Choose your first growth channel the way you would choose anything with a cost attached: by testing a small number of options cheaply, keeping the one that shows real signal, and cutting the rest without sentiment. The word itself is doing work here. To choose, as the Cambridge Dictionary puts it, is to select from several options — and the selection only means something if the options were real and the test was honest.
The operating problem is simple. Early-stage companies rarely die because no channel could have worked. They die because three channels were tried at once, each underfunded, each judged on vibes, and the runway was gone before any of them got a fair hearing. This guide gives you a framework for running that selection as arithmetic instead of instinct.
One caveat up front: this is general business information, not financial advice, and every company's constraint set is different. A services business with 90-day sales cycles and a self-serve software product should not reach the same answer.
What has to be true before any channel can work?
Nothing downstream matters until the product retains the customers a channel brings in. A channel is a pipe. If the pipe feeds a leaky tank, more water is just a faster way to watch the level drop. Before you spend a dollar or an hour on acquisition, you need evidence that customers who arrive actually stay and pay long enough to matter.
You do not need certainty. You need a direction of travel. If you cannot yet point to any measurable sign that the product is landing — usage, repeat behaviour, an unprompted second purchase — fix that first. Our earlier piece on the measurable signs of product-market fit covers what to look for before you scale anything. And if churn is the leak, the first ninety days of the customer relationship are usually where it lives; we have a separate guide on fixing that window. This connects to our earlier piece, The 7 Measurable Signs of Product-Market Fit (Before You Scale Anything).
Then stop tinkering and start testing. A channel test run on a product that loses its customers is not a channel test. It measures nothing.
Which channels should be on the shortlist?
Most early-stage companies face the same menu. The honest version of the list looks like this:
- Outbound to a named list — cold email or direct outreach to people you can describe precisely.
- Paid acquisition — search or social ads where you buy intent or attention at a posted price.
- Content and search — writing that compounds slowly and costs time rather than cash.
- Referrals — a structured way for happy customers to bring the next one.
- Communities and partnerships — showing up where your buyers already gather.
The shortlist should be three items, not five. Three is enough to have a real choice; five guarantees each test is starved. Cut using two filters. First, precision: can you actually name who buys, or is your target "everyone with a budget"? If you cannot name them, outbound and communities beat broad paid spend. Second, cash: if money is the binding constraint, channels paid in time (content, referrals, partnerships) rank ahead of channels paid in cash.
For a deeper read on the paid option, see our bootstrapper's guide to the first paid campaign. For the slow-burning option, the piece on SEO as compounding rent explains why it rewards patience and punishes impatience in equal measure.
How do you run a channel test that proves something?
The failure mode is not testing badly. It is testing without a kill criterion, so every channel survives long enough to drain the account. Run the test as a process with fixed steps:
- Define the question. Not "does marketing work?" but "can this channel deliver a customer for less than the customer is worth to us, at our current conversion rates?"
- Set a budget in advance — money, time, or both. Write it down. A test with no ceiling is not a test; it is a habit.
- Set a decision date in advance. Ninety days is a common working horizon for a first test, because it is long enough for a slow channel to show movement and short enough to survive a mistake. Adjust to your own sales cycle: a channel with a 90-day deal cycle cannot be judged in 30.
- Instrument it. You need to know where each new customer came from. If you cannot attribute a single customer to the channel, the test returns no data — which is itself a result.
- Decide on the date. Continue, double down, or kill. No extensions granted on feeling.
One test per channel. Don't change the offer, the audience, and the creative all in week three, because then you will not know what moved. Change one variable at a time, or accept that you are learning, not measuring.
Track the results on a single screen rather than a sprawl of dashboards. Our guide to the five-number growth dashboard covers the minimum set worth watching.
What does "it worked" actually mean?
A channel works when the arithmetic closes. That means the cost to acquire a customer sits comfortably below what that customer returns over a sensible period, and the channel can scale without the economics collapsing. Two traps hide in that sentence.
The first trap is judging a channel on cheap clicks or sign-ups rather than paying customers. Vanity volume is the most common way a bad channel looks good for a quarter. The second trap is scaling a channel that only works at small size. Some channels have a fixed pool of demand; the first hundred customers come cheap and the next thousand do not exist. Ask what has to be true for the channel to work at ten times the spend. If the answer is "the market gets bigger", the channel does not scale.
Also distinguish a channel from a loop. A channel is a pipe you pay to keep water flowing through. A loop is a structure where customers create customers — referrals being the classic case. When a loop works, growth compounds; when only a channel works, growth is a treadmill you keep paying to run. We unpack the difference in Growth Loops vs. Funnels, and the referral-specific design choices in our referral program guide.
What this means for your next ninety days
Pick three channels. Write down, before you start, what each must show and by when. Spend no more than you wrote down. Kill without ceremony anything that misses its own bar, and put the freed budget behind the one survivor.
Our analysis of why founders get this wrong is blunt: the channel decision feels like a strategy decision, so it gets debated; the test itself feels like execution, so it gets improvised. Reverse that. The choice deserves ten minutes of filtering. The test deserves a written plan.
And keep the receipts. The post-mortem of a failed channel test — what you assumed, what the data said, what you'd do differently — is among the most useful documents a company produces. The same is true of a failed company, which is a cheaper lesson to borrow than to live.
Finally, tie the channel decision back to the funding decision. A channel that needs six months and real spend before it proves itself has implications for how much money you need and when you raise. The arithmetic that connects those choices is covered in The 4 Numbers That Decide Whether You Raise or Bootstrap.
