01GTM metrics are a short list with owners
A go to market metric is a number you chose in advance because it tells you whether a stage of the plan is working. It is not every chart a product tool can draw. If nobody will look at it on the same weekday every week, it is not a GTM metric yet. It is a report.
The GTM plan is the calendar of tasks. The GTM playbook is the set of moves you repeat after launch. GTM metrics sit beside both: they tell you which tasks to keep and which moves to stop. The words themselves are defined in the glossary. This playbook is only about which numbers to keep, and how to put them in a sheet a small team will actually fill in.
Name one owner for the sheet. That person does not need to be an analyst. They need to update the row and bring it to the weekly review. Everyone else can argue with the row. Nobody else should invent a second sheet.
02What each stage is allowed to prove
Stages exist so you do not judge a pre launch week by revenue. Pick the stage you are in, then pick the one question that stage can answer.
| Stage | The question | The metric to write down |
|---|---|---|
| Before anyone can buy | Are the right people paying attention, and can a stranger get value? | Useful replies, and activation on the private build |
| Launch window | Did the announcement reach buyers, and did they try? | Launch counts on X, signups, activation of new users |
| Early use | Do people come back and pay? | Retention in a window you chose, conversion to paid or to a sales conversation |
| Repeatable motion | Does a new user pay back the cost of winning them? | A plain payback view, plus inference cost per active user |
One row in that table is the headline for the week. The others can stay as context. A weekly review that tries to move five headlines at once moves none of them.
Before launch, write down substantive replies from people who fit the buyer, and how many outside testers reached a first useful result. A raw follower count does not answer either question. In the launch window, add signups and the X counts below, and keep activation in the same row so a loud day with no first result is visible immediately. After people can use the product, retention and conversion become the headline. Inference cost joins when usage moves the model bill. Payback joins when you have both a cost of winning customers and money they have paid.
03Activation, conversion and retention
These three words get used as if they were software features. They are definitions you write in a sentence and then refuse to change mid week.
Activation is the first useful result, not an account created and not a failed prompt. Write that sentence in the user's words and count people who hit it. If you change the sentence, start a new column.
Conversion is the step from trying to paying, or to a real sales conversation if the product is not self serve. Pick one. Do not add both into one cell. The GTM strategy page is where the motion is chosen. The metric only records it.
Retention is people active in a later window after they activated. A week and a month are enough. "Active" should mean they did the job again. Logins are a weak stand in.
Change the sentences when the product changes, not when a week looks bad.
04Payback, without a finance team
Payback means the cost of winning a customer is covered by what that customer pays. You do not need a model with a dozen assumptions.
For a cohort you can name, write what you spent to win them (ad spend and anything you paid creators; founder time can stay as hours) and what they have paid so far. If the payments have not caught the spend, that cohort has not reached payback. There is no universal number of months that makes a plan good. The sheet shows the gap so the next month can shrink spend, raise price, or keep more of the cohort. Do not hide a bad gap by redefining the cohort after you see it. Note whether you sell an annual contract or a small recurring plan, so later readers do not compare different businesses.
05Inference cost per active user
Classic software often treats hosting as a rounding error. An AI product can spend real money every time someone has a successful session. GTM metrics that ignore that spend will celebrate usage that makes the business worse.
Once a week, add up what you paid model providers, or the internal compute you can attribute to the product, for that week. Divide by the number of users who did the core job at least once in the same week. That ratio is inference cost per active user.
Read it next to activation. A rising ratio with flat activation means serving the product got more expensive without more people reaching a useful result. Look at retries, oversized models, and background calls before you change the go to market plan. If the ratio is stable and activation is rising, spend the meeting on retention.
06Launch metrics on X
X is the lead channel for the launches this site plans, so the launch row needs counts from X that a founder can write down without a data team. Keep them separate from product metrics. A repost is not activation.
For the launch post and the few posts around it, record:
- Replies that are real questions or objections from people who fit the buyer, not every notification.
- Reposts and bookmarks, as a rough sign the post was worth keeping.
- Follows that arrived with the launch, with a note if you can tell they match the buyer.
- Clicks to the one link you asked people to use, taken from whatever link count you already trust. Do not add a second link so the sheet has more columns.
If the launch uses video, X's own Media Studio describes a specific view metric. In the Media Studio analytics help page (archived May 15, 2026, because X help pages block automated checks), a view is counted when someone watches for at least 2 seconds with at least half of the player in view, and the figures are logged in views only. The same page separates organic views from promoted views, and it warns that metrics are aggregated by video, not by post, so reposting the same file adds up. Use those labels in the sheet if you have Media Studio. Do not mix them with a casual "views" number from a screenshot of the post.
Paid is a different system. X's conversion tracking overview (archived September 19, 2026) says conversion tracking measures actions people take after viewing or engaging with ads, through the X Pixel on the site or the Conversions API from a server. If you run ads, install that before the launch week so the sheet is not empty afterward. Read the page for the consent notes before you turn it on. It is measurement for ads, not a substitute for the organic counts above.
Check the live help pages before you depend on a definition. The archives are dated because the live sites often refuse automated fetches, and the labels can change.
07Set up the sheet
One tab. One row per week. Freeze the header. That is the whole setup.
Columns that earn their place:
- Week ending date.
- Stage, in the words of the table above.
- Headline metric for that stage, as a single number.
- Activation count, with the sentence you are using written in row 1 and never only in someone's head.
- Conversion count, same way.
- Retention count for the window you picked.
- Inference spend and active users, so the ratio can be recomputed when you fix a mistake.
- X launch counts, filled during a launch week and left blank later rather than stuffed with ordinary posting.
- A note of one or two sentences: what you changed, and what you will not change next week.
Fill the sheet the same day each week. Do not add a column mid meeting, and do not delete a bad week. Keep definitions in the header row. Leave out anything you cannot act on this month, including a blended score that hides a bad retention week, and do not track a second definition of the same word. The AI GTM playbook is for positioning and trust. This page is for which number decides the argument. A spreadsheet or a document table is enough if the owner can update it in one short sitting.
08A weekly review that stays short
Open the sheet. Read the headline cell. Ask three questions only.
- Did the definition stay the same as last week?
- Did the number move for a reason you can name?
- What is the one change in the plan before next week?
If you cannot name a reason, do not invent one in the note. Write "not sure" and look at the raw events before the next review. A wrong story in the note becomes next month's strategy if nobody challenges it.
If you want the definitions and the sheet filled in for your own product, the custom AI GTM plan is $499. The sample plan shows how a written plan is structured, and the GTM score shows which part of the launch is weakest right now.
Pick one metric per stage, write it in a sheet you actually open, and add inference cost once people are using the product.
Questions about this playbook
01What are the main GTM metrics for an AI product?
Activation, conversion, retention and a plain payback view, plus the launch counts you record on X and inference cost per active user. The list should stay short enough to review every week.
02Which GTM metrics matter before launch?
Before launch, track whether the right people are paying attention and whether a stranger can reach a first useful result. Revenue metrics can wait until someone can actually buy or start a trial.
03How should a team count a launch on X?
Write down replies worth answering, reposts, bookmarks and follows from the launch post, and separate organic video views from promoted views if you use video. For ads, X conversion tracking is a different system.
04What is inference cost per active user?
It is model or compute spend for a period divided by users who did the core job in that period. A jump with flat activation means the product is getting more expensive to serve without getting more useful.
05Do we need a metrics tool on day one?
No. A single sheet with one row per week is enough until the same definitions have survived a few reviews. Add a tool when the sheet becomes the bottleneck, not before.