FAQ · AI

“Can’t I just do this myself with ChatGPT?”

Part of it: yes. ChatGPT writes a good email once you tell it to whom and why. The rest: no. It does not find 300 matching companies in the commercial register, does not back claims with a source, does not remember your lead list next week, does not watch websites and does not know whom it has already written to. b2breachout uses the same model families as ChatGPT — the difference is not the model but the data pipes, the memory and the process that keeps running every week on its own.

Auf Deutsch lesen

What ChatGPT does well — and where it stops

ChatGPT is strong at writing: outreach copy, variants, tone, translations, ideas for your positioning. If you know the company, the person and the occasion, you get a usable text in seconds. Nobody disputes that, and b2breachout uses the same kind of language model for its drafts (Anthropic Claude, OpenAI as fallback).

The problem is what comes before. Before anyone can write a text, you need the right company, evidence that it matches your profile, the person who decides, and a reason to reach out right now. That part is 90 percent of the work in outbound — and exactly there a chat window has no tools. Ask ChatGPT for “50 family-owned metal-working companies in the canton of Aargau with 20 to 200 employees” and you get a mix of real, outdated and invented entries — no address, no legal form, no source, and a different list next time.

The data pipes ChatGPT does not have

b2breachout is not a prompt but a system with connected sources. Each of them delivers something a language model cannot know by itself:

  • Commercial register Zefix (Switzerland): official company search by name, legal form and canton — up to twelve register queries per run, current and free. Result: real, existing companies with seat and status.
  • SHAB publications: management and board members with their official role are read from the register entry — the decision-makers ChatGPT can at best guess.
  • Google Places: address, phone, website and location data per company; queries are spread across your prioritised regions.
  • Firecrawl: web search in directories plus reading the company website with evidence — team page, offering, news. This produces enrichment, contact emails and signals.
  • Apollo.io (optional): company size for the size filter, contact data and target roles such as procurement or engineering when the website does not name them.
  • Weekly monitoring: monitored companies are checked automatically every Monday for website changes and news; new signals arrive as a digest.
  • Your memory: customer profile, positioning, priorities, every lead with status, contact history and draft versions — permanently, not until you close the chat tab.

What a single run actually does

A discovery run derives search terms from your entire customer profile (register terms for Zefix, descriptive terms for Places and web search), queries the sources, merges the hits and removes duplicates, filters junk and exclusions deterministically, scores the candidates with the language model against your weighted criteria, enriches the best ones immediately with website data and stores everything with a source. Companies already in your list or disqualified are skipped; your feedback on the last run changes the next one.

That takes a few minutes and runs without you. In ChatGPT, each of these steps would be its own prompt, each source a copy-paste, and none of it would know about the previous step.

The honest calculation: building it yourself

Of course it can be rebuilt. With a coding assistant you have a script over a weekend that queries Zefix and lets ChatGPT score a list. Then the real work starts: API keys and costs for Places, Firecrawl and Apollo; deduplication and data quality (name variants, “Your Electro | Partner”); pre-filters so hospitals do not land under “health”; a database for leads, status and history; a cron job for monitoring; quotas so one run does not cost 200 francs; login, backups, and the legal building blocks for Switzerland. Then you maintain all of it whenever an API changes.

That is a software project, not a prompt — and your business is selling, not running data pipelines. b2breachout is exactly that project, operated, maintained and set up for Swiss and European markets. Whether it pays off is a matter of your deal size: in most B2B industries a single new customer pays for a year of the platform.

Side by side

ChatGPT (chat)ChatGPT + own scriptsb2breachout
Finding companiesFrom memory, partly invented, no sourcePossible if you connect Zefix/Places/Firecrawl yourselfBuilt in: Zefix, Places, directories, with a source per company
Fit scoringPer request, not reproducibleBuild it yourself, incl. weights and exclusionsWeighted criteria, reasoning, learning loop from your feedback
Decision-makersGuessedParse SHAB yourselfFrom SHAB, team pages, target roles, optional Apollo
Buying signalsNoneOwn crawler + cron requiredWeekly monitoring with classification and digest
MemoryEnds with the chatOwn database requiredLeads, status, contact history, draft versions — permanent
DraftsGood if you supply the contextGood, context from your scriptsFour channels, your positioning, the current signal, approved drafts as style reference
Law (CH)On youOn youNo auto-send, postal address and opt-out in the draft, blocklist
OperationsNoneYou: APIs, costs, maintenanceUs: operations, quotas, updates, EU data hosting

The human stays in the loop — and ChatGPT may stay too

b2breachout sends nothing automatically. Every draft is reviewed, adapted and sent by you from your own mailbox; the platform learns from what you approve. If you like to rewrite drafts in ChatGPT on top, you can keep doing that — the draft is the blank, not the finished message. The value lies in what happens before the text.

The essentials in five points

  1. ChatGPT writes copy. b2breachout finds, sources, scores and monitors the companies there is copy to write about.
  2. Same model families, different tools: commercial register, SHAB, Google Places, Firecrawl, optional Apollo — connected, not guessed.
  3. Memory and repetition: leads, status, feedback and weekly monitoring keep running long after the chat is closed.
  4. Building it yourself is possible — it is a software project with ongoing operations, not a prompt.
  5. No auto-send: you approve, the platform does the work before.

Run the comparison: ask ChatGPT for your target companies — and start a run with your customer profile in parallel.

Related question: “I already have HubSpot — why b2breachout?”