salesperson. Investors will ask us one thing: are we viral?

Alex Tutor

"Viral" has a number: the viral coefficient, k. It's how many new users each user brings: invitations per user times the share of invitations that turn into new users.

Alex Tutor

k = \frac{\text{invitations}}{\text{user}} \times \frac{\text{new users from invitations}}{\text{invitations}} = \frac{\text{new users from invitations}}{\text{users}}

You

Over what time, though? Someone can keep inviting people for years.

Alex Tutor

Fix a window, or older cohorts will always look more viral. Here: invitations sent in each user's first 30 days, and the invited people who sign up within them.

Alex Tutor

Dataset · 6 rows

Invites by sign-up cohort, January–June 2026

Everything counted within 30 days of each user's sign-up.

Download CSV

Columns
cohort: Sign-up month
new_users: Users who signed up that month
invites_sent_30d: Invitations the cohort sent in each user's first 30 days
invites_accepted_30d: Invitations accepted within those 30 days
accepted_by_existing_users: Accepts by people who already had a Halves account
invited_new_users_30d: Accepts that created a new account
median_days_to_invitee_signup: Median days from the inviter's sign-up to the invited person's sign-up

Data

cohort	new_users	invites_sent_30d	invites_accepted_30d	accepted_by_existing_users	invited_new_users_30d	median_days_to_invitee_signup
2026-01	4646	5737	1947	487	1460	6
2026-02	5106	6452	2075	519	1556	6
2026-03	5395	7018	2275	569	1706	6
2026-04	5504	7200	2406	602	1804	6
2026-05	5987	7290	2340	585	1755	6

Show 1 more row

2026-06

12904

25379

8793

1759

7034

3

Alex Tutor

Question 1

What is the viral coefficient k of the January 2026 cohort (new users from its invitations within 30 days, per cohort user)? Round to two decimals.

Your answer

Units: new users per user, e.g. 0.35

Theo

My dashboard says 0.42 for January: 1,947 accepted invites over 4,646 users.

Alex Tutor

Those accepts include 487 people who already had Halves and just joined one more group. Good for engagement, but it isn't growth.

You

So the numerator is new accounts created from invites: 1,460, not 1,947.

Alex Tutor

Yes. Growth counts new people. Everything else is activity.

Theo

Existing users joining more groups still matter to me. They keep groups alive.

Alex Tutor

They do, and they belong in an engagement metric next to k, not inside it.

Alex Tutor

Question 2

Theo's dashboard computes k for January as 1,947 ÷ 4,646 = 0.42. What's wrong with it?

Choose one answerNothing: every accepted invite adds a member to a groupIt should divide by invitations sent, not by usersIt counts 487 existing users who joined another group, which isn't growthIt should use a 7-day window instead of 30 days

Maya

0.31 is a long way from 1. So invites don't really matter for growth?

Alex Tutor

With k below 1, invites alone can't grow Halves: each generation is smaller than the last and the chain dies out. But everyone you bring in some other way pulls a chain of extra people behind them.

Alex Tutor

\text{total} = \text{seed} \times (1 + k + k^2 + \dots) = \frac{\text{seed}}{1 - k}

Valid for k below 1.

You

So May's ad budget bought more users than the ad reports show.

Alex Tutor

Let's see how many more. In May, paid social and store search ads brought 2,169 users. Pool the January–May cohorts for k, leaving June aside for now, and work out what those 2,169 turn into.

Alex Tutor

Question 3

Pool the January–May 2026 cohorts to get k. With that k, how many users in total do the 2,169 users that paid channels brought in May turn into, counting every generation of invites?

Your answer

Units: users

Alex Tutor

The second number is viral cycle time: how long from a user joining to the people they invite joining. The median here is 6 days, and 3 in June.

You

A shorter cycle doesn't change the total, it changes how fast it arrives?

Alex Tutor

For k below 1, yes: those 978 extra users arrive within weeks, not months. Above 1 the chain never dies out, and a shorter cycle makes it compound far faster.

Theo

So the ad money comes back faster. I like that part.

Alex Tutor

Question 4

With k below 1, halving the viral cycle time raises the total number of users that May's 2,169 paid users eventually bring.

TrueFalse

Theo

June's k is 0.55! That goes in the investor deck.

Alex Tutor

June is trip season: groups form for a holiday and pull in friends within 3 days. January to May give 0.29–0.33. Quoting June as Halves's k promises growth autumn won't deliver.

You

And some invited people would have installed Halves anyway, from the store.

Alex Tutor

That's the last honest step. To know how much invites add, hold them back for a small random share of new groups, say hide the invite prompt for 5%, and compare sign-ups.

Maya

And what would that holdout tell us that k doesn't?

Alex Tutor

How much of the 0.31 is truly extra. If half of those people would have come anyway, invites are worth half as much as k suggests.

Alex Tutor

Question 5

Which practices make a reported k honest? Select all that apply.

Choose all that applyA fixed window after sign-up, the same for every cohortCounting only accepts that create a new accountQuoting the best month so investors see the potentialReporting k by cohort and seasonHolding invites back for a small random group to see how many invited people would have come anyway

Maya

So what can we say to investors without embarrassing ourselves next quarter?

Alex Tutor

The measured k with its window and season, what it does to every other channel, and what we haven't measured yet.

Alex Tutor

Question 6Short answer · AI-checked task

Write the virality paragraph for the investor deck in 3–4 sentences: the k you'd quote and how it's measured, what it does for growth, and what's not yet known.

Write your answer and get a score with feedback from our AI reviewer.

Log in to get AI feedback

The rule: k counts new people within a fixed window, per cohort and season. Below 1 it's a multiplier, not an engine, and that's still worth a lot.

Alex Tutor

That’s the lesson. You answered every task — nicely done.

Next lesson: Channels and attribution