如何使用 Claude 進行業務開發：Connor Skill 指南

Yes, but not for what most guides suggest. Almost all advice about AI for business development is really advice about cold email, which is a different job. Use it to build a partnership thesis, to construct a scoring frame, to work out the questions to ask on a first call, and to stress-test a deal before you commit a quarter to it. Do not use it for volume outreach, and never let it supply a fact about the other company.

In sales the offer is already defined and you are talking to a buyer with a budget and a problem, over weeks or months, and you lose to a competitor or to no decision. In business development what you build together does not exist yet, the counterparty is buying nothing, the cycle runs for months or years with long silences that mean nothing, and it dies quietly in month four when the champion moves team. That difference decides what AI is good for: volume and follow-up discipline in sales, structure and stress-testing in BD.

Because the real answer depends on facts it does not have. It will tell you a target would gain expanded distribution or access to a new segment, which is fluent and true of almost anyone. What actually decides a partnership is whether they already have three partners doing exactly this, whether their strategy this year is expansion or margin, whether the team you would work with is growing or being wound down, and who inside gets credit if it works. A model has none of that and will not tell you it has none of it.

It will, and the scores are close to astrology. Paste fifty companies and ask for a fit score out of ten and you get fifty ranked numbers with one-line justifications that look like analysis and come from the vibe of the brand name. The fix is not to abandon scoring but to make the model build the frame and leave every cell empty: criteria you agree with, a named evidence source for each, and blanks you fill from that source. Then the ranking means something.

Firmographics. Headcount, funding, revenue, executive names, product details, and especially who their existing partners are. These come back specific-looking and are sometimes fiction. One wrong fact about the recipient's own company is disqualifying in a way a generic email is not: generic reads as lazy, wrong reads as careless, and partnerships run on the belief that you will be careful. Instruct the model to write VERIFY and leave a gap rather than filling it.

Not whether the company benefits. Companies do not sign things, people inside them do, and they do it when saying yes helps them personally. The operative question is who inside the organisation gets promoted if this works, and what their boss is measuring them on this quarter. If you cannot name that person and that number, you do not have a partnership opportunity yet, you have an interesting idea.

Run a pre-mortem before you commit the time. Ask the model to write the five most likely failure stories twelve months out, each with the month it went wrong: the champion left, it was never anyone's priority so it was never resourced, procurement outlasted the enthusiasm, the economics only worked for one side at real volume, or both sides did the easy half and nobody did the integration. For each, the earliest signal and the one question you could ask in the next two weeks to surface it now.

The download is a ZIP with SKILL.md at the root of the archive, not inside a nested folder. In the Claude desktop app open Customize then Skills, upload the ZIP and toggle it on. Skills need code execution enabled, under Settings then Capabilities, or by an owner in organisation settings on Team and Enterprise. Anthropic's help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise, so if you are on the free plan check Settings then Capabilities for your own account rather than trusting either page. In ChatGPT or Gemini, paste the contents of SKILL.md into custom instructions instead.