GTM Compression

In 2022, if a founder had asked me what the shortest honest timeline was from here is the idea to the site is live and we can take money, I would have said six weeks on a good day and twelve on an ordinary one. That was the scoping-to-launch arc every studio I knew was quoting. The first two weeks were discovery. The next six or eight were design and development. The last two were, in the polite phrasing of the statement of work, launch preparation — which, in practice, meant a frantic scramble to assemble landing pages, pricing pages, analytics, a mailing list and some sort of an answer to the question who, specifically, is this for. Marketing was the thing that happened at the end of the engagement, after the real work was done, by whoever had the least to do that fortnight.

That timetable is gone. A brief that would have taken six weeks in 2022 can, in 2026, move from first conversation to a paying-customer-facing product in six days. The build is a Tuesday, to borrow the phrasing of The Bottleneck Moves Again; the rest of the week is the thing that used to be its own two-month engagement. What has happened, quietly, is not that marketing got faster. What has happened is that marketing stopped being the end of the engagement and became the whole shape of it. You cannot ship something in six days and then spend three months working out how to sell it; the asymmetry looks silly as soon as you lay it on a calendar. So everything after the build compresses to match the build, and the question becomes whether the compression is genuine or whether we are just running the old routine at double speed and calling it new.

We think the compression is genuine on three axes, and we think those axes compress on different clocks. They are worth walking in turn, because founders and studios get them in the wrong order more often than not.

Pricing — fastest.

Pricing compresses faster than anything else on the go-to-market shelf, and it does so because the experiment has no physical cost. A price point is a variable. You can change a variable many times a day, route different users to different variables through a modern product-analytics platform wired up to a simple feature-flag service, and watch the conversion curve respond in near real time. What used to be a quarterly exercise — pricing-committee meeting, new deck, new sales script, a week of internal alignment, a month of rollout — is, in the current tooling, a week. A founder I was on a call with in early 2026 tested five distinct price points inside a single week. At any given moment during that week, five live variants of her checkout page were running against five randomly assigned cohorts of inbound traffic, and the attribution layer was attaching a conversion number, a cart-value number and a seven-day retention number to each variant without her writing a line of SQL.

The reason this works is not that the underlying question is easier. Pricing is still one of the hardest questions a business has to answer, and the philosophical bit of it — what is this thing honestly worth to the person paying for it — has not compressed at all. The mechanical bit has compressed. The cost of running a pricing test used to be the cost of assembling the test. By the time you had a landing page, a variant, an analytics hook and a billing integration that could charge two prices in the same week, you had burned most of the time you would have spent running the test at all. The modern stack compresses the assembly cost to something close to zero. Feature flag, variant copy, billing route, dashboard. Ten minutes of set-up per variant; ten minutes of tear-down. The rest of the week is looking at the numbers and deciding what they mean, which is where the human judgement belongs.

I want to be careful about one thing here, because pricing compression is the axis most likely to be misread as move fast and break things. It is not. A founder running five price points in a week is not guessing faster; she is reducing the cost of asking a question honestly. The question is the same question businesses have always asked. The new thing is that the answer arrives inside the same week she framed the question, which means she can ask it again the following Monday with what she has learned. That is a feedback loop closing.

Positioning — second fastest.

Positioning compresses on the next clock down, which is the clock of a working day rather than an hour. The reason is that positioning is a language problem, not a variable-substitution problem. You cannot feature-flag a story. What you can do, and what the founders we have watched through 2025 and 2026 are doing, is run three or four landing-page variants a day against the same inbound traffic, each with a different hero sentence, a different first screen, a different promise and a different shape of call-to-action. The AI-assisted copy tool produces the first draft of each variant in minutes. An edge-deploy workflow pushes the variant live in under a minute after the copy is approved. The analytics surface, the same one underneath the pricing tests, attaches a time-on-page, a scroll-depth and a conversion-to-next-step number to each variant inside the same day.

What you are testing, when you run positioning variants on this cadence, is the sentence that matters most in any business, which is the one the customer would use to describe the product to a colleague. In 2022 that sentence was a thing a founder argued about with a marketing consultant for six weeks and then committed to, because the cost of re-testing it was too high. In 2026 that sentence is a thing a founder tests seven ways on a Tuesday and then decides about on a Thursday, because by Thursday the data has arrived. The sentences that come out of the AI draft are, as with any AI draft, mediocre on arrival and need a human pass before they go live. But the quantity of sentences that can be drafted, deployed and measured in a day has risen by about an order of magnitude, and that changes what kind of question positioning becomes. It stops being a creative bet and starts being a search, in the proper sense of the word: a search across a space of framings for the one that the market recognises.

This is the axis where I have seen founders most often fall back into the old habit, which is to fall in love with a single sentence before the data is in and then defend it against all of the evidence. The new tooling leaves the habit intact and makes the evidence arrive faster, which makes the defending look sillier sooner. Studios who are good at this work have a gentle, repeated practice of pulling the founder back to the scoreboard on Thursday afternoon, so the Monday version of them does not inherit an argument they already lost.

ICP — slowest but still compressed.

The Ideal Customer Profile is the slowest of the three axes to move, and the reason is that ICP compression bumps up against the physical world in a way pricing and positioning do not. To test a new ICP hypothesis, you need to put the product in front of a cohort of people who plausibly fit it, which means you need a list of those people, which means you need someone — or something — to have done the list-building. Lists are slow. Enrichment tools, even the good ones, can only do so much against the raw friction of finding twenty-five right-shaped companies with twenty-five right-shaped decision-makers inside them, matching contact records to those decision-makers and getting outbound to them in a format that does not immediately flag as automated spam.

Even so, the compression on this axis is real, and it is conservatively four-to-six times faster than the 2022 baseline. The founders we have watched run two ICP hypotheses in parallel in a week — hypothesis A, mid-market logistics operators with 50–200 headcount; hypothesis B, early-stage fintech platforms with a compliance function built in-house — by sourcing two lists of fifty companies each, running two slightly different outbound sequences against them, and reading the response rates and the qualified-call rates at the end of the week. By feature-flag standards that is a slow experiment, but one that in 2022 would have taken the better part of a quarter to run end-to-end — one outbound agency to brief, one to onboard, one campaign to cycle, one round of reporting to wait for. The list-builders are faster now. The sequencing tools are faster now. The analytics layer that tells you whether the replies from cohort A were of a different quality than the replies from cohort B is faster now. And the founder, who in 2022 would have watched this from a distance through a weekly report, is now inside the experiment daily, reading the individual replies and adjusting the sequence copy in the afternoon.

Where ICP compression is fragile is the point at which the founder tries to conclude from a small sample. A week’s outbound against fifty accounts is not a statistically clean test, and anyone who has run this work knows that the signal-to-noise ratio at that sample size is closer to noise than signal. What the compression allows is a better-directed next week. Last week’s hypothesis A beat hypothesis B at the response-rate level; next week runs a sharper version of hypothesis A against a larger list, and a new hypothesis C emerges from what the replies revealed. ICP, done honestly, is still a twelve-week process. What has compressed is the iteration cadence inside that process: where 2022 ICP work was a single pass through a quarter, 2026 ICP work is six or eight passes through the same quarter, each one slightly better informed than the one before it.

A quick note on the stack, because this chapter will age faster than the rest of the book if we let it.

The three axes above are described in terms of the category of tool each one requires: a modern product-analytics platform, an edge-deploy workflow, a usage-based billing layer, a feature-flag service, an enrichment and outbound layer for the ICP work. Those categories are going to outlast the names currently sitting inside them. At the moment of writing, the specific stack the founders we work with most often assemble is feature flags (via a dedicated service), PostHog for product analytics, Vercel for edge deploys, Stripe for usage-based billing, and one of several enrichment / sequencing tools on the outbound side. I am writing those down because naming tools honestly is a studio discipline and the book argues for doing so; I expect half of them to be replaced or rebranded inside eighteen months. The pattern — analytics + deploy + billing + outbound, each one programmable, each one measured — is the thing that outlives the brand names. Readers in 2029 will substitute whatever their equivalent stack is, and the chapter will still do the work.

What is actually compressing, across all three axes, is the cost of asking a question. In 2022 a pricing question cost a quarter; in 2026 it costs a week. In 2022 a positioning question cost six weeks; in 2026 it costs a day. In 2022 an ICP question cost a quarter; in 2026 it costs a week, and four iterations of the week give you what the quarter used to.

None of this removes the expensive part of selling. The part that is genuinely expensive (building the brand, earning the trust, accumulating the reputation that means a prospective customer picks up the phone when the name shows up in her inbox) has not compressed at all, and the previous two chapters have been our attempt to describe why it has not. The GTM compression described here is the compression of everything downstream of that expensive part. Without the brand and the trust, the compression accelerates the motion of pedals that are not connected to a chain. With them, it turns what used to be a quarterly rhythm into a weekly one, and a studio and a founder who are working well together can adjust a business on a weekly cadence in a way they could not in the old economy.

Shipping is the cheap part; selling is the rest of the year.