Your First 100 Customers Don't Come From a Channel Strategy

The most common early-stage marketing mistake I see is a good marketing plan executed roughly a year too early.

A founder comes to me with a channel strategy. Paid social for awareness, search for intent, content for the long game, an email programme to nurture, sensible budget split, sensible targets. It is a competent plan. I have built plans like it for brands with real scale, and at scale they work.

For the first hundred customers it is close to useless, and worse, it is expensive.

What the first hundred are actually for

The first hundred customers are not a distribution problem. They are a learning problem.

You are not trying to find the cheapest way to acquire many customers. You are trying to find out which customers get enough value to stay, and what language makes them understand that before they buy. Until you know that, scaling a channel scales a question mark.

This is why the channel-first plan burns runway. Channels are efficient at delivering volume against a known message to a known audience. Both unknowns at the start. So you end up paying to distribute a message you have not validated, to an audience you have guessed at, and the resulting data is nearly uninterpretable — a poor result cannot distinguish between wrong channel, wrong audience, wrong message, and wrong product.

Four hypotheses, one number. That is not a test, it is a coin flip with an invoice.

Sequence over mix

What works instead is a sequence, and it is deliberately unscalable at the start.

Phase 1: Twenty conversations, zero automation

Find twenty people who plausibly have the problem. Not a segment — twenty named humans. Talk to them. Sell to them manually, by hand, in whatever awkward way works: DMs, email, introductions, a call.

Two things come out of this that nothing else produces. First, the actual words people use for their problem, which are almost never the words on your website. Second, an honest read on whether the problem is urgent or merely acknowledged. Plenty of products solve real problems nobody is in a hurry to fix, and that distinction does not survive a survey but is obvious in conversation.

Expect an ugly conversion rate. It is not the point. The output of phase one is language and qualification criteria, not revenue.

Phase 2: One channel, one message, until it repeats

Now take the best-performing message from those conversations and put it in exactly one channel. One.

The instinct is to spread across three or four to see which works. Resist it, for two reasons: at low volume you will not reach significance in any of them, and you will not have the attention to execute any of them properly. Four badly run channels teach you nothing except that everything is mediocre.

Which channel does not matter nearly as much as people think. What matters is picking the one where your specific audience already congregates and where you can iterate messaging quickly. Run it until you get a repeatable result — meaning you can predict roughly what a given effort produces. Usually somewhere between thirty and sixty customers.

Phase 3: Instrument, then diversify

Only now does the channel-mix conversation make sense, because now you have three things you did not have before: validated message, defined qualification criteria, and a baseline conversion rate to compare a second channel against.

This is also the right moment to build the CRM foundation — not before. Early on, lifecycle automation is premature: too few customers to segment meaningfully, and the flows you would build encode assumptions you are about to disprove. I say this as someone who does CRM work for a living. Building it at twenty customers is genuinely wasted effort.

At a hundred, with a validated message, it becomes the highest-leverage thing available.

The retention question, asked early

One thing I push hard on that early teams routinely defer: measure retention from the very first cohort, however small and however statistically meaningless.

Not because the number is reliable at that size. Because it forces a definition. What does a retained customer look like for this business? Used the product twice? Renewed? Still active at ninety days?

Teams that defer this question until they have "enough data" end up scaling acquisition against an undefined outcome, which is how you get a company that grows customer count and flat revenue for four quarters. By the time the retention problem is undeniable, the acquisition machine is built and the org is organised around it, and unwinding that is far harder than defining it early.

Acquisition without a retention definition is not growth. It is churn with better paperwork.

Where AI genuinely helps early

Small teams should use AI aggressively here, in specific places.

It compresses the research phase substantially — synthesising conversation notes, clustering the language people actually use, drafting message variants faster than you could alone. For a team of two or three, that is a real multiplier on the slowest part of phase one.

It is also good for the unglamorous operational load: drafting sequences, building briefs, keeping a fragmented stack coherent.

What it does not do is tell you whether the problem is urgent. That comes from twenty conversations where you hear how people talk about it and whether they have already tried to solve it themselves. There is no shortcut, and the teams that try to synthesise their way past it tend to produce very fluent marketing for a product nobody needs quite yet.

The version I would run

If I had to compress it: talk to twenty people by hand. Extract their language. Pick one channel and one message and stay there until the result repeats. Define what retention means before you have enough data to measure it. Build the CRM foundation at a hundred, not at twenty. Then, and only then, have the channel-mix conversation.

It looks slower on a plan. It is considerably faster in practice, because you spend the runway learning instead of distributing a guess.