Produces the touches and every reply branch for one-to-one outreach, with a research gate that decides whether you may send at all and a rejection branch that stops selling.
We have not measured this one. It ships on the strength of its contents, not against a control.
What it knows that a general instruction does not, starting with the part that is genuinely inverted. The default reply to a rejection tries once more: three further reasons, a discount, a different angle. This file says stop selling completely, ask exactly one question, which is what would have won the business, and reconfirm a specific date to check back. The claim underneath it is that the rejection reply is the highest-yield message in the sequence: a no has already cost the recipient the awkward part, answering one question costs nothing, and the answer is the only place the real objection exists in the buyer's own words.
Around that sit a research gate and a close. The gate is a test rather than an instruction: you may send only if you can name one quantifiable thing you will do for this specific recipient, and if you cannot, the correct output is not a message, it is an hour of research or the recipient removed from the list. The close is directive rather than interrogative, because a question hands the recipient a scheduling task while a proposed time hands them a yes or a no, and the ask carries an odd duration, nine minutes rather than thirty, which is memorable and truthfully small.
One gap is declared rather than papered over. The method this is built from specifies short and persistent and gives no touch count and no interval. Inventing a number and dressing it as a finding is precisely the failure this directory exists to avoid, so the file states four touches over fourteen days, labels it as our default, explains why we chose it, and gives the rule for adjusting it.
Who it is not for. If your motion is volume, this will slow you down, and it is meant to. If the recipient is already warm, most of the machinery is unnecessary. If nothing about the target is findable, the file's own decision rule tells you not to send.
Measuring one skill honestly costs about twenty model sessions: five runs with it, five without, on real material, each output graded alone by a session that is not told the other arm exists, against a rubric written by somebody who never saw the skill. We have not spent that on this one yet, so it ships labelled rather than ships silently.
How it would be measured. Tier A. Both arms receive the same four invented outreach targets, each with a public research pack of a website, a job posting and two published posts, where two targets contain a findable quantified specific and two deliberately do not, plus a scripted reply for each branch. Outputs are graded mechanically for a subject line carrying a quantified term or a referral name, exactly one ask in touch 1 that is not the terminal ask, each touch shorter by word count than the one before it, an imperative close rather than an interrogative one, a rejection branch containing zero product claims and exactly one question, a commitment branch naming a specific weekday and time, a complete forwardable block in any introduction request, and a refusal to send for the two targets with no findable specific.
Most of this is mechanically gradable by someone who never opened the file: whether the subject line carries a quantified term or a referral name, whether touch 1 contains exactly one ask and whether that ask is the terminal one, whether each touch is shorter than the last by word count, whether the close is imperative, whether the rejection branch contains zero product claims. The expensive part is the fixture. A fair test needs invented targets with real-looking public research packs, including at least two where no quantified specific can be found, because refusing to send is a scoreable behaviour and the most likely place a control diverges. It also needs scripted replies for all four branches, a day of writing before a single run. Our honest expectation is that a strong model already writes a decent touch 1 unprompted, and that the gap sits in the branches.
The rule that decides pass or fail was written down before any run was executed and it does not move afterwards. It is in the method note on the hub, along with the full results table including every skill that was tested and cut.
Stated plainly, because a skill that claims everything is useful for nothing.
name and description in the file's frontmatter, so you can also invoke it by name.A warm introduction beats every message in this file, and it is not close. If one person will forward a paragraph to the target, do that instead, and use the sequence only for people you cannot reach that way.
A person with a phone and one hour is the other honest alternative. Two calls to the right switchboard often produce the name, the timing and the current blocker faster than a research pass, and that information is current.
Asking a model directly for a cold email produces something serviceable and predictable: a compliment, a paragraph about the sender, three benefit bullets and a request for a call at your convenience. The pieces this tends not to produce unprompted are the refusal to send when nothing specific can be found, the escalation ladder that keeps the terminal ask out of touch 1, and any branch at all for a no.
Tooling covers the half this ignores. Address finding, verification, deduplication and stopping a sequence when somebody replies are solved problems, worth paying for, and orthogonal to what to write. Buy the tool, and do not let its template library write your first touch.
