Claude를 영업 관리자로 활용하는 방법: Daniel Skill 가이드

Yes, and at one specific job better than you would expect. Every deal rests on claims: some you have observed, with a trace, and the rest you were told. Sorting a deal narrative into those two columns is a text classification task, which is what these tools are actually for, and the model does it without the loyalty that stops a human doing it honestly. It also will not get tired on deal number eleven, which is where a manager's scepticism usually runs out. What it cannot do is tell a rep their deal is not real. That part is still yours.

Observed is something that happened and left a trace: a security questionnaire arrived, legal asked for the DPA, three people came to the second call, a date was given in writing, procurement raised a ticket. Asserted is something you were told: budget is approved, the CFO is on board, the competitor is out. Assertions are rarely lies. They are usually a champion repeating what they believe, filtered through a rep who badly wants it to be true. Which column a claim sits in is the single most useful fact about a deal.

Not on its own, and you should be suspicious of any number it produces unprompted. There is no close rate for your business anywhere in its training data, so a confident percentage or an industry benchmark is invented. Give it your own historical conversion rates by stage and let it do the arithmetic, with an explicit instruction to write a placeholder rather than supply any figure you did not give it. Then ask which deals the number is most sensitive to, and which committed deals are standing on assertions.

Point it at the rep rather than the deal. Its instinct is to hand you the play for the stuck deal, and if you pass that on you have closed one deal and taught nothing. Ask instead what the deal reveals about how that person sells, what single habit would change the most, and what question would let them notice it themselves in the one-to-one. Then ask what you should expect to see in their next three deals if the coaching landed, so you can check in a month rather than guess.

Because one deal is an anecdote. A manager coaching from the deal that most recently went wrong is coaching noise, and everyone does it because recall is easy and pattern-finding across a quarter is not. Reading five call notes and finding what repeats is a job a model does well and a busy human does badly. The important step comes after: classify the pattern as rep skill, process, product or pricing, or market. Five deals that died at procurement is a pricing problem, and coaching the rep for it is useless and unfair.

Rehearse the conversation with it, yes. Draft the document, no. Performance improvement plans, written warnings and termination rationales are employment-law instruments, the requirements differ by country and often by contract, and they end up in front of lawyers. Wording that sounds firm and fair in a chat window can create real exposure. That work goes through HR and legal, who also know what your own process requires.

Treat it as a data-protection decision rather than a productivity one. A recorded call contains identifiable information about people who agreed to be recorded by your company, not to have it pasted into a chat tool, and what you may do with it is governed by your privacy notice, your data processing agreements and your customer contracts. The answer also depends on your account tier. Settle it once with whoever owns data protection before it becomes a habit. Working from your own notes about a call rather than the transcript sidesteps most of it.

The download is a ZIP with SKILL.md at the root of the archive rather than inside a nested folder, which is the usual reason an upload fails. 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. 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 there is no upload step, so open SKILL.md, copy the contents and paste them into custom instructions.