Claude for sales: the setup at 1, 5 and 20 people

Claude for sales means handing Claude the work that surrounds a sales conversation: researching the account, preparing the call, drafting the follow-up and writing the CRM notes. The rep keeps the conversation, the judgement and the close. What changes as the team grows is where Claude reads from, which tools it reaches, and how much runs before anyone asks.

We run this on our own pipeline first, then build it for clients, shaped to how they sell. It sits inside a wider Claude for GTM system, with outbound, content and data around it. This page stays on the selling side of Claude for sales: the rep's day first, then the setup at one, five and twenty people.

01What a rep hands Claude in a normal day

Six jobs come up for every seller we work with. None of them is the call itself. All of them decide whether the call goes well and whether the deal moves afterwards.

Account research before the first touch

Research is the step a busy rep skips first. The Prospect Researcher builds one snapshot from a company domain or a person's LinkedIn profile. It covers the company, the people and what changed lately: hiring, tooling, expansion and leadership moves. The rep reads two or three angles in place of twenty open tabs.

We keep research and writing apart on purpose. The rep sees the facts before anyone drafts a word, so a weak account gets dropped before it costs a sequence. Every skill in this guide sits in our skills library, and the best Claude skills for GTM is the shortlist we hand new teams.

Call prep: a brief before every call

Our own version started from a simple frustration. Account executives often walk into a call without the context, either because they did not look or because they could not find it.

So the brief opens with how the lead got there: the positive reply, the booked call, the discovery call, and what today's call is for. Then the account. What the company does, its size and revenue, who else from there is on LinkedIn, and which decision makers are likely to join.

It closes with the objections likely to come up, what to raise and what to leave alone. All of it is built from the prospect and from what you sell. The Pre-Call Prep skill reads your own history first, then adds web research. One run in the morning covers the whole calendar, because the minutes before a call always go to something else.

Follow-ups that do not wait for memory

I have never met a head of sales happy with how their team handles follow-up. It stays manual even in businesses that automated everything around it. A CRM task reminds you a deal exists and leaves the words to you.

Claude reads the real email threads and shows which deal has been silent the longest. It drafts the next message from the full sequence that already happened and the problem the buyer wants solved. The rep reviews it and sends it. After a call, the Post-Call Debrief reads the transcript and lists who owes what, with a recap ready to edit.

CRM notes, written back from the transcript

Prospects ask us on sales calls where all this activity gets logged. Fair question: a brief and a follow-up are worth little if the CRM still shows nothing happened.

Claude writes the call summary, the next step and the agreed fields back through the CRM's API. We work mostly in HubSpot and Attio, and any CRM with an API takes the same write-back. It only holds up on clean fields and clear stages, so start with our CRM setup checklist if yours has drifted.

Scoring the call afterwards

David runs our discovery calls, mostly alone, so nobody sits in to tell him how one went. On one call, a single true line about how much sending infrastructure we can set up made a prospect file us as a spray-and-pray agency. They had come for a white-glove approach, and everything after that sentence went through one thought.

A recording summary would not have caught it. A summary tells you what you talked about, and leaves out where you lost them. So Claude scores our calls against a rubric we wrote for the way we sell. It checks eight things:

• Pain quantified
• ICP fit confirmed
• Stakeholders mapped
• Commercials raised
• Objections surfaced
• Proof used
• Next step agreed
• Stayed on their problem

The rule that makes it work: every score quotes the call, with a timestamp. No quote, no score. On pain quantified it might say the rep took the answer flat, with no revenue figure to push against later. On ICP fit, that the rep hedged, when the fix was to say why they fit or test whether they do.

Each call ends with one thing to change and one thing to keep. When we build this for a client, the rubric is theirs, because they lose deals for different reasons than we do. We wanted that one feature shaped to how we sell and connected to the prep and the follow-ups.

02Claude for sales at one person

A founder who sells, or a single rep, needs little: Claude in chat, one project, and a handful of skills for the tasks that keep coming back. The whole stack is three tools: Prospeo finds and verifies contacts, Smartlead sends, and Attio is the CRM.

The project is where the sales context lives. Put the offer, the pricing logic, the ICP, the proof you are allowed to use and the objections you hear most into it. Then add the two skills that pay back first: call prep and the post-call debrief. Paste the transcript in after each call and take the recap, the follow-up and the notes out.

