Growth hacking in 2027: what I think is actually coming
the honest version of a predictions post. three shifts I think are real, one I think is overhyped, and how you would know if I am wrong.
the short version: most predictions posts are marketing. this one commits to specific claims and tells you what would prove each of them wrong, because a prediction you cannot lose is not a prediction.
I write this as someone whose whole job is distribution. I scaled ZuAI from 10,000 to 2 million users at a $0.02 blended CAC, and everything below is shaped by that, including the biases.
First, the thing that will not change
channels decay. Andrew Chen called it the law of shitty clickthroughs, and it has held for every channel anyone has ever found: as more people use a channel, its response rate falls, forever.
that is the one safe prediction for 2027. whatever works brilliantly next year works less well the year after, not because you got worse but because it got crowded. every specific tactic below is temporary for that reason.
What I think is coming
one. buying becomes agent-mediated, and your funnel gets a new gatekeeper.
people are already asking an assistant what to use rather than searching and comparing. by 2027 I expect a meaningful share of software discovery to happen inside a model’s answer, not on a results page.
that changes the job. you are no longer optimising a page a human reads, you are trying to be the thing a model repeats. that rewards being clearly described, consistently described everywhere, and corroborated by sources the model trusts. it punishes anything that only makes sense visually.
how you would know I am wrong: if referral traffic from assistants stays a rounding error through 2027 while classic search holds steady.
two. creative volume stops being a differentiator.
testing 150 creatives a month was an advantage in 2024 because it was expensive and annoying. generation costs have collapsed, and by 2027 everyone will be able to produce more variants than they can evaluate.
when volume is free, the constraint moves to judgement: knowing what to test, and having a measurement setup that can tell a real winner from noise. the creative testing planner exists because most teams already run more variants than their budget can resolve, and that gap gets worse as volume gets cheaper.
how you would know I am wrong: if the teams winning in 2027 are simply the ones producing the most, with no measurement edge.
three. retention becomes the whole game for consumer AI.
acquisition costs rise, and AI apps have a specific problem: serving heavy users costs real money, so a badly retained user is not merely worthless, they are expensive. that pressure only increases.
I expect the successful consumer AI companies of 2027 to look boring from the outside. fewer channels, more work on the first session, pricing that reflects actual serving cost. the retention ceiling calculator makes the arithmetic obvious: at 8% month one retention, doubling your spend doubles your ceiling, while doubling retention does far more and costs nothing in media.
how you would know I am wrong: if a new cheap acquisition channel opens and stays cheap for more than a year.
What I think is overhyped
fully autonomous growth agents that run your campaigns end to end.
the forecasts point the right way and the money is real. Gartner-style projections have agentic spending overtaking chatbots and assistants by 2027, and the orchestration market tripling to over $30 billion. one widely reported forecast also expects more than 40% of agentic projects to be shelved by the end of 2027 on cost, unclear value and missing controls.
both of those can be true. the tooling gets much better and most attempts still fail, because growth decisions are mostly judgement calls about incomplete data, and the expensive mistakes are the ones nobody knew to check. I use automation heavily, including n8n for the repetitive parts, and I would not hand it the budget.
how you would know I am wrong: if a team you respect runs paid acquisition end to end on an agent for two consecutive quarters and holds CAC flat.
What I would actually do in 2027
the same as 2026, honestly. find one channel that works before hiring against it. run enough volume to learn something. measure retention before spending on acquisition. treat every channel as temporary, because it is.
that is unglamorous, and it is why predictions posts exist: the real answer does not change much, so people write about the future instead.
if you want the current-year version of this argument, with the evidence and the strongest case against it, read growth hacking is still real in 2026.
Sources: the channel decay principle is Andrew Chen’s law of shitty clickthroughs; agentic AI spending and the orchestration market forecasts are summarised by G2.