You can build the pipeline. Here's how to decide what it should do.

The builder's trap

If you come from engineering, the GTM stack is the easy part. You can connect a CRM to an enrichment tool, score accounts, push them into a sequencer and log every reply. The hard question is the one nobody hands you as a ticket: what should this pipeline be optimising for, and how will you know it is working?

For example, if you build a scoring model before anyone has written down the ICP, the model encodes whatever the person who asked for it believed that week. If you scale an outbound sequence before anyone has checked whether those customers stay long enough to repay what it costs to win them, you have automated a loss. Neither problem shows up in the code. Both show up three months later in the numbers.

GTM engineering is the discipline of building go-to-market systems with engineering methods. The engineering part you already have. What follows is the decision part, in the order we would settle it.

1. Write the ICP and the offer down as standing documents

Start with what the company believes about who it sells to, what it offers and why it wins. Not as a prompt you paste into every chat, but as a file the agent must read before it produces anything that depends on it.

In GTM Brain, every new brain starts with a draft strategy.md. Running /gtm-brain-strategy starts a 10 to 15 minute conversation with the founder or the head of sales, growth or GTM. It asks plain questions (who are your best customers, why did they pick you, which deals do you say no to, how do you know you are on track) and fills the document. Each point in it ends with a label: Confirmed, Hypothesis or Gap, with its source. The person approves it with "approve the strategy".

From there, /gtm-brain-document keeps the narrower documents the work must follow: a target customer definition with the trigger that makes a company buy and what disqualifies it, the offer, pricing, the tone of voice. A document carries a kind: document marker and says what kinds of work it applies to. Before the agent writes copy, builds a list or drafts a playbook, it reads the documents that apply in full and says which one it followed.

That changes what you build. Your list filters, scoring rules and sequence copy now trace back to a document someone approved, and when the document changes, you know what to rebuild.

2. Separate evidence from hypothesis

The second decision is about honesty: which of the beliefs in those documents are supported, and which are guesses you are about to spend money on?

/gtm-brain-market-desk-research is built for this. It starts with a 5 to 8 minute intake about the market, your offering and your best proof, shows you a research frame and a proposed set of sources to approve, and then runs sourced desk research. The report covers market and segment growth, customer pains, the buying committee (influencers, champions, main decision makers and blockers), competitors and your current right to win, with every material claim labelled Confirmed, Hypothesis or Gap and linked to its source.

Two properties matter for a builder. It is a diagnosis only: it will not design your product, your pricing or your positioning, so it cannot quietly turn a hypothesis into a plan. And it keeps attention signals apart from demand. For example, repository activity or search interest show that people are looking, not that they will pay. That distinction is the one most dashboards blur.

3. Check whether the motion pays, with the inputs on the table

Now the numbers. A builder can calculate CAC and LTV in a spreadsheet in ten minutes. The work is in making the inputs comparable, and that is where /gtm-brain-unit-economics earns its place. Its method, written by Lech Kaniuk, insists on a few rules: one customer unit for both LTV and CAC, monthly units throughout, fully loaded CAC (people, tools, agencies, not only ad spend), contribution margin instead of revenue, LTV over 12 and 24 months instead of a lifetime figure, and both paybacks shown next to LTV to CAC. It ends in one decision: PUSH, FIX, STOP or MEASURE, with at most three actions.

Here is how that looks for Northstar, the fictional sample company from the GTM Brain site. Every number below is invented for illustration.

Northstar charges 400 a month per account with a 75% contribution margin, so each customer contributes m = 300 a month. Over the last six months, fully loaded outbound cost 54,000 and brought 12 new customers. The sales cycle is about two months, so a six-month window is long enough to measure CAC. CAC is 54,000 / 12 = 4,500. Northstar has no reliable churn data yet.

The simple payback says a customer must stay 15 months to repay what it cost to win them. The retained payback, which accounts for churn, moves that to 18 months if 2% of customers leave each month and past two years if 6% do. Since Northstar has not yet observed customers staying that long, and 12 new customers is too small a sample to push a channel on, the method's answer is MEASURE: collect retention for the accounts already won before raising outbound spend.

For the builder, that answer is a specification. The most useful thing to build next is not a bigger sequence. It is the tracking that records, per account, when it started paying and whether it is still active, so the next review can move from a churn scenario to observed retention. The review can be saved as a decision note, and each missing input becomes an open question that shows up in /gtm-brain-todo until someone supplies it.

4. Write the playbook with its gates

Only now does the pipeline itself get designed, and it gets designed as a playbook rather than a script. In GTM Brain a playbook is a workflow: a file in workflows/ with three sections. Before starting lists what must be settled first, for example the segment, the offer and the success criteria. Steps is a numbered list where each step names a tool category (crm, email, calendar, sequencer, ads, analytics) and an optional gate, such as after 14 days, on 2026-12-03 or when the engaged accounts are known. Learn lists what to read back when the run ends.

/gtm-brain-workflow drafts one from what the brain already knows, starts a run once the brief is confirmed, and tracks each step with its date. It is worth being exact about what that means: the workflow records progress and tells you when a step is due, at the start of a chat, in the todo list and on the dashboard. It does not send an email or launch a campaign by itself. Each step is done in chat through your connected tools with their own permission prompts, or by you, and then recorded.

For an engineer, the gates are the interesting part. They are where a human decision sits inside an automated process, written down in advance, so that "when the engaged accounts are known" is a checkpoint in the run rather than a message someone forgot to send.

Then build

With the four decisions written down, the build is easier to scope and easier to defend. The scoring model implements the approved ICP. The list excludes what the document excludes. The sequence runs as a tracked workflow with gates. The dashboards measure what the unit economics review said was missing. And when a step you repeat becomes routine, /gtm-brain-skill turns it into a reusable skill kept in the brain.

/gtm-brain-unit-economics
Review outbound for the last six months. Fully loaded spend was [amount],
new customers [number], price per account per month [amount], contribution
margin [percent], sales cycle about [months]. We have no churn data yet.
Show every assumption and tell me what to measure first.

Next: One brain per client →