Nothing runs on its own here, and at this size that is the right call. The plan sits in the founder's head and gets re-briefed every session. That holds until a second person starts selling.

03Claude for sales teams of five: one shared library

At five people the plan moves out of heads and into files. A shared GitHub repository holds the skills as a library everyone reads from, and MCPs connect it to the live tools. The ICP, the offer and the proof become a version-controlled file, so that work gets done once instead of five times.

For the sales team, the gain is consistency. Every rep's brief reads the same objections file and the same proof. When one rep improves the call prep skill, the other four get the improvement the next time they run it.

The stack grows to eight tools. HeyReach opens LinkedIn next to email. Attio holds the relationships and Supabase holds the data everyone reads from. BounceBan checks the catch-all addresses Prospeo cannot confirm on its own. Call notes now land in Attio from the transcript, so the pipeline view reflects what was said.

Where Claude Code for sales comes in

Five people is where Claude Code earns its place. Chat is fine for one person's prep. A shared library, connectors into the CRM and a write-back step need something that reads files and calls tools directly. The wider stack it runs on is laid out in Claude Code for GTM.

The first governance question arrives at the same time: who is allowed to add to the library. Owning that library is the work of a GTM engineer, and it is the core of GTM engineering as a discipline.

04Twenty people: the sales day runs on a schedule

At twenty the library becomes the product. Enough people build at the same time that the system needs its own opinion about how the work gets done. For the reps, the visible change is that nobody has to ask for the brief any more.

Reps are briefed in the morning on the calls that day. After the calls, we run the scoring by hand for now, because the routine that triggers it per call is not built yet. Follow-up drafts wait in a queue for review, and briefs and alerts land in Slack, where the team already works.

• Signals feed the research. RB2B names who is on the site and Trigify watches the market, so a brief can say why an account is moving now.
• Replies turn into meetings. EmailBison carries the sending volume, and iClosed turns a positive reply into a booked call.
• Every layer gets a second source. Apollo and BlitzAPI sit next to Prospeo, and DeBounce sits behind BounceBan.
• One reasoning layer over every tool. Claude decides what runs next, with Clay doing enrichment for outbound and n8n moving data between tools.
• The funnel sits in one place, from positive reply to closed-won, with the real conversion rate between every stage.

Credentials never reach the model, and anything irreversible stops at a human. This is the setup we run internally. It is also what we build into a client's GTM brain, shaped to how they sell.

Three tools at one person, eight at five, twenty-one at twenty. That progression came out of the team size, and it was never a shopping list. Pick one tool per job and add the next only when the team needs it.

05The surface and connectors at each size

Seen from the rep's chair, Claude for sales changes in five places as the team grows. The table follows one rep through each setup.

The two ways this goes wrong are opposite. Governance built at one person is process for a team of one. Chat sessions still running at twenty means twenty people building twenty versions of the same thing. The setup gets re-picked each time the team outgrows the last one.

06What stays human when you use Claude for sales

The line sits in the same place at every size. Claude prepares, drafts, scores and records. A person does everything that reaches the buyer or cannot be undone.

• The conversation. The brief tells the rep what to raise. Reading the room stays with the rep.
• Every send. Recaps and follow-ups stay drafts until a person reads and sends them.
• Commercials. Pricing, discounts and terms are a person's decision, on the call or after it.
• Which deals to chase. Claude lists the silent deals. The rep decides which ones deserve the next message.
• The fix after a bad call. The score points at the minute it turned. Saying it better next time takes practice.

07How to start using Claude for sales this week

1. 01Write the sales context downPut the offer, ICP, proof, pricing logic and common objections into one Claude project. Every later job reads from it.
2. 02Start with call prepRun a brief for tomorrow's calls and compare it with what you would have looked up yourself. Fix the gaps in the context, then run it again.
3. 03Add the post-call debriefPaste each transcript in and take out the recap, the follow-up draft and the CRM notes. Send nothing you have not read.
4. 04Connect the CRMOnce the notes are reliable, write them back through the CRM's API so the pipeline shows what was said.
5. 05Move to a shared repo at the second sellerWhen another person starts selling, move the context and the skills into GitHub so both of you brief Claude from the same files.

The guide below takes a new user through picking a Claude surface, connecting tools over MCP and writing prompts for sales work.