Sean Ellis
Growth advisor, coiner of "growth hacking", author of Hacking Growth, former head of growth at Dropbox and Eventbrite
Where to hear and read about Sean Ellis1 episode or test
Sean Ellis coined the term growth hacking, led growth at Dropbox and Eventbrite, and wrote Hacking Growth. The most useful thing to learn from him is to find the users who would be very disappointed to lose your product, then fix activation, engagement and monetization before chasing acquisition.
Customer acquisition is so hard that if you're not really efficient at converting and retaining and monetizing people, you're going to really struggle.
The original growth hacker reveals his secrets | Sean Ellis (author of “Hacking Growth”)
Sean Ellis: the story and the lessons
1Who Sean Ellis is and what he has built
Ellis coined 'growth hacking' to name a different way of doing marketing. He saw Facebook, LinkedIn and Twitter treating growth as scrutiny of everything for its impact, not as textbook awareness-building, and named it with a couple of friends.
The term was divisive from day one. He wouldn't pick another, since something divisive is less likely to go unnoticed.
He has led growth at several well-known startups. He was on the founding team of LogMeIn, which sold for over $4 billion, and was head of growth at Dropbox and Eventbrite. He has also helped Microsoft and Nubank refine their growth strategy.
He invented the ICE prioritization framework and was an early user of freemium. Hacking Growth has sold over 750,000 copies.
His blog writing seeded TikTok's early growth work. TikTok's founding growth team in Singapore told him their early work was based on his blogging, before the book came out.
He had helped many unicorns but none of the very biggest companies, so it felt good to hear.
2The Sean Ellis test for product-market fit
One question separates must-have users from everyone else. He asks 'how would you feel if you could no longer use this product?' with the choices very disappointed, somewhat disappointed, not disappointed, or already stopped using it.
The goal is to find the 'very disappointed' group, who are the vein to dig into, and ideally learn who they are and how many there are.
The question began as a fix for lukewarm feedback. At Xobni, a satisfaction question would have given lukewarm answers from senior management, and flipping it to losing the product gave honest ones.
He reused it at Dropbox and every company after. The 40% line was not in the first version, since the question began as a filter for feedback from users who truly cared.
Ellis surveys only people who have really used the product. He samples randomly from users who reached the activation moment, used it at least twice, and used it in the last week or two.
Visitors, signups and demo viewers are excluded. When testing a new onboarding, he surveys only people who went through it. He wanted at least 30 responses, a hunch others later confirmed as a standard floor.
The 40% target is a shared signal, not a pass mark. The line came from watching many startups. Its main use is giving the team one target for when to start growing aggressively.
He bends it by culture: Nubank uses 50% because Brazilians skew optimistic, and 30% may be enough in Hungary. Nubank also holds each feature launch until 50% say they'd be very disappointed. 'We're not going to aggressively start to grow until we hit this target.'
He reads high scores against switching costs. Webs.com, a legacy website builder facing Wix and Weebly, scored about 90% very disappointed because users had invested heavily in their sites. Eventbrite scored the second highest he had seen.
The signals still guided growth: Webs had been flat for a year, then grew significantly over the next 12 months. He sees the score as a function of both switching costs and utility.
The survey keeps its value after retention cohorts arrive. The survey is a day-one leading indicator that needs no analytics system. Cohorts confirm retention later but never explain why.
If must-have users start churning once you grow, he says to retrench and rethink product-market fit. The test doesn't fit one-off products like a movie or workshop, where he uses NPS.
3How Ellis turns must-have users into growth
Lookout went from 7% to 40% in two weeks. The product bundled backup, find-my-phone, a firewall and antivirus, but most of the 7% cared about antivirus. Lookout repositioned on antivirus and made the first onboarding step set it up and say 'you're now protected from viruses.'
It reached 60% six months later. In his words, it 'set the right expectations and then speed to value.' Lookout became an early unicorn.
He asks why the benefit matters to get the ad hook. He asks an open question on the primary benefit, then four benefit statements to a different group, then 'why is that benefit important to you?'
At Xobni the benefit was finding things faster in email and the reason was 'I'm drowning in email.' Testing 'drowning in email?' as the hook worked well.
