7. Asking for the First Commitment

The digest has a price, a sample page and an interview sheet. No owner has seen any of it. This chapter puts the offer in front of trade owners and asks for something real. It also builds the record that stops a pleasant conversation from turning into a false belief about demand.

The conversations themselves are short. You show an owner the one-paragraph offer, the invented sample digest and the price of two hundred and seventy dollars a month. Then you ask the sheet's three questions. What do missed enquiries cost you now, in money or time? Would you start a paid two-week trial at this price? What would have to change for a yes? Many good guides to running those talks exist elsewhere. What matters here is what happens after you hang up, because that is the part an agent can do well or badly.

Turning conversations into a decision

Start with your raw notes. Say you talk to five owners in a week, by phone, in person and by email. After each one you type what you remember into a plain text file in the project folder, one file per owner. Type it rough and fast, but put their exact words in quotation marks wherever you caught them. Every answer quoted in this chapter is invented to show the method. An invented yes proves nothing about whether real owners want a digest. Only real conversations can do that.

Next you open Claude Code in the project folder and give it a job. The request goes something like this. Read the interview notes in this folder. Fill the interview record using its format, one answer per line. Each line holds the owner, the question, the exact quote, the date, how it was given, and the commitment field. Use the evidence-checked-drafting skill. Do not put any words in quotation marks unless they appear in the notes.

The skill is what makes this safe. You built it in an earlier chapter, and it carries the evidence rules. Every claim needs a numbered source and a word-for-word quote. Silence in the sources is written as "not reported", and the work ends with a claim-check list. Here each owner's notes file counts as a numbered source. So every line of the record points back to a file, and every quote can be found in it.

The commitment field is where the real judgment lives, so it helps to be exact about what goes in it. A commitment is something that costs the owner. A deposit counts. So does agreeing to a paid trial at the stated price, or agreeing to route their enquiries into a shared folder. So does telling you a real figure for what they already spend. A compliment costs nothing. "That looks handy" goes in the record as polite interest, with nothing committed. That rule was written onto the sheet before anyone was asked, so a mood after a good call cannot change it.

Here is an illustrative spread. Owner one, a plumber, by phone: "Would've saved me last Tuesday, honestly." Asked about a trial: "Maybe, send it over." Commitment: polite interest. Owner two, an electrician, in person: "Looks good." Asked about a trial: "Not right now." Commitment: none. Owner three, a roofer, by email: "I'd pay for this but not two-seventy. More like a hundred." Commitment: none, with a price objection noted. Owner four, a joiner, by phone: "Nice idea." Commitment: polite interest. Owner five, a heating engineer, in person: "Yes, I'll do the two weeks. Send me the terms." Commitment: agreed to a paid trial at the stated price.

Notice what owner one sounds like. "Would've saved me last Tuesday" feels like proof. It is a real problem, said warmly. But "maybe, send it over" commits nothing, so the field says polite interest. The record holds both the warmth and the lack of commitment, side by side, on separate lines.

Now ask Claude Code for a short summary for your own notes. Suppose its first draft opens like this: "Strong interest from three owners, with one ready to start." That sounds fair. Owners one, four and five were all friendly. But read the claim-check list at the bottom. The phrase "strong interest from three owners" has no record line behind it. No commitment field says strong interest. Two lines say polite interest, and one says agreed to a paid trial. The check marks the sentence unsupported. The skill then forces the draft back to what the lines say: one commitment, a paid trial at the stated price, from owner five. Four committed nothing. Owners one and four showed polite interest.

That correction is worth slowing down for. The agent did not lie. It wrote the kind of summary a hopeful person writes, smoothing three friendly people into one hopeful number. The model is good at fluent, plausible sentences. Fluency is exactly the risk when the subject is your own hopes. The structure catches it. A summary may only say what a record line says, and every summary sentence must cite a line. You do not have to be in a sceptical mood on the day. The check is sceptical for you.

Suppose instead the spread had come back with no trial agreed by anyone. The honest summary then reads "zero commitments, two polite interest". It is a dull sentence, and it is the one you need.

