Skip to content

Customer acquisition with AI

From first research to personalized outreach: this guide shows how B2B teams run their entire outbound process from a single chat window — based on their own ICP (Ideal Customer Profile — the written definition of your dream customer) instead of generic AI copy.

by · updated

TL;DR
  • Customer acquisition with AI doesn't mean sending more emails faster — it means scoring target accounts against your own ICP before anyone picks up the phone.
  • Via the MCP standard, your LLM accesses your sales data directly — target lists, ICP fit, and call scripts are created in the chat, not across five tabs.
  • Cold outreach stays bound by the same legal rules even with AI support: in B2B, calls require presumed consent, and email marketing still requires opt-in.

Why traditional customer acquisition keeps getting more expensive

The mechanics of recent years were volume: more contacts, more sequences, more touchpoints. Exactly that volume is now the problem. B2B decision-makers receive interchangeable messages every day, response rates are dropping, and every additional touchpoint costs more than the last.

The expensive part is rarely the conversation itself — it's the preparation. Who is even a fit? What is the company doing right now? Which of our references is relevant? This research sits scattered across CRM, website, LinkedIn, and notes — and it gets redone for every single contact.

AI accelerates the wrong thing here when it lacks context: it produces copy faster, but copy that concretely fits no one. It becomes useful only when it works from the same foundation as your team — your ICP, your positioning, your data.

B2B outbound directly from your LLM

The difference is not the model — it's the connection. Via MCP (Model Context Protocol), an LLM like ChatGPT, Claude, or Gemini can access GrowthKit like a tool. It reads your stored target-customer definition, scores accounts against it, and returns the results right inside the same conversation.

In practice: you ask the chat for the twenty accounts with the best ICP fit, get the reasoning per account, and have a call script drafted for the first one — without switching windows and without re-explaining the context.

The context persists. What you store once about your target audience, your objections, and your reference customers is available in every future session — including for colleagues on the same team.

Which sales software supports MCP?

MCP is an open standard for connecting tools to AI assistants. It's young, and the number of vendors running their own MCP servers is only just growing. GrowthKit provides such a server — for Claude and for ChatGPT.

When evaluating options, the decisive question is less the feature count than which data the server actually knows. An MCP connection that only returns generic search results is no substitute for a sales data foundation. It becomes useful where your ICP, your accounts, and your history sit behind it.

Connecting GrowthKit to your AI assistant

Setup is a one-time task that takes a few minutes: generate an access token for the MCP server in your account settings, add GrowthKit as a connector in your assistant — and the connection is available in every new chat. If the connection fails, it's almost always the token: expired, copied incompletely, or generated for a different account. A new token resolves these cases.

The details — step by step, with screenshots — are described on the respective pages: set up GrowthKit MCP for Claude, set up GrowthKit MCP for ChatGPT, and the Chrome Extension for research directly in the browser. Setup differs between assistants only in the connector settings.

What can I actually do with it?

Three workflows teams typically switch first — each entirely in the chat, with no export and re-import.

Build a target list. Describe your ideal segment once, properly; scoring accounts against that ICP then happens automatically, including the reasoning per match.

Prioritize accounts. Instead of sorting by deal size or gut feeling, you sort by fit. The ranking can be challenged in the conversation — and corrected when your market works differently.

Prepare the conversation. Openers, discovery questions, and objection handling are drafted fresh per account, based on what's known about the company. Structure, not a script — the human leads the call.

Glossary

ICP (Ideal Customer Profile)
The written definition of your dream customer — industry, company size, typical problems, buying situation. It's the benchmark every contact list is scored against, instead of sorting by gut feeling or deal size.
LLM (Large Language Model)
An AI language model — the technology behind ChatGPT, Claude, and Gemini. It processes and generates text in conversation; via a connection like MCP it can additionally access external tools and data.
MCP (Model Context Protocol)
An open standard that lets AI assistants like Claude call external tools and data sources. Instead of copying data into the chat, the assistant calls the provider directly — with your access rights and in your context.
UWG (German Unfair Competition Act)
The German law governing when advertising is legally permissible. § 7 UWG is especially relevant for cold outreach: calling businesses is allowed under presumed consent, while email marketing requires the recipient's explicit consent.

Frequently asked questions

Try customer acquisition with AI

Start free, store your ICP, get your first accounts scored — no credit card required.