“Try content, outbound and paid” sounds diversified. For an early team it usually means three underpowered tests with different audiences, offers and feedback. Nothing runs long enough to teach the company why a customer arrived or whether the economics could ever work.
Channel-fit practice connects the product, market and model; systematic channel testing prevents favourite tactics from winning by instinct; and positioning determines whether the chosen audience can understand why it should change. Together they turn “try growth” into a falsifiable channel hypothesis. Products adapt to channels because founders do not control the channel's rules.
Choose one channel hypothesis
The first channel is the one whose riskiest assumption you can test.
01 · Audience
Find the behaviour
Where does this group already seek or share?
02 · Product
Match the motion
Can the experience fulfil the channel promise?
03 · Economics
Afford the route
Can price and margin support acquisition cost?
04 · Learning
Read the result
Will a small test change the channel belief?
Give each channel one job
Search captures existing intent. Founder-led outbound creates a conversation with a narrow identified group. Community borrows trust. Product loops turn use into new distribution. Paid media can scale a message whose conversion and economics are already legible. These jobs are different; judging all channels by cheap clicks hides the question they were meant to answer.
Also ask whether the channel can become a loop rather than a recurring purchase of attention. A useful output may create a referral, shared artefact, collaborator invitation or new supply that brings the next customer. The loop is valuable only when it is native to the product and the new participant can reach value; a forced invite is merely a funnel with extra friction.
- Reach: can the channel access the chosen customer in a context where the problem matters?
- Native action: what behaviour does the channel reward, and can the product produce it?
- Economics: can expected value per customer support the cost and time to acquire one?
- Evidence: what observable result would weaken the channel hypothesis?
Test the constraint before scaling the activity
Choose the riskiest link: audience availability, message response, conversion friction or payback. Run the smallest test that exposes it. Twenty carefully chosen conversations can teach more about an enterprise motion than thousands of untargeted impressions. A landing-page test may suit existing search intent. Preserve the audience, message, offer and outcome so the next test revises a named belief.
Use AI for channel memory, not channel spray
An agent can research a bounded audience, adapt an approved message to the channel and compare responses with the original hypothesis. It should not publish everywhere because generation is cheap. Volume without channel fit turns uncertain strategy into a larger reputational and measurement problem.
Give the agent one channel hypothesis to run
The agent needs a commercial hypothesis, not a request to “do growth”. Bind one audience, one channel job and one offer to a short evidence window. Its work is to carry the test from preparation to observed response without changing the strategy halfway through.
- Evidence pack: the accepted customer group and exclusions, customer language, current alternative, approved promise and proof, price and margin constraints, prior channel results, consent rules and exact source conversations.
- Working brief: write the hypothesis, channel job, target cohort, message, offer, primary outcome, evidence window, budget or time cap and the result that means continue, revise or stop.
- Permitted action: research the bounded cohort, prepare channel-native drafts and build the measurement sheet. Where an authorised provider supports it, the agent can execute the reviewed batch and collect replies and delivery evidence.
- Review gate: a person approves claims, audience, external sends, publication and spend. The agent cannot expand the cohort, change the offer or switch success metrics because early results look inconvenient.
- Learning record: keep the exact audience, message version, send or publication receipt, responses and outcome definition together. Ask the founder to accept, revise or reject the proposed belief change before the next test begins.
The agent gets better because it can compare like with like: the belief before the test, the action actually taken and the outcome that followed. It is learning the company's distribution system, not memorising whichever copy received the most attention.
A useful next move
Score three plausible channels from one to five on audience reach, native product fit, economic fit and speed of learning. Pick the channel whose weakest assumption can be tested this week. Define the stop rule before launching.
Continue with positioning before promotion, or read how to test whether the offer earns commitment.