You know the moment. The call went well, the prospect said yes, and now you have to put a number on paper. So you scope the job from memory. You remember the deliverables because the client talked about them for twenty minutes. You forget the part nobody mentioned: the account that doesn’t exist yet, the access you’ll spend a morning chasing, the integration that works on the third try.
Those are the hours that disappear. You don’t lose them in the proposal. You lose them three weeks later, at eleven at night, doing work you never priced because it wasn’t in your head when you wrote the quote.
If you run social in-house, you have your own version of this. It’s the campaign brief that forgot the approval round, or the launch plan that assumed legal would turn things around in a day. Same problem. The work you can’t see is the work you can’t plan for.
I talked with Corey Walker about this on Episode 12 of The AI Social Playbook. Corey has run The Marketing Specialist for twenty years and co-wrote Instagram for Dummies, and she built a Claude Project to write her client proposals. I expected the story to be about speed, and her draft does come back in about two minutes. The part that stuck with me was something else she said almost in passing: because Claude writes out the scope in so much detail, it reminds her what she forgot to charge for.
How the setup works
Corey’s project has three pieces, and none of them are complicated.
The instructions hold her rate card. In plain sentences: what she charges per hour, what she generally charges when a client wants three posts a week, what she charges for a Mailchimp campaign. These are the rules Claude prices from.
The context holds her past proposals. She uploads her most recent ones, the ones that reflect her current rates, so Claude learns her branding and how she describes each service.
The prompt stays short. For the example she showed, the client had a CRM and a Mailchimp account with three emails already written, and needed the two connected through Zapier with the automation built. Her prompt gave that background and then said: “based on my hourly rate and my estimated four hours to complete this task, please create a proposal.”
Notice where the hour estimate comes from. Corey supplies it. Claude doesn’t decide how long the work takes. She does, and Claude turns her estimate into a document.
What came back was a full proposal: cover page, scope of work, investment, how the integration would work, payment terms, next steps. It also included a section on what’s not included, which she never asked for. Claude worked that out from the scope.
Read the scope before you read the price
Most of us jump straight to the number when the draft comes back. Corey’s experience says to start one section earlier.
When Claude spells out every step of a job, the steps you skipped in your head show up on the page. In Corey’s words, it’ll remind her, “Oh yeah, you’re gonna have to do that part, and that part’s a real pain, and you should really be charging another hour for that.” Her example was a Facebook and Instagram ads job where the client didn’t have an ad account yet. Setting one up means time in what she called “another load of time and horror that is Meta Business Suite.” That’s an hour she would have done for free.
So the first pass on any draft is a scope review. Go line by line and ask whether each task is something you actually priced. Anything that isn’t gets one of two treatments: you add it to the price, or you move it into what’s not included so the client knows it’s extra. Both are fine. Doing it silently for free is the only wrong answer.
In the Blueprint for this episode, I added one line to the project instructions that Corey didn’t describe: ask Claude to end every draft with a separate list of tasks the scope requires that your prompt didn’t mention, each with an hour estimate. It turns a happy accident into a step you can count on.
Treat Claude’s number as a second opinion
Once the scope is right, look at the price. This is where Corey overrules Claude, and she does it in both directions.
I asked her about a job Claude priced at $3,500 that she’d normally quote at $1,500. She works with a lot of small business owners, and she knows $3,500 isn’t in their budget for that kind of project. She also knows her own speed: “Something that might take someone three hours, I could do in an hour.” Claude has no way to know either of those things. So she backs the price off.
Then she said the part I think matters more. Sometimes she looks at Claude’s number and thinks, “I should be charging more. Like, why am I selling myself short?”
That’s the value. Claude prices from your rules every single time. You price from your rules plus your mood, your memory of the last client who pushed back, and how much you want this one. When the two numbers disagree, the gap is worth a minute of honest thought. Sometimes you know something Claude doesn’t. Sometimes you’ve been discounting out of habit.
And if your rules only live in your head, Claude has nothing to go by. Writing your rate card into the project instructions forces a useful exercise: deciding what you actually charge, in writing, before the next prospect asks.
Where it breaks
This isn’t set-and-forget, and Corey was candid about that.
The first few runs took tweaking. Most of it was giving Claude enough information up front to price correctly. Branding took longer. Claude wouldn’t render her logo and left a question mark where it belonged, so she replaced it with her business name as text. Her pink and brown brand bar came out sized wrong until she worked through a few rounds of corrections. She says it seems to be retaining those fixes now.
Two limits are worth naming. The scope is only as complete as the job description you give it. Claude can infer a missing ad account from “Facebook and Instagram ads.” It can’t infer the client’s internal approval process if you never mention it. And the hour estimate is still yours. If you guess low, Claude will write a beautiful proposal around the wrong number.
None of this is Claude-specific, either. ChatGPT, Gemini, and Copilot all have a projects or notebooks feature with an instructions field and somewhere to upload documents. The names differ. The setup is the same.
The bigger lesson
Corey still waits about a day before she sends a proposal, even though it takes her minutes to produce. She doesn’t want a client thinking it was too easy. I like that, because it says something about where she puts AI in her process. Claude does the drafting and the arithmetic. She handles the judgment and the relationship.
That split is the thing to carry to whatever tool you pick up next. Ask two questions about it. What can this tool see that I tend to miss? For Corey’s proposals, it’s the full list of tasks. And what do I know that it never will? For her, it’s the client’s budget and her own speed. Give the tool the first job and keep the second one for yourself.
Corey is already planning to build the same kind of project for her contracts. That’s how this tends to go. You fix one document you produce over and over, you learn what the AI is good at, and then you move on to the next one, whether that’s a contract, a campaign brief or a monthly client report.