--- name: cold-outreach-sequence description: Writes a one-to-one cold outreach sequence for a specific named recipient: the research gate that decides whether to send at all, the subject line, an escalating ask across a small number of shortening touches, a directive close with an odd-duration meeting request, and a defined branch for every reply type including acceptance, silence, a pending commitment and rejection. Also covers forwardable introduction requests, same-night follow-up after a live meeting, and press or guest-post pitches. This skill should be used when writing to a stranger to ask for something, when planning the follow-up cadence for outreach that has gone unanswered, or when deciding what to send after a no. --- # Cold outreach sequence ## The claim this skill is built on In one-to-many email the scarce resource is the offer. In one-to-one outreach the scarce resource is the reply, and everything in the sequence is arranged around the cost of producing one. The obvious approach fails in a way that is easy to miss because it looks like effort. You build a list of five hundred plausible targets, write a good template, personalise the first line with the company name, and send. Response is close to zero, so the conclusion drawn is that the volume was too low, and the fix applied is more volume. It does not work, because a generic message depresses response at the point of reading and no amount of follow-up rescues a message that was ignored on its merits. The follow-up inherits the original's problem. So the first move is not writing. It is a gate that most of your list will fail. ## Step 0. The gate: the quantified-specific test You may send to this person only if you can complete this sentence about **this** recipient: > I will [do a specific thing] to [a specific thing of theirs], [by a number, to a number, or within a time]. Three invented examples of the shape, none of them about a real organisation: three of your five locations take bookings by phone only, for six hours a day. Your documentation set has 43 pages and no search. Your onboarding email is 740 words and the only link is in the last paragraph. If you cannot complete that sentence, you are not ready to send, and there are exactly two acceptable outputs. Either do the research, which realistically costs 30 to 60 minutes per recipient, or remove the recipient from the list. The list gets shorter. That is the intended result, and it is the mechanism by which the rest of the sequence becomes affordable. **Why the gate is a gate rather than a preference.** A generic message is not slightly worse than a specific one, it is a different category of message. It is filed as a broadcast on sight, before the second sentence, and that decision cannot be reversed by touch 2. The marginal value of following up on a generic first touch is close to nothing, which is why volume as a strategy keeps disappointing the people running it. ## Step 1. The subject line carries the specific, or the referral name Put the quantified thing in the subject line, expressed in their terms rather than yours. Their terms means the words they would use about their own operation, not your product category. **Where a mutual contact exists, the referral name goes in the subject line instead.** It is the strongest opener available and it displaces everything else. Use one or the other, never both: a subject line carrying a name and a number is doing two jobs and lands as neither. Keep it short enough to survive truncation on a phone, roughly 30 to 45 characters, because mobile clients cut the line at a width you cannot predict and the front is the only part you control. Never write a subject line that becomes a different sentence when truncated. ## Step 2. Touch 1 carries the smallest viable ask Never open with the full request. Write down the terminal ask first, the thing you actually want, then place touch 1 at least two rungs below it. A workable ladder, smallest first: a yes or no confirmation that this is even their problem; a nine-minute call; a look at one artefact you have already made; an introduction; a paid engagement or a contract. If the terminal ask is the contract, touch 1 asks for the call at most. Touch 1 is under 120 words and has three parts: the quantified specific in one or two lines, one line establishing why it is you asking, and the close. There is no paragraph about your company, and there is no list of benefits, because both are answers to questions nobody has asked yet. **Exactly one ask per message.** Two asks is a decision, and a decision gets postponed. ## Step 3. Close with a directive, not a question "Let us talk at 3:15 on Thursday" outperforms "how do you feel about talking sometime soon". The second hands the recipient a scheduling task, which goes onto a list they already resent. The first hands them a yes or a no, which they can answer in three words while walking. Propose exactly one time, not three. Three times is a scheduling task wearing a helpful coat. Add one fallback line inviting them to name a time that works instead, and stop. **Use an odd, specific duration.** Nine minutes rather than thirty. It is memorable because it is not a default, and it truthfully signals a small cost. The rule that comes with it: if you ask for nine minutes, you take nine minutes, and you say so at minute eight. An odd duration you then overrun is worse than a round one, because it converts a precise promise into evidence that your promises are decorative. ## Step 4. The touch count and the interval, which is our default rather than a finding **This is the part where the source method is silent, and rather than invent a number that sounds sourced, here is what we chose and why.** What the method specifies: the follow-ups are short, they get shorter, and persistence beats length. What it does not specify anywhere: how many touches, or how far apart. There is no number to report, so any number presented here as a finding would be fabricated. The following is our default. It is a choice, not a result. **Our default: four touches over fourteen days. Touch 1 on day 0, touch 2 on day 3, touch 3 on day 7, touch 4 on day 14. Then stop.** Three reasons we picked it, and they are reasons rather than citations. First, gaps longer than about a week mean the next message arrives as a fresh cold message rather than as a follow-up, so the sequence loses whatever recognition the earlier touches built. Second, a sequence needs a terminal touch, because a sequence with no end is a subscription the recipient never agreed to, and the fourth touch exists specifically to close the file. Third, touch 3 is where the question changes from "will you" to "why not", and you need at least one touch after that to act on an answer. **The rule for adjusting it**, which matters more than the number: - **Adjust on the value of one yes.