With a clean record, you apply the stopping rules. They were written in an earlier chapter, before any interviews, for this moment. You ask Claude Code to test each rule against the record, cite the lines that decide it, and state the outcome. Then you read its reasoning yourself, because this decision is yours.

The first rule is about price. If owners take a trial only at a lower price, narrow the offer and drop delivery work rather than run below the floor. In the spread above, one owner said a lower figure. The same week, another owner agreed at two hundred and seventy. So the rule's condition, a trial only at a lower price, is not met. Owner three's line stays on file as a warning about price. It is not a verdict.

The second rule is about the parked text-back. If owners keep saying missed calls hurt more than the summary, bring the text-back back. Nobody in the spread said that, so the rule does not fire. Agents tend to overreach here. One plumber saying "would've saved me last Tuesday" is about a missed enquiry. It is not a claim that the text-back matters more. The decision has to cite the actual words.

The third rule is about commitments. If several owners commit nothing, park the digest. Count by the record's own definition: polite interest means nothing was committed. So owners one and four committed nothing, just like owners two and three. That makes four of five, and read as written, the rule says park. You may decide the rule was worded badly. Perhaps one full-price trial should count for more than four warm maybes. If so, that is a revision of the rule, not a reading of it. Write the new wording down, for example "if no owner commits, park the digest", and date it. Note that you changed it after seeing the answers, which is exactly when a rule is easiest to bend. Only under that written revision does the outcome become proceed: send the trial terms to owner five.

Now picture the other two outcomes with different invented answers. Say every owner who showed any interest said some version of "cheaper and I'd try it". Then the first rule fires. You narrow the scope, perhaps cutting the weekday run to fewer evenings or dropping a kind of enquiry. That lowers your cost and keeps the price above the floor. You do not just discount, because two hundred and seventy dollars was set at twice a floor of about a hundred and thirty-three. Discounting toward a hundred puts every delivered digest under water. Or say all five answers were warm and none committed anything. Then the third rule fires, and the digest is parked, however nice the conversations felt.

Each run of this process leaves the same three things. There is a record with one answer per line. There is a summary where every sentence cites a line. And there is a decision that names the rule and the lines that settled it. Next month, with more conversations, you run it again. The record grows, and the decision is made the same way. Any line of work that depends on people saying yes can use this: freelance design, a bakery's wholesale offer, a software tool. The commitment field is the only part that has to be defined for the trade.

The terms and the payment request

Owner five said yes, and asked to see the terms. That yes creates an obligation before any money moves. Once someone pays you, you have to deliver what you said you would, in a form they can hold you to. So the terms get written now, before the payment request goes out.

A paid trial agreement for a small service needs a few things on one page. It needs the deliverables. It needs what is out of scope. It needs the trial's length and price. It needs when payment is due. It needs how either side stops. And it needs what happens when you miss a promise. Each of those prevents a specific argument later. Without an out-of-scope list, "can it text them back too?" turns into free extra work. Without a stopping rule, it is unclear whether the trial quietly becomes a monthly bill. Without a credit for a late digest, a missed evening turns into a disagreement about what is owed.

You already have most of this, because the digest's scope file was written in plain sentences. So you ask Claude Code to draft the agreement from the scope file only, using the evidence-checked-drafting skill, with every term citing the line of scope it came from. A term with no source line gets marked as an open question, not filled in with something that sounds standard.

The draft covers the following. The deliverables: one page each weekday evening by seven, local time, listing every enquiry from the shared folder, unreachable ones first, ending with a count check, for up to thirty enquiries a day. The out-of-scope list: no replies, no texts or emails, no touching the phone, voicemail or website, no booking or quoting, no chasing payments, no weekends, no guessing missing details. The customer's part: they route their voicemail transcripts, form messages and emails into the shared folder. A digest can only list what reaches the folder, so this sentence is what makes the count check fair to both sides. The late-digest credit: a credit for that day, as the scope already promises.

