Startup distribution: why it now comes before the product
Hugo Chamberland
8 min
After ChatGPT launched, most people at OpenAI thought a chatbot couldn't be the product. Peter Thiel told Sam Altman otherwise: they had created a new kind of customer. Startup distribution starts with that sentence.
In OpenAI Understands Something Important and Rare, David George of a16z argues that in a world of abundant intelligence, the winner is not the one with the best product. It is the one best at creating customers and reaching them. That holds well beyond AI labs. This article explains why distribution now comes first for a startup, what it changes in a Belgian context, and how to build it into a first version.
When anyone can build the product, the product stops being the bottleneck. The customer is.
What is a startup distribution strategy? A startup distribution strategy is the plan for how a product reaches the people who will pay for it: which channel, which first users, which offer, and how each is measured from launch. It is decided before the product is built, because it changes what gets built.
The a16z piece makes the point at the scale of AI labs. Models are easy to swap, prices fall for everyone, so switching costs and pricing stop protecting anyone. What is left is the ability to create a new behaviour and reach people with it. A startup faces the same squeeze, one level down: if a competent team can ship your product in six weeks, the product alone protects nothing.
Building got cheap. Customers didn't.
The failure data has not moved with the cost of code. CB Insights analysed 431 VC-backed startups that shut down since 2023. Running out of capital tops the list at 70%, but the report calls it the final cause, not the root one. Underneath it: poor product-market fit in 43% of cases, bad timing in 29%, and unsustainable unit economics in 19%. The median company had raised $11 million before closing.
None of those causes is fixed by writing code faster. They are fixed by finding out earlier who buys, through which door, at which price.
💡 In short: a cheaper build moves the risk, it doesn't remove it. The money saved on development is only saved if it buys an earlier answer to "who pays?"
Distribution isn't what you do after launch. It's a design decision before it.
Most founders treat distribution as a marketing phase: build first, then find users. That order made sense when the build took a year and most of the budget. It no longer does. The channel you pick shapes the onboarding, the pricing page, the data you collect and even the language of the interface.
What must be true before the first sprint
Four things, each concrete enough to check:
- One channel, named. Not "social and SEO and partnerships", but the single door the first hundred users will come through.
- A list of the first people you can reach directly, by name, before any ad spend.
- A price, written down, even if it changes after the first ten conversations.
- A way to pay on day one. A product nobody can buy proves interest, not a business.
The Belgian twist: your first channel is a language choice
Belgium had 11,867,634 inhabitants on 1 January 2026, split across Flanders (6.9 million), Wallonia (3.7 million) and Brussels (1.26 million). Three regions, three official languages, separate media and often separate professional networks. A Brussels startup that launches "in Belgium" is in fact launching in two or three small markets at once, with a third of the effort in each.
The practical answer is to pick one region, one language and one network first, and expand once that channel converts. For a B2B product, crossing into the Netherlands or France with the same language can be a shorter step than crossing the linguistic border at home.
⚠️ Watch out: three channels half-built. A landing page in three languages, an ad account, a partner programme and a newsletter, all at 30%. None produces enough signal to learn from. One channel at 100% beats three at a third.
Three ways to reach customers, and who ends up owning them
The a16z piece describes three ways to deliver a product to many people: a standalone product, partnerships, or a platform others build on. The labels come from the AI frontier, but every startup picks one of them, often without noticing.
| Route | Speed of reach | Who owns the customer | What you learn |
|---|---|---|---|
| Standalone product | Slow at first | You | Everything: usage, objections, what people try to do |
| Partnership or reseller | Fast | The partner | What the partner chooses to tell you |
| Platform others build on | Slowest | You, through the builders | What people build, which is the widest signal |
The standalone route: slow, but you keep the learning
ChatGPT is the extreme case of the first row. OpenAI reached 900 million weekly active users by February 2026 through its own product, and every one of those sessions taught it what people want from the model. For a startup at a tiny fraction of that scale, the logic holds: owning the interface means owning the feedback.
The partnership route: fast, but someone else holds the relationship
A reseller, an integrator or a marketplace gets you in front of many customers quickly. The partner, however, owns the customer's trust and decides what the customer does with your product. The a16z authors note that customers may not even know you are behind it. That is fine for reach. It is a problem if the partner later builds your feature itself.
👉 Choose the route before the stack. A product sold through a partner needs an API and clean onboarding for someone else's team. A standalone product needs payment, analytics and a self-serve first session. They are different first versions.
Users are not customers: the limits of the distribution argument
Distribution wins attention. It does not guarantee revenue. When OpenAI announced its 900 million weekly users, it also reported 50 million paying subscribers, about one user in eighteen. The largest distribution engine in software still converts a small share of its reach.
For a startup, the lesson is to measure the right thing from the first week:
| Measure | What it tells you | Trap |
|---|---|---|
| Sign-ups | The channel works | Says nothing about value |
| Weekly active use | The product is useful | Can hide a free tier nobody upgrades from |
| Paid conversions | Someone values it enough to pay | Small numbers early, so read them as signals, not rates |
| Repeat purchase or renewal | The model can run again | Takes months to observe |
Two other limits deserve a mention. First, distribution cannot rescue a product nobody needs: it only makes that fact visible faster, which is still worth paying for. Second, the source has a stake in its own thesis. a16z is an OpenAI investor, a point raised by readers under the article. The argument about distribution stands on its own; the prediction that OpenAI will win is a separate claim.
🔍 Test before you scale a channel: of the last 20 people who signed up, how many did something only a real customer would do, like inviting a colleague, importing their own data or entering a card? If the answer is close to zero, more reach will only multiply the zero.
Four distribution mistakes founders make in the first version
1. Launching with no way to pay
A free beta collects compliments; a payment link collects proof.
2. Tracking page views instead of events
A traffic pixel tells you people arrived, not which step of the first session they abandoned.
3. Handing the first customers to a partner
The fastest reach in month one can become the partner's customer base, and its learning, by month twelve.
4. Building the list after the product
The people who said yes to an early demo are your first channel; without their contacts saved, every launch starts from zero.
💡 What the four share: each one postpones the moment you learn who pays. In a market where building is fast, that moment is the only real deadline.
How we build distribution into a first version at Nightborn
Step 1: pick the channel in the scoping workshop
In a four-hour session, we decide the one channel, the first audience and the success criteria before any feature list. Deliverable: a problem statement, the channel hypothesis and the numbers that will confirm or kill it. The format is on our Product Discovery page.
Step 2: ship the commercial layer with the product
The first version goes out with a landing page, a tested price, payment in place (Stripe or equivalent) and event-level analytics with funnels, not just a traffic pixel. One channel is wired end to end, not three halfway. More on our MVP Design & Development page.
Step 3: keep the list you can come back to
Every contact from demos, sign-ups and early conversations is stored and segmented from week one. The next release, the price change or the second product starts with people who already know you.
📌 Checklist for a first version: one named channel, a price on the page, a way to pay, events tracked on the first session, and a list of people you can write to tomorrow.
Startup distribution: the first version is a channel test, not a product launch
What to remember:
- Startup distribution is decided before the build, because the channel shapes the first version.
- Failure data hasn't moved: poor product-market fit sits behind 43% of recent shutdowns, whatever the cost of code.
- In Belgium, the first channel decision is also a language and region decision.
- Users are not customers: measure payments and repeat use, not sign-ups.
The implication for your next build: write the channel, the price and the first fifty names before the first ticket. If the list is hard to fill, that is the most useful thing you will learn this month, and it cost nothing to build. If you want to test that list against a real first version, it is the starting point of every scoping workshop we run.
