What is AI-native GTM? The order to test 5 new-business hypotheses
AI-native GTM tests a new business's five hypotheses against customer responses, starting with the top unproven one.
Key points
- Write 5 hypotheses; mark unproven rows
- Buyer unclear? Test customer discovery
- Narrow the promise to fit AI and staff
- Widen acquisition only after repeat use
Contents
Write 5 hypotheses on one sheet before you start
A&A perspective
AI-native GTM is a way of running a new business in which you write its five hypotheses on one sheet, test them against how customers respond, and revise them. GTM stands for go-to-market, the work of deciding who a new product is for and how to bring it to them.
A&A perspective
Automate & Augment (A&A), the company writing this article, splits that work into five hypotheses: customer discovery, the promised outcome, delivery, retention and acquisition channels.
A&A perspective
AI-native means building AI into the work of testing those five hypotheses from the start. It is not the same as using AI to mass-produce sales emails and proposals, because the more you can produce, the harder it becomes to see which hypothesis is wrong.
Hypothetical example
Take a hypothetical case. A 60-person bookkeeping firm launches a monthly service as a new business. AI first does the reconciliation, matching invoices against incoming payments, and a staff member checks the result before it goes back to the client.
Hypothetical example
Give each hypothesis one row on the sheet, and write down the question to test, what counts as customer evidence, the work given to AI and what a person decides. In this example, the customer-discovery row asks who has the problem and who pays. The evidence is the problems you heard from the firms you met, and the person who decides to pay.
A&A perspective
When you mark rows, do not count anything AI produced as customer evidence. A list of target customers, a prototype screen and a proposal deck are all made inside your company. Evidence is something the customer does: what they told you, what they paid for, and whether they asked again the next month.
A&A perspective
Start testing from the top marked row. If the buyer turns out to be someone else, both the promise and the acquisition channels have to be rebuilt. When a row higher up proves wrong, the results of the rows below it cannot be used. The sections that follow cover the promised outcome and delivery together, then retention and acquisition together. The same sheet also works as material to show your manager or board.
Buyer unclear? Test customer discovery first
A&A perspective
While it is unclear who has the problem and who will pay, test customer discovery first. Customer discovery means finding the people with the problem, and the buyer, by actually meeting and talking with them.
From the sources
Stripe, a payments company, publishes a sales guide for founders. The guide recommends that founders actively find their first few customers themselves. It says early customers help you refine the product and understand where you stand in the market.
From the sources
The guide also recommends researching where customers gather and looking there for recurring problems, frustrations and the other products people use. After a conversation, it says to ask plainly whether the person can make the purchase decision alone.
From the sources
A partner at Y Combinator (YC), which invests in and supports newly founded companies, gave similar advice in a 2018 Q&A: figure out the actual buyer, the decision-maker who signs the bill.
A&A perspective
A&A sees this as the row a new business most often gets wrong. In the bookkeeping example, the person struggling with reconciliation may be the accounting clerk, while the person who decides to pay may be the company president. When the user and the payer differ, the way you word the promise changes too.
A&A perspective
At this stage, use AI for research. Have it list candidate firms from industry groups and public information, and sort interview notes into what the customer actually said and what is guesswork. Choosing whom to meet and identifying the buyer stay with the staff member.
A&A perspective
A company with existing customers may already have evidence for this row. If you have already heard the problem and the buyer from current customers, move on to the next row. How to record those interviews is covered in "Customer validation doesn't end with AI research: four columns to record."
| Hypothesis | Work given to AI | What a person decides |
|---|---|---|
| Customer discovery | Sorting public information and interview notes | Whom to meet, and who the buyer is |
| Promised outcome | Drafting the reconciliation | What goes inside and outside the promise |
| Delivery | Reconciling by fixed steps | Handling exceptions and answering the client |
| Retention | Drafting what changed from the previous month | Whether to continue or revise the promise |
| Acquisition channels | Routing inquiries and researching them in advance | Which channels to widen and which to stop |
Narrow the promise to what AI and staff can keep
A&A perspective
Once the buyer is clear, narrow the promised outcome to what AI and your staff can keep. The promised outcome is the result the customer pays to receive. The delivery hypothesis is whether you can keep that promise every month.
From the sources
Paul Graham, who has invested in newly founded companies, wrote about advice to early founders in a 2013 essay. He says he sometimes encourages them to get deeply involved, as if they were building for a single customer.
From the sources
But he says this is safe only while you are not being paid for it. Once customers start paying you by the hour, they expect you to do everything.
A&A perspective
A&A sees this as the reason to fix the scope of the promise first. If the extra care you gave during the trial goes into the promise, you cannot keep it once the service becomes paid. In the bookkeeping example, the promise ends at returning the reconciliation result, and re-investigating exceptions is handled separately.
