Distribution Engineering Is a Discipline

I The claim

Distribution engineering is the work of specifying, building and instrumenting a product’s route to people with the same rigour as the product itself. We think it is a discipline, with inputs, outputs and known ways to fail, and that most teams treat it as a mood.

The definition has three parts: a stated offer hypothesis, a test with a threshold set before the data arrives, and a measured handoff into operations. A team that leaves one of them out is doing something else, usually posting.

II Building got cheap. The route did not.

The supply of products went up fast. The Next Web, citing The Information, reported that new app submissions to Apple’s App Store rose 84% in a single quarter as vibe coding went mainstream, and that 2025 closed with 557,000 new submissions, the largest annual wave since 2016. Product Hunt’s own quarterly report counted 2,345 launches in the second quarter of 2026, up from 1,508 a year earlier.

The supply of attention did not move, since nobody acquired more hours in which to look at new products.

Stripe Atlas has figures on what happened to outcomes over the same period. Solo founders made up 63% of the C corps formed through Atlas so far in the second quarter of 2026, an all-time high by its count. Among solo-founded startups, median revenue in the first six months fell 23% year over year in 2025, while the top decile rose 19%. Four years ago the top decile made about 34 times the median, and in 2025 it made 61 times.

Our reading is that more people can build, about the same number can get found, and so the middle sinks. Stripe does not claim that cause and we cannot prove it. But the same analysis found that top-decile solo founders took 51% of their revenue from international sales, against 2% for the median. A gap that size describes where the product was sold, and a difference in features would not produce it.

III What the discipline specifies

Engineering starts with a spec. Ours for distribution has five parts, in this order.

1. The offer, written as a claim that can fail. Who it is for, what they get, what it costs, and what result would prove us wrong.
2. The test. A landing page, a funnel, A/B variants, and a threshold written down before any traffic arrives.
3. The warm read. People we trust see it first. Then we watch real users try it, before anything goes out cold.
4. The opening sequence. For an offer that passed its test: keywords, one press release, then channels opened one at a time so we can tell which one did the work.
5. The handoff. Operations receive an offer with numbers attached.

AI agents work alongside every step, drafting the variants, building the pages and pulling the numbers. They do not get a vote on what counts as a win, because that was written down before they started.

The threshold is the part people skip, and it is the part that makes the rest honest. Kromatic’s Real Startup Book puts it as a step in its landing page procedure: “Set your success threshold before you launch.” A team that picks the bar after seeing the result will find, every time, that the result cleared it. We have never met a founder who looked at a 3% conversion rate and said that they had privately been hoping for 4%.

The same guide carries a warning we repeat often: “Leaving an email is cheap; an email signup is not a purchase.” So the spec also has to say what kind of evidence the test produces. A waitlist shows interest and a card number shows rather more, and the handoff should say which of the two it is carrying.

IV Why this is not growth marketing

Growth marketing reports in volume: posts published, emails sent, impressions served, leads generated. Those are fine things to count once you know the route works, but before that point they measure effort, and a struggling launch already has plenty of effort.

Distribution engineering reports a validated route and a decision, and the decision can be go, change or kill. A test that ends with “this offer missed the threshold, change the offer” has done its job, and done it weeks before the same news would have arrived as a bad quarter. A growth plan has no place to record that outcome, so it records activity instead and keeps going.

The order of work differs too. The growth instinct is to open every channel at once and see what sticks, whereas we open them in sequence, warm before cold, because five channels opened on the same day produce one blended number and no idea whose it is.

There is outside support for warm before cold. Growth Unhinged surveyed 300 software go-to-market leaders for its 2026 State of B2B GTM report. The survey is self-reported and was sponsored by go-to-market vendors, so read it as direction. It ranked network and introductions as the best channel and intimate events second, and its author wrote: “Cold outreach is officially oversaturated, and AI SDRs aren’t living up to the early hype.” The channels that worked were the ones where somebody already trusted somebody. That is where we start, and it is also the cheapest place to find out your offer is confusing.

V The handoff

A product team would not ship without tests, logs and a way to roll back. The route deserves the same, and the handoff is where you find out whether it got them.

What operations should receive fits on a page: the offer as it was stated; the threshold and the result against it; which channels were opened, in what order, and what each one returned; and what was tried and failed, so nobody pays to learn it twice. If the person who ran the work left tomorrow, somebody else could run the route next month from that page.

Most launches hand over a login to the ad account and a Slack thread, and the company then spends its next six months rediscovering what it already knew.

We run this as Kill or Ship, in two stages, and the test comes before the build. Stage one, The Straight Answer, is two weeks: research, a live landing-page test with real users against a pass/fail number agreed in writing on day one, and a verdict of go, change or kill. Stage two, Ship it, is three more weeks and happens only if stage one passes; that is when we build a working version, launch it, set the keywords, send the one press release, open the channels in sequence and hand over. Two weeks is long enough to run one honest test and short enough that nobody can hide in it. If your route to users lives in one person’s head, has no number it was supposed to hit, and has never been allowed to fail, it has not been engineered yet.