Three terms came back flagged, because the scope file was silent on them. The trial price was one. Say you set it at half the monthly price, a hundred and thirty-five dollars, for the two weeks. That is an illustrative choice, and once you write it into the scope the term can cite it. Payment timing was the second. You decide payment is due before the first digest, so you never deliver work you have not been paid for. The third was how the trial ends. You write that the trial ends after two weeks and does not roll into a monthly charge on its own. Either side can stop by email, and a monthly service starts only if both agree in writing. The draft also needs to say what form the late-digest credit takes. During a trial, a credit can only reduce a later bill if there is a later bill, so the agreement should say plainly that it comes off a later invoice or is paid back. Each of these decisions is yours. The agent's job was to notice that the scope did not answer them.

Some parts of a contract depend on where you and the customer are. These include limits on liability, who owns the work and the data, rules on automatic renewal from a trial into a paid plan, which law governs a dispute, and how customer data must be handled. Rules on auto-renewal are strict in some places. That is one reason the trial here does not auto-renew. Whether a given clause holds up, or whether a required one is missing, is checked against an authoritative source for your jurisdiction, such as your government's guidance for small businesses or a lawyer. It is not settled by the agent. Claude Code can list which clauses depend on local law. It cannot tell you that your draft is enough where you live.

Now the payment. A payment processor is a company that takes a card payment from a customer and later pays the money into your bank account, minus a fee. Stripe is a mainstream one, and it offers two simple ways to ask for money. A Payment Link is a hosted payment page you create in the Stripe Dashboard and send as a web address. An invoice in Stripe Invoicing is a bill emailed to a named customer, with a link to pay. Either works for one trial. A Payment Link needs no customer setup. An invoice ties the payment to a named customer and a line item, which keeps the records tidier as customers add up. Before either works, Stripe needs an activated account in live mode, completed identity checks and a connected bank account. You set those up yourself, as the account owner.

For the trial, the payment link would be a product called something like "two-week trial", with a one-time price of a hundred and thirty-five dollars in US dollars. Collecting the customer's email should be turned on so a receipt goes out. Claude Code can draft all of this: the product name, the description taken from the agreement, and the amount. It can prepare the request. The step where money is asked for or moved is yours. You check three things before anything goes out. The recipient: is this the owner who agreed, at the address they gave? The amount: does it match the price in the agreement? The destination: is the money going to your connected business bank account? An agent that can send payment requests to the wrong person, or for the wrong amount, can do real damage in seconds. Preparing a request is drafting. Sending one is a money decision.

Now follow the hundred and thirty-five dollars. Stripe's standard fee for a domestic US card is two point nine percent plus thirty cents per successful charge. Two point nine percent of a hundred and thirty-five dollars is about three dollars ninety-two. Add thirty cents and the fee is about four dollars twenty-two. So about a hundred and thirty dollars seventy-eight reaches your Stripe balance. An international card adds another one and a half percent, and one more percent if currency has to be converted. If you later refund the charge, Stripe keeps the original fee. That means a refunded trial costs you money even though no service was kept.

The money also does not arrive at once. A new Stripe account's first payout waits seven to fourteen days after the first successful payment, and that wait cannot be waived. After that, US accounts normally pay out two business days after a charge. The bank then takes another day or two to clear it. So the price you charged, the fee, and the cash in your bank are three different numbers on three different dates. A trial paid on day one may not be spendable cash until after the trial has ended.

This is why each payment carries an identifier. Stripe gives each payment a PaymentIntent identifier, starting with the letters p-i. That is the master record of the payment, used for status and refunds. If you use an invoice, it has its own identifier starting with i-n. When the money lands in your bank, it may arrive bundled with other payments as one payout, with an identifier starting with p-o. Without those identifiers, a bank line saying "Stripe, one hundred and thirty dollars seventy-eight" cannot be matched to a customer or an agreement.

So you make a simple payments record, a spreadsheet or a plain file in the project folder. Each line holds the date, the customer, the trial agreement it relates to, the gross amount, the fee, the net amount, and the transaction identifier. The payout identifier is added when the money arrives. One request, one line, one identifier. Later, an agent can match bank lines to this record. It can only do that if the identifiers are there.

The trial agreement is drafted from the scope, with three gaps decided by you and the local-law clauses marked for checking. The payment request is prepared at a hundred and thirty-five dollars, waiting for your check of recipient, amount and destination. The payments record has its columns and no rows. Its first line gets written when a real owner pays.