Ellis ignores somewhat disappointed users, or mines only those who share the core benefit. These users call the product a nice-to-have, and tuning to their feedback can dilute it. 'It becomes kind of good for everyone but not great for anyone.'
Superhuman instead looked at nice-to-have users who wanted the same benefit as must-have users and asked what they still needed. Such users usually have a commodity use case and an easy alternative.
He prefers a passionate niche over a broad lukewarm base. He segments results by use case. If one group is 60% very disappointed and a larger one sits at 35%, choosing between them is a strategic call that depends on how much money has been raised.
His view: 'Your biggest competition when you're really innovating is just being irrelevant.' He has never seen the percentage fade, and says companies that fail with it have an execution problem.
4How Ellis fixed activation at LogMeIn
A three-month roadmap freeze lifted signup-to-usage from 5% to 50%. Only 5% of signups ever ran a remote control session, and paid channels capped at $10,000 a month. The CEO froze the product roadmap and the team focused on getting signups to usage.
The same channels then scaled to $1 million a month, with a three-month payback and 80% of new users coming by word of mouth.
He emailed people who quit at the download step. A cheap channel sent 200,000 people a day and about 20,000 registered, but 18,000 never downloaded. More than 10 tests had failed, so the team emailed them in a note that looked like it came from customer service.
They said it seemed too good to be true for free. The winning test offered a choice of a paid trial or the free version, with a big check mark by the free one, and downloads rose 300%.
He defines activation as an early experience, then checks it against retention. He starts with two or three qualitative guesses of when someone has had a good enough taste, then looks for correlation with long-term retention. At LogMeIn, no remote control session meant no use.
He avoids goals like 'used it 100 times' because they correlate with retention but come too late to act on. There may be two or three valid aha moments.
He frames the problem before trying tactics. He quotes Charles Kettering: a problem well stated is a problem half solved. At the download step the two levers were more desire and less friction.
The team also studied instant messengers' install flows, and the result was many small gains, not one big one.
5Sean Ellis's order of work and how he picks channels
Ellis fixes activation first and acquisition last. His order is activation, then engagement and referral, then revenue, and only then channels. Acquisition is too competitive to win with an inefficient funnel.
Activation often leaks because it falls between product, which is focused on the roadmap, and marketing, which is focused on bringing people in. Collecting emails only means reacquiring users later.
He wants two or three channel hypotheses before joining a company. He says he is not confident of finding channels if he starts with none, and expects a few more to appear later.
The right channel comes from what is unique about the business, and it is often a mix. Bounce uses SEO for people searching 'luggage storage Paris' and signs at about 10,000 partner locations for people dragging bags past.
The same lens gave opposite channels at LogMeIn and Dropbox. LogMeIn grew almost entirely on paid search because a competitor, GoToMyPC, was creating the category with radio and TV and spending tens of millions a month. At Dropbox nobody was searching, so user-get-user loops made more sense.
The difference is demand generation versus demand harvesting. LogMeIn harvested GoToMyPC's demand with a free version.
A VC pushed him to talk to customers, and his tests improved. He spent his first five years measuring and testing everything and told LogMeIn's lead VC, 'I don't care what they say, I care what they do.' To appease the VC he began one customer conversation a day.
His experiments soon improved. He also asks customers how they found the product and how they usually find products like it, which gives channel ideas.
6Referral, freemium and North Star metrics according to Ellis
Referral incentives only speed up what already works. Dropbox had strong word of mouth before its two-sided incentive program. Ellis says it is a great accelerant when it's already working but can't fix a product people don't talk about.
At LogMeIn he skipped incentives for fear of breaking an engine where 80% of new users came by word of mouth. He saw the two-sided design work first at Zabni, from Jamie Simonoff's earlier tests. An intern, Albert Knee, built the Dropbox program and stayed on as his growth partner.
He says freemium needs two great products. The free version must be good enough to spark word of mouth, and the premium version differentiated enough to earn upgrades.
Teams worry about the upgrade, make the free version weak, then wonder why word of mouth is thin. He was among the first to use freemium and took a while to work out how it functions.
He picks a North Star that counts units of customer value. He starts from the must-have value named by the most passionate users and chooses a metric that is not a ratio and can rise over time. Examples are Amazon monthly purchases, Airbnb nights booked, Uber weekly rides and Eventbrite weekly tickets.