** If a single yes is worth more than a month of your time, more touches and longer gaps are justified, up to roughly six touches spread across ninety days. If a yes is worth an hour of your time, two touches is the honest ceiling and the research gate probably should have cut this recipient already. - **Adjust on evidence the address is alive.** An automatic out-of-office reply, a click on a link, or a public role change all say a human is at the other end, and justify one further touch beyond your default. A hard bounce, or a role change away from the problem you named, means stop today rather than at touch 4. - **Any reply exits the schedule.** An explicit no, or a "not now" with a date attached, moves the recipient to the branches below and cancels every remaining scheduled touch. This is the single most common automation failure in outreach and it is worth checking by hand. - **Re-entry, at most once.** A recipient who never replied may be approached once more after roughly ninety days, and only if something has genuinely changed on their side, which means a new quantified specific. The same sequence sent twice is not persistence, it is a bounced cheque. **What would change our default:** running two arms on comparable lists and comparing reply rates per recipient. Until somebody does that, this number is a considered choice and is labelled as one. ## Step 5. Touch 2, 3 and 4 get shorter The mechanical rule: each touch is shorter than its predecessor by word count. If touch 1 is 110 words, touch 2 is under 80, touch 3 under 50, and touch 4 under 30. Persistence beats length, because the response rate is driven by how easy the message is to answer rather than by how completely the case is made. **Touch 2** adds exactly one new fact and repeats the same directive close. New fact means new to them, not a restatement: a second observation from your research, or the artefact itself. **Touch 3 changes the question from "will you" to "why not".** One sentence asking what would need to be true for this to be worth nine minutes, with no product claims attached. This converts a non-answer into information, which is a better return than another attempt at the same yes. **Touch 4 closes the file** in a line or two: you are assuming the timing is wrong, you are stopping, the door is open. This message draws replies at a rate that surprises people the first time, which is the reason it exists as a message rather than as silence. Three things never appear in any follow-up: a re-explanation of the case made in touch 1, the phrase about bumping this to the top of an inbox, and a fake reply prefix on a thread that never had a reply. All three read as pressure, and the third reads as a trick because it is one. ## Step 6. The reply branches, which are the load-bearing part A sequence without branches is a broadcast with delays. Every reply type gets a defined response. ### Accepted Move immediately, the same day. Reinforce that the decision was correct by restating the one quantified thing in their words, then name the next concrete step and its date. Do not celebrate, and do not expand the scope in the acceptance message. Where you can, **do the first piece of work before the agreement is firm**. Finished work is a stronger motivator than the promise of it, and a person looking at the thing behaves differently from a person imagining it. Two constraints keep this from being naive: only do work whose cost you can absorb entirely, and prefer work that is not useful to them without you, because a deliverable that can be taken and used is a deliverable that will be. ### Rejected, and this branch inverts the default **Stop selling.** Not fewer claims, zero. Not one line about what they may be missing. The instinct after a no is to try once more, and continuing to pitch after a no forecloses the follow-up entirely, because it tells the recipient that saying no did not work and the only remaining option is to stop reading you. The rejection reply is three sentences and contains three things: 1. An acknowledgement, with no counter-argument attached. 2. **Exactly one question: what would have won the business.** Not a survey, not two questions, not a request for feedback in general. One specific question, easy to answer in a sentence. 3. **A specific date to check back**, named as a date. Not "down the line", not "in a while". This draws a reply from a substantial minority of the people who just rejected you, which makes it the highest-yield message in the sequence and the reason the branch exists at all. What you get back is the real objection in the buyer's own words, which exists nowhere else, and the person who tells you why is disproportionately the person who returns later. ### Silence Assume the cause is unknown rather than negative. The plausible causes are wrong person, wrong time, message triaged unread, message never delivered, or the reader is on leave. Only one of those is a no. So design the next touch to surface the cause rather than to repeat the case. The highest-value version is a single routing question: if this sits with someone else, who is it. That is answerable by a person who has no interest in your offer and will often be answered anyway, because it costs nothing and it gets you off their desk. ### Commitment pending Someone has said yes in principle and nothing has happened. **Name a specific day and time for the return, never at your convenience**, and remove every step you can from their side: propose the time, write the two-line agenda yourself, attach the thing rather than linking to a place where the thing can be requested. The governing rule is that the next action belongs to you until the meeting exists. Every message that ends with the next action on their side extends the pending state by however long their week is. ## Step 7. When the goal is an introduction, write the forwardable email The person doing you the favour should have to forward, not compose. Asking someone to introduce you and leaving them to write it is a tax they pay for your benefit, and it is why introductions stall. Structure the request in two parts. Above the line, a short note to your contact asking whether they are willing, and giving them an easy way to decline. Below the line, a complete block addressed to the target that they can forward untouched: under 120 words, carrying the quantified specific and the small ask, written in a voice that will not embarrass them. Give both sides a stated reason it makes sense for them. An introduction that only benefits you spends your contact's credibility, