Most people’s first session with an AI assistant goes the same way. You ask it something clever, it answers well, you think “that is impressive,” and then you close the tab and go back to work. Nothing changes, because nothing was actually handed over.
The fix is to stop treating it as a conversation and start treating it as a new hire with an unusual skill profile: fast, tireless, well read, no memory of your business unless you provide it, and no judgement about what matters to you.
Here is a first week that ends with something different in the business.
Day one: give it the context a new hire would get
Before asking for anything, paste in what a new employee would receive in their first hour. What the business does. Who the customer is. The five services you actually sell and what each costs. How you talk to people. What you will not do.
Keep this somewhere you can paste it again, or use a project so it persists. Nearly every disappointing AI output traces back to this step being skipped. The tool was not guessing badly; it was guessing, because you did not tell it.
Day two: the quote follow-up you keep not sending
Take a real quote you sent three weeks ago that went quiet. Give the context, the quote, and the silence, and ask for three follow-up messages of different tone: one short and practical, one that adds a piece of useful information, one that closes the loop and lets the person go politely.
Send the one you would have written if you had the time. The win is not the writing quality. It is that the message left the building.
Day three: turn your last job into a case study
Describe a project you finished: what the customer’s problem was, what you did, what changed. Ask for a structured write-up with the specifics kept and the adjectives removed.
Then check every number in it. If you cannot verify a figure, cut it or say where it came from and over what period. Invented results are the fastest way to lose a careful buyer, and AI will produce them confidently if you let it.
Day four: connect it to your actual numbers
This is the step that separates a novelty from a tool. Claude connects to QuickBooks, PayPal, HubSpot, Canva, Docusign and your email suite, which means it can work from what your business did rather than from your description of what it did.
Useful first asks: which invoices are overdue and by how long, what the cash position looks like against what is landing this month, which enquiries from the last quarter never got a second contact.
That last one usually produces an uncomfortable list. It is also usually the fastest money in the business.
Day five: write down what worked
Whatever produced good output, save the exact instruction. You are building a small library of things that work, and that library is the actual asset. The tool will keep changing. Your description of how your business talks and what it needs will not.
How to tell whether the output is any good
Three questions. Would you have sent this yourself? Does it contain a specific fact that is true and checkable? Does it sound like your business or like a business?
If the answer to the third is “a business,” the context you gave it was too thin. Go back to day one.
And keep a rule that anything customer-facing gets read by a person before it goes out. Not because the tool is unreliable, but because your name is on it.
If you would rather have this set up properly than experiment for a month, schedule a call with us. We will look at where your time actually goes and build the workflow around that, not around a tool.