Revenue should follow value, not guide daily action. He caps the team discussion at about 30 minutes: 'If you give a team 30 days, they'll take 30 days.'
A metric can stretch a product's natural usage cycle. Facebook moved its North Star from monthly to daily active users and became more addictive. A monthly metric credited one visit even if a user came 10 times.
After reaching 1 billion monthly users, they chased 1 billion daily.
7Running growth experiments across teams
Ellis uses ICE so ideas from anyone can be compared fairly. A systematic score lets you explain why an idea was or wasn't chosen, which keeps contributors engaged in a high-velocity testing program.
He sees RICE's reach as unnecessary because reach is already part of impact, and admits he may be protective of his own idea. More tests beat fewer, and better tests beat bad ones.
He says data-driven acquisition no longer wins by itself. Early on, heavy testing plus some creativity beat rivals focused on CPM. Now most online marketers test a lot, so a company must be efficient at converting, retaining and monetizing too.
His summary: things are just getting harder. The hard part is getting marketing, product, sales and customer success to test together, and few later-stage companies have managed to copy it.
He expects AI to take ego out of cross-team decisions. A system that flags an underperforming area and suggests experiments is easier for product and marketing teams to accept than direction from each other. Analysis becomes the next bottleneck AI can relieve.
He already asks ChatGPT how Sean Ellis would answer an advice request, then edits the draft, since his book and writing are indexed. He reviews every answer.
8How Ellis works and what he values
He joins companies just after product-market fit. He works full-time as part of the team for three to six months, about once every year or two.
He prefers to start at around 40% or a bit less, because these systems are hard to build retroactively.
He offered refunds to protect a reputation he valued at $5 million. Two early interim clients seemed unhappy, so he offered full refunds. He valued his reputation at about $5 million, so $20,000 was not worth risking it.
One client took the refund and the other declined. The VCs who had introduced both gave him his first two term sheets, and his company's pre-money valuation ended up at more than double the figure he had put on his reputation.
He says weak answers usually come from skipping the obvious question. With Oleg Yakubenkov at GoPractice.io he teaches that data science is asking the right questions. In workshops the obvious ones are why users aren't downloading, and who considers the product a must-have.
He notes BlackBerry as the limit of the must-have test: users would have been very disappointed to lose the keyboard, until the iPhone made it unimportant.
The original growth hacker reveals his secrets | Sean Ellis (author of “Hacking Growth”)
Sean Ellis explains how to use the product-market fit survey to find your must-have users and then grow in the right order: activation, engagement, referral, revenue, and only then acquisition. His case studies include Lookout going from 7% to 40% in two weeks and LogMeIn lifting signup-to-usage from 5% to 50%, which took paid channels from $10,000 to $1 million a month.
Ellis coined 'growth hacking' to name a different way of doing marketing. He saw Facebook, LinkedIn and Twitter treating growth as scrutiny of everything for its impact, not as textbook awareness-building, and named it with a couple of friends.
The term was divisive from day one. He wouldn't pick another, since something divisive is less likely to go unnoticed.
He has led growth at several well-known startups. He was on the founding team of LogMeIn, which sold for over $4 billion, and was head of growth at Dropbox and Eventbrite. He has also helped Microsoft and Nubank refine their growth strategy.
He invented the ICE prioritization framework and was an early user of freemium. Hacking Growth has sold over 750,000 copies.
His blog writing seeded TikTok's early growth work. TikTok's founding growth team in Singapore told him their early work was based on his blogging, before the book came out.
He had helped many unicorns but none of the very biggest companies, so it felt good to hear.
One question separates must-have users from everyone else. He asks 'how would you feel if you could no longer use this product?' with the choices very disappointed, somewhat disappointed, not disappointed, or already stopped using it.
The goal is to find the 'very disappointed' group, who are the vein to dig into, and ideally learn who they are and how many there are.
The question began as a fix for lukewarm feedback. At Xobni, a satisfaction question would have given lukewarm answers from senior management, and flipping it to losing the product gave honest ones.
He reused it at Dropbox and every company after. The 40% line was not in the first version, since the question began as a filter for feedback from users who truly cared.
