Performance Marketing for Startups: The First 90 Days

Most early-stage marketing budget does not fail because the ads were bad. It fails because it was spread across four channels, none of which received enough volume to produce a conclusion, and at the end of the quarter nobody could say what had been learned.

Here is a sequence that avoids that.

Day 0: Check whether you should be spending at all

Paid acquisition amplifies existing demand. It is poor at creating demand that does not exist, and expensive as a way to test whether people want your product.

Before committing budget, look for evidence of pull:

• Are people coming back? Early retention is the strongest signal you have.
• Has anyone bought without being personally convinced by a founder?
• Is there organic word of mouth, however small?
• Can you describe, in one sentence, who this is for and what it replaces?

If those are all weak, the constraint is not distribution — it is the product or the positioning. Ads will make that constraint more expensive, not less. Any agency that takes a media budget from a company in this position without saying so is selling you something you should not buy.

Weeks 1-2: Pick one channel

The instinct to hedge across several channels is the single most expensive mistake at this stage. Four channels at a quarter of the budget each produce four inconclusive results.

Choose based on how your buyers currently behave:

Google when demand already exists. If people search for the problem you solve — “invoice software for freelancers,” “physiotherapy in Pune” — capturing that intent is cheaper than creating it. High-intent search is usually the correct first channel for anything solving a recognized problem.

Meta when the category is new or visual. If buyers do not know to search for you, or the product is discovered rather than sought — consumer goods, novel categories, impulse purchases — paid social lets you interrupt the right audience with a proposition they were not looking for.

There are exceptions, but if you are genuinely uncertain, the question “do people already search for this?” resolves it most of the time.

Weeks 2-3: Fix the destination before sending traffic

Sending paid traffic to a page that does not convert is a costly way to learn your page does not convert.

The minimum before spending: one page that states plainly what the product is and who it is for, a single obvious action, load times that do not lose mobile visitors, and conversion tracking you have personally tested end to end.

That last one is not optional. A meaningful share of early campaigns run for weeks against broken tracking, producing data that leads to confidently wrong decisions.

Weeks 3-8: Run a test large enough to conclude something

A test needs enough conversions to distinguish signal from noise. As a rough working floor, aim for a budget that will produce at least 30 to 50 conversions within the test window. Below that, you are reading randomness.

If your budget cannot produce 30 conversions on your target action, use a shallower conversion event — email signups, qualified leads, trial starts — and treat it as a proxy while you gather volume. Optimizing toward an event that fires twice a week does not give the platform enough to learn from either.

Keep the structure simple. A small number of campaigns, a clear audience, three or four creative variants testing genuinely different propositions rather than button colours. Complexity at this stage mostly fragments your data.

Then leave it alone. Daily intervention during a learning period is the second most common way early tests get ruined.

Weeks 8-12: Judge it on economics, not ROAS

This is where the decision actually gets made, and where surface metrics mislead.

The numbers that matter:

Customer acquisition cost against contribution margin. Not against revenue. If a ₹4,000 order carries ₹1,200 of contribution margin after cost of goods, shipping, payment fees, and returns, then a ₹1,500 CAC loses money on every sale — while showing a ROAS above 2.5x that looks fine on a dashboard.

Payback period. How long until a customer repays acquisition cost. Under three months is comfortable for most early companies; past six, you are effectively financing growth and need to know that deliberately rather than by accident.

Repeat behaviour. Whether paid-acquired customers behave like your best organic ones or churn faster. A channel that acquires structurally worse customers can look efficient for a quarter and damage the business over a year.

The decision

At 90 days you should be able to say one of three things:

The economics work. Scale deliberately — increase budget in steps, watch whether efficiency holds, and expect it to degrade somewhat as you move past the cheapest audience.

The economics do not work yet, but the problem is identifiable. Conversion rate, margin, or targeting is the constraint. Fix the specific thing and retest once.

This channel is wrong for this business. Stop. Try the other channel, or accept that your growth is not going to come from paid acquisition at this stage. This outcome is not a failure — a clear negative answer for a defined cost is a genuinely good result, and far better than another optimistic quarter.

What to avoid throughout

Optimizing before you have signal. Statistical noise is not a trend. Making daily changes based on a handful of conversions produces motion, not learning.

Comparing yourself to published benchmarks. Other companies’ CAC figures reflect their margins, their category, their brand awareness, and occasionally their creative accounting.

Adding channels to escape a bad result. If channel one is not working, understand why before starting channel two. Otherwise you repeat the same mistake with a different logo.

Treating the agency’s dashboard as the source of truth. Platform-reported conversions are consistently generous. Reconcile against actual orders in your own system, monthly, without exception.

The startups that get this right are rarely the ones with the largest budgets. They are the ones that resisted spreading thin, ran one test properly, and were willing to accept an unwelcome answer.