From the sources
Anthropic, an AI developer, split ways of building with AI into two in a 2024 article: workflows, where AI runs along steps set in advance, and agents, where the AI chooses its own steps.
From the sources
It recommends starting as simply as possible and adding complexity only when needed. A note at the top of the article says much of the tooling it describes has changed since publication.
A&A perspective
Test whether you can keep the promise with the simple form that runs on preset steps. AI drafts the reconciliation by fixed steps, and a staff member looks at the items that do not match and decides how to answer the client. Write into the promise how you handle items AI cannot settle, such as a payment made under a different name.
A&A perspective
The evidence for this row is a record of trial requests returned as promised. Also keep what the staff member corrected and how long checking took. If the effort is not worth the price, narrow the promise or fix the steps before moving on.
Widen acquisition only after clients ask again next month
A&A perspective
Widen acquisition channels only after you confirm that clients ask again the next month. An acquisition channel is the path by which you meet customers and receive inquiries. As in customer discovery, the staff member gathers the first customers one firm at a time.
From the sources
In a 2017 conversation, YC partners said to check first whether people keep using the product before building a team for growth. They added that a user merely coming back does not count as meaningful repeat use. For a lodging booking service, the example counts a completed booking, not just a visit to the site.
A&A perspective
In the bookkeeping example, check whether the client asks for reconciliation again the next month and uses the result in its month-end accounting work. If requests keep coming but the client redoes the matching in-house without using your result, go back to the promise row. How to count retention is covered in "Judge new-venture retention by first-month customers: 3 fixes."
From the sources
Clay, which sells a tool for researching sales prospects, has published a case study about its customer Lovable. Lovable is a Swedish company whose service lets people build web apps by giving instructions in plain language.
From the sources
The case study says inquiries flooded in, and Lovable needed to guide individual users to a self-serve path and connect enterprise buyers to the sales reps suited to their use case.
From the sources
When an enterprise form comes in, Clay's system researches the company on the spot, matches it to the right rep, and gives the rep a brief before the first conversation. The person who built the system says he tests something small, confirms what works and then adds volume.
A&A perspective
What A&A borrows from this case is the order. Lovable built its routing after inquiries had already overflowed. If a new business with no demand yet builds the routing first, there are no inquiries to route.
A&A perspective
Once clients start asking again the next month, widen the channel those clients came through. In the bookkeeping example, that is whichever of outreach to current clients, referrals or search articles brought in the clients who kept ordering. Put AI on routing and researching inquiries; the staff member decides which channels to widen and which to stop.
Summary
A&A perspective
AI-native GTM means writing a new business's five hypotheses on one sheet and testing first the top row that lacks customer evidence.
A&A perspective
Write the five hypotheses on one sheet, and count customer actions, not AI output, as evidence
A&A perspective
While the buyer is unclear, interview firms one at a time and test customer discovery first
A&A perspective
Once the buyer is clear, narrow the promised outcome to what AI and staff can keep
A&A perspective
Widen acquisition channels only after confirming clients ask again the next month
AI-native GTM means writing a new business's five hypotheses on one sheet and testing first the top row that lacks customer evidence. While the buyer is unclear, test customer discovery first; once the buyer is clear, narrow the promised outcome to what AI and staff can keep. Widen acquisition only after confirming clients ask again the next month, and if they do not, go back and revise the promise.
FAQ
How is AI-native GTM different from AI sales?
AI sales means using AI in sales work. AI-native GTM is the whole approach of testing the five hypotheses, from customer discovery to acquisition channels, with AI built in. Sales is one part of it.
Can we test all five hypotheses at the same time?
If your company talks with customers often, testing them side by side can be faster. But if you misjudge the buyer in the customer-discovery row, the results of the other rows become unusable too. Even when you test them side by side, revisit the lower rows once the upper rows have results.
What should we show our manager or board?
Show the five-hypothesis sheet: which rows have evidence, which do not, and which rows you will test in the next month. Adding what the testing will cost and where you will stop if a hypothesis proves wrong makes it easier to answer the approvers' questions.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- 最初の顧客 10 社を獲得する方法
Stripe · Publication date not stated
Accessed 2026-10-23 - Growth AMA with YC Partner Gustaf Alströmer
Y Combinator · 2018-10-29
Accessed 2026-10-23 - Do Things that Don't Scale
Paul Graham · 2013-07
Accessed 2026-10-23 - Building effective AI agents
Anthropic · 2024-12-19
Accessed 2026-10-23 - Growth Office Hours with Anu Hariharan and Gustaf Alstromer
Y Combinator · 2017-12-01
Accessed 2026-10-23 - How Lovable uses Clay to scale one of the fastest-growing startups in history
Clay · Publication date not stated
Accessed 2026-10-23