Ellis surveys only people who have really used the product. He samples randomly from users who reached the activation moment, used it at least twice, and used it in the last week or two.
Visitors, signups and demo viewers are excluded. When testing a new onboarding, he surveys only people who went through it. He wanted at least 30 responses, a hunch others later confirmed as a standard floor.
The 40% target is a shared signal, not a pass mark. The line came from watching many startups. Its main use is giving the team one target for when to start growing aggressively.
He bends it by culture: Nubank uses 50% because Brazilians skew optimistic, and 30% may be enough in Hungary. Nubank also holds each feature launch until 50% say they'd be very disappointed. 'We're not going to aggressively start to grow until we hit this target.'
He reads high scores against switching costs. Webs.com, a legacy website builder facing Wix and Weebly, scored about 90% very disappointed because users had invested heavily in their sites. Eventbrite scored the second highest he had seen.
The signals still guided growth: Webs had been flat for a year, then grew significantly over the next 12 months. He sees the score as a function of both switching costs and utility.
The survey keeps its value after retention cohorts arrive. The survey is a day-one leading indicator that needs no analytics system. Cohorts confirm retention later but never explain why.
If must-have users start churning once you grow, he says to retrench and rethink product-market fit. The test doesn't fit one-off products like a movie or workshop, where he uses NPS.
Lookout went from 7% to 40% in two weeks. The product bundled backup, find-my-phone, a firewall and antivirus, but most of the 7% cared about antivirus. Lookout repositioned on antivirus and made the first onboarding step set it up and say 'you're now protected from viruses.'
It reached 60% six months later. In his words, it 'set the right expectations and then speed to value.' Lookout became an early unicorn.
He asks why the benefit matters to get the ad hook. He asks an open question on the primary benefit, then four benefit statements to a different group, then 'why is that benefit important to you?'
At Xobni the benefit was finding things faster in email and the reason was 'I'm drowning in email.' Testing 'drowning in email?' as the hook worked well.
Ellis ignores somewhat disappointed users, or mines only those who share the core benefit. These users call the product a nice-to-have, and tuning to their feedback can dilute it. 'It becomes kind of good for everyone but not great for anyone.'
Superhuman instead looked at nice-to-have users who wanted the same benefit as must-have users and asked what they still needed. Such users usually have a commodity use case and an easy alternative.
He prefers a passionate niche over a broad lukewarm base. He segments results by use case. If one group is 60% very disappointed and a larger one sits at 35%, choosing between them is a strategic call that depends on how much money has been raised.
His view: 'Your biggest competition when you're really innovating is just being irrelevant.' He has never seen the percentage fade, and says companies that fail with it have an execution problem.
A three-month roadmap freeze lifted signup-to-usage from 5% to 50%. Only 5% of signups ever ran a remote control session, and paid channels capped at $10,000 a month. The CEO froze the product roadmap and the team focused on getting signups to usage.
The same channels then scaled to $1 million a month, with a three-month payback and 80% of new users coming by word of mouth.
He emailed people who quit at the download step. A cheap channel sent 200,000 people a day and about 20,000 registered, but 18,000 never downloaded. More than 10 tests had failed, so the team emailed them in a note that looked like it came from customer service.
They said it seemed too good to be true for free. The winning test offered a choice of a paid trial or the free version, with a big check mark by the free one, and downloads rose 300%.
He defines activation as an early experience, then checks it against retention. He starts with two or three qualitative guesses of when someone has had a good enough taste, then looks for correlation with long-term retention. At LogMeIn, no remote control session meant no use.
He avoids goals like 'used it 100 times' because they correlate with retention but come too late to act on. There may be two or three valid aha moments.
He frames the problem before trying tactics. He quotes Charles Kettering: a problem well stated is a problem half solved. At the download step the two levers were more desire and less friction.
The team also studied instant messengers' install flows, and the result was many small gains, not one big one.
Ellis fixes activation first and acquisition last. His order is activation, then engagement and referral, then revenue, and only then channels. Acquisition is too competitive to win with an inefficient funnel.
Activation often leaks because it falls between product, which is focused on the roadmap, and marketing, which is focused on bringing people in. Collecting emails only means reacquiring users later.
He wants two or three channel hypotheses before joining a company. He says he is not confident of finding channels if he starts with none, and expects a few more to appear later.
The right channel comes from what is unique about the business, and it is often a mix. Bounce uses SEO for people searching 'luggage storage Paris' and signs at about 10,000 partner locations for people dragging bags past.
The same lens gave opposite channels at LogMeIn and Dropbox. LogMeIn grew almost entirely on paid search because a competitor, GoToMyPC, was creating the category with radio and TV and spending tens of millions a month. At Dropbox nobody was searching, so user-get-user loops made more sense.
The difference is demand generation versus demand harvesting. LogMeIn harvested GoToMyPC's demand with a free version.
A VC pushed him to talk to customers, and his tests improved. He spent his first five years measuring and testing everything and told LogMeIn's lead VC, 'I don't care what they say, I care what they do.' To appease the VC he began one customer conversation a day.
His experiments soon improved. He also asks customers how they found the product and how they usually find products like it, which gives channel ideas.
Referral incentives only speed up what already works. Dropbox had strong word of mouth before its two-sided incentive program. Ellis says it is a great accelerant when it's already working but can't fix a product people don't talk about.
At LogMeIn he skipped incentives for fear of breaking an engine where 80% of new users came by word of mouth. He saw the two-sided design work first at Zabni, from Jamie Simonoff's earlier tests. An intern, Albert Knee, built the Dropbox program and stayed on as his growth partner.
He says freemium needs two great products. The free version must be good enough to spark word of mouth, and the premium version differentiated enough to earn upgrades.
Teams worry about the upgrade, make the free version weak, then wonder why word of mouth is thin. He was among the first to use freemium and took a while to work out how it functions.
He picks a North Star that counts units of customer value. He starts from the must-have value named by the most passionate users and chooses a metric that is not a ratio and can rise over time. Examples are Amazon monthly purchases, Airbnb nights booked, Uber weekly rides and Eventbrite weekly tickets.
Revenue should follow value, not guide daily action. He caps the team discussion at about 30 minutes: 'If you give a team 30 days, they'll take 30 days.'
A metric can stretch a product's natural usage cycle. Facebook moved its North Star from monthly to daily active users and became more addictive. A monthly metric credited one visit even if a user came 10 times.
After reaching 1 billion monthly users, they chased 1 billion daily.
Ellis uses ICE so ideas from anyone can be compared fairly. A systematic score lets you explain why an idea was or wasn't chosen, which keeps contributors engaged in a high-velocity testing program.
He sees RICE's reach as unnecessary because reach is already part of impact, and admits he may be protective of his own idea. More tests beat fewer, and better tests beat bad ones.
He says data-driven acquisition no longer wins by itself. Early on, heavy testing plus some creativity beat rivals focused on CPM. Now most online marketers test a lot, so a company must be efficient at converting, retaining and monetizing too.
His summary: things are just getting harder. The hard part is getting marketing, product, sales and customer success to test together, and few later-stage companies have managed to copy it.
He expects AI to take ego out of cross-team decisions. A system that flags an underperforming area and suggests experiments is easier for product and marketing teams to accept than direction from each other. Analysis becomes the next bottleneck AI can relieve.
He already asks ChatGPT how Sean Ellis would answer an advice request, then edits the draft, since his book and writing are indexed. He reviews every answer.
He joins companies just after product-market fit. He works full-time as part of the team for three to six months, about once every year or two.
He prefers to start at around 40% or a bit less, because these systems are hard to build retroactively.
He offered refunds to protect a reputation he valued at $5 million. Two early interim clients seemed unhappy, so he offered full refunds. He valued his reputation at about $5 million, so $20,000 was not worth risking it.
One client took the refund and the other declined. The VCs who had introduced both gave him his first two term sheets, and his company's pre-money valuation ended up at more than double the figure he had put on his reputation.
He says weak answers usually come from skipping the obvious question. With Oleg Yakubenkov at GoPractice.io he teaches that data science is asking the right questions. In workshops the obvious ones are why users aren't downloading, and who considers the product a must-have.
He notes BlackBerry as the limit of the must-have test: users would have been very disappointed to lose the keyboard, until the iPhone made it unimportant.