Growth Marketing for Startups: Channels and Metrics
·
Growth marketing for startups from a performance marketer: a channel table, the CAC and retention numbers to track, and how to run weekly experiments.
Growth marketing for startups is the practice of finding, testing and scaling the channels that bring in customers who stay, using small experiments and hard numbers instead of big campaigns and gut feel. You pick a metric that reflects real value, run cheap tests across a few channels, keep what pays back and kill the rest quickly.
I've spent most of my career on the performance side of marketing. Cost per install. Cost per lead. Verified users. The world where a client pays for results, and if the numbers don't hold, the campaign stops. That world teaches you one thing fast: reach is cheap, results are expensive.
## How growth marketing for startups differs from brand marketing
Brand marketing asks: how many people saw us, and how do they feel about us?
Growth marketing asks: how many of them became customers, what did each one cost, and are they still here a month later?
Both matter eventually. Early on, a startup can rarely afford the first question without answering the second. Brand budgets are patient. Runways aren't.
The other difference is speed. A growth team runs many small tests rather than one big launch, because the [Law of Shitty Clickthroughs](https://andrewchen.com/the-law-of-shitty-clickthroughs/) is real. Andrew Chen's version: over time, all marketing strategies result in shitty clickthrough rates. He points out that the first banner ad in 1994 had a 78% clickthrough rate. Every channel decays as more people pile into it, so you need a steady flow of new experiments.
## The numbers that matter
Before channels, the numbers. Get these right and channel decisions become much easier.
- **Customer acquisition cost (CAC).** What you spend to get one paying customer. The [a16z guide to 16 startup metrics](https://a16z.com/16-startup-metrics/) makes a point I wish more founders knew: paid CAC matters more than blended CAC for judging whether a business works, because blended numbers mix in free users and hide how your paid campaigns are really doing.
- **Lifetime value (LTV).** The same a16z piece defines it as the present value of future net profit from a customer, and warns against calculating it on revenue. Use margin.
- **Cohort retention.** Of the users who joined in a given week, how many are still active after 4, 8, 12 weeks?
- **Payback period.** How many months until a customer's margin covers what you spent to get them. In a tight runway, short payback beats high LTV.
- **Cost per result.** Cost per install, per lead, per verified user, per first purchase. Define "result" as close to revenue as you can.
i used to report on clicks because clicks always went up. Funnily, the business didn't.
## Growth marketing channels compared
Here's how I think about the main channels for a young company. Cost levels are relative, and "first metric to watch" is the number I'd look at in week one before I trust anything else.
| Channel | Best for | Cost level | First metric to watch |
|---|---|---|---|
| Founder led outreach | B2B, high value deals, first 50 customers | Low cash, high time | Reply rate and meetings booked |
| Search content and SEO | Products people actively search for | Low cash, slow to pay back | Organic sign-ups per article |
| Paid search | High intent buyers, clear problem keywords | Medium to high | Cost per lead or first purchase |
| Paid social | Consumer apps, visual products, broad audiences | Medium | Cost per install or sign-up, then day 7 retention |
| Performance networks (CPI, CPL, CPA) | Apps and lead gen where you pay only for results | Variable, pay per result | Cost per verified user and fraud rate |
| Influencers and creators | Consumer, gaming, lifestyle | Medium to high | Tracked sign-ups per creator |
| Communities (Discord, WhatsApp, Reddit) | Gaming, developer tools, niche hobbies | Low cash, high time | Active members and referrals |
| Partnerships and co-marketing | Borrowing a trusted audience | Low to medium | Sign-ups per partner |
| Referral programmes | Products people naturally talk about | Low | Invites sent per active user |
| Events and esports activations | Gaming, youth brands, local markets | High | Leads or sign-ups captured per event |
Brian Balfour's [Four Fits framework](https://brianbalfour.com/four-fits-growth-framework) has a line that saves a lot of wasted budget: "Products are built to fit with channels. Channels do not mold to products." A low price app can't afford a sales team. A complex B2B product won't sell from a 15 second video. Match the channel to how your customer already buys.
### A word on pay per result models
Performance models (cost per install, cost per lead, cost per action) are a good fit for startups because you pay when something happens. But they have traps I've seen many times:
- **Define the result tightly.** An "install" that never opens the app is worth nothing. Pay for verified users or a post-install event where you can.
- **Watch for fraud.** Bots, fake leads and incentivised installs that churn on day one. Check quality by source, every week.
- **Look past day one.** A cheap install with zero day 7 retention is the most expensive install you'll buy.
- **Cap budgets per source.** Scale the sources that hold quality, cut the rest without emotion.
Gaming is where I've seen these models used most heavily. If you're building in that space, my look at [India's generative AI in gaming market](/blog/india-generative-ai-in-gaming-market) covers why user acquisition costs and distribution matter so much there. And for the live events side, see [why esports finals work as commerce engines](/blog/esports-finals-commerce-engines).
## How to run growth experiments
The engine of growth marketing is a weekly loop. Here's the version I use.
1. **Hypothesis.** "If we change the landing page headline to the user's own words, sign-up rate will rise."
2. **Metric.** One number that decides the result, defined before you start.
3. **Smallest test.** The cheapest way to learn. A new ad set, one email, one page variant.
4. **Time box.** One or two weeks. Long enough for signal, short enough to stay honest.
5. **Decision.** Keep, kill or iterate. Write the result down, even when it fails.
Keep a simple experiment log: date, hypothesis, channel, metric, result, decision. After a quarter, that log becomes the most valuable growth document you own. It tells new hires what's already been tried, stops you repeating failed tests, and shows investors that your growth comes from a process rather than luck.
Run two or three experiments a week at most. More than that and you won't know what caused what.
My [startup growth plan template](/blog/startup-growth-plan) slots these experiments into a 13 week calendar, so each week has a goal, a channel and a metric.
## Adapting for AI search
A lot of discovery now starts inside AI answer engines as well as search results. Content that answers a question clearly in the first lines, cites real sources and stays consistent across the web tends to get picked up. I wrote about the practical changes in [how AI search is changing SEO](/blog/ai-search-is-changing-seo-how-to-adapt-your-digital-marketing-strategy).
## When to spend, and when not to
Paid growth amplifies whatever is already true. If retention is weak, paid makes the leak bigger. If you're still searching for your first fans, read [how to grow a startup when you have more time than money](/blog/how-to-grow-a-startup) first; manual tactics will teach you more per rupee.
For the full picture of how marketing fits into product fit, channel fit and scale, see my pillar piece on [startup growth](/blog/boost-startup-effective-growth-strategies).
## What I'd do this week
- [ ] Calculate paid CAC and payback period separately from blended numbers.
- [ ] Pick two channels from the table that match how your customer buys.
- [ ] Define one "result" that is as close to revenue as possible.
- [ ] Write three experiment hypotheses and run the cheapest one.
- [ ] Pull day 7 retention by acquisition source.
Every channel gets worse over time… the only real edge is learning faster than it decays.
Vikasifications