Not subscribed? Sign up to get it in your inbox every week.

Last month, a corporate card company shipped an LLM router, and nobody blinked.

Ramp Router sits in front of your AI calls and, in Ramp's words, "sends every AI request to the lowest-cost approved model that clears your quality bar." The company says the same system cut its own model bill by about 30%.

On paper this is infrastructure software from a fintech, about as far from expense reports as a launch can drift. In practice everyone understood it in one read, because Ramp has told the same story since the original card in 2019:

Ramp saves you money.

The card that flags wasted spend fits the sentence, and so do travel booked under policy, procurement, treasury, and banking. So when AI tokens became, as Ramp's launch copy puts it, "the fastest-growing spend category," the router had a slot in the story waiting for it. The launch barely needed marketing, because the marketing had been running for seven years.

That's the subject this week: the order of operations. Most companies build a product, then hire marketing to explain it, then send sales to push it. I think that sequence is backwards. The motion that works runs marketing first, building second, sales third, as one continuous thing: write the story of the company, let the story decide what gets built, and let sales close people who believe it.

First, terms. The marketing I mean is the story: who the customer is, what you get them, why each product exists, and how the whole roadmap rolls up into one vision that explains why the company exists. (I sold marketing software for five years; discount my bias as needed.) Ads, launches, and logos come later and matter less. Written at the end of the process, the story has to be reverse-engineered from whatever got built, which is why so much SaaS marketing reads like a feature list with adjectives. Written at the start, it becomes the strategy document everyone else executes against.

The effects show up everywhere at once, because a company tells its story to more audiences than prospects. The same sentence has to work on a customer deciding whether to renew, an investor deciding what you're worth, a candidate deciding whether to join, and the engineer deciding which of two features survives the sprint. When the story is crisp, those decisions all point the same direction without a meeting. When it's muddled, every audience invents its own version, and you find out during the renewal call, the board meeting, and the sprint review, respectively.

Amazon turned this order into process decades ago. Under Working Backwards, a team that wants to build something first writes the press release and FAQ for the finished product, in customer language, as if it had already shipped. Leadership reads the document, and a press release that bores them kills the project before an engineer is assigned. Kindle, Prime, and AWS all started life as fictional press releases. The mechanism is the point: the press release is the cheapest possible test of whether a product has a story, and it costs a week instead of a roadmap slot.

Last week I argued that AI collapsed the cost of the second product and the unit of competition is shifting to the persona. This week is the other half of that machine. When any competent team can ship a credible v1 in a quarter, features converge, and they converge from every direction at once. What refuses to converge is the story, because a story compounds with every product that fits it and every year it stays the same. Ramp's router will be copied within quarters. The sentence it slotted into took seven years, and that is the part a competitor can't easily ship.

The spine I wrote about last week is the technical half of this argument: the products share one data spine. The story is the same fact living in the customer's head. When the two match, every launch makes the next launch cheaper to explain, which is exactly the compounding you watch in Ramp's product history. When they drift apart, you get the catalog problem: a roadmap of reasonable products that nobody, including your own team, can explain in one breath.

There's also a new audience reading the story now: the software. The context files your agents run on (the CLAUDE.md at the repo root, the skills library, the system prompts) all begin with a paragraph about what the company does and for whom. That paragraph is marketing, written for a machine that makes a hundred small decisions a day based on it. A muddled story now confuses your agents in the same week it confuses your sales team.

Two moves, both cheap.

First, measure your story debt. Ask five people (a sales rep, an engineer, your newest hire, a customer if you're brave) to finish the sentence "This company exists to ___ for ____ ." Count the distinct answers. One answer means the story is doing its job. Three answers means three different roadmaps are quietly being built in three different heads, and marketing's most urgent audience is internal.

Second, run the press-release test on whatever sits at the top of your roadmap. Three paragraphs in the customer's voice, plus one line stating how the product rolls up into the company sentence, written before the first ticket. If the document excites nobody, you just saved a quarter of building. Therefore the worst case of marketing first is a wasted week of writing, while the worst case of building first is a launch nobody can explain.

Build, market, sell made sense when building was the hard part. The story is the hard part now. Write it first.

Till next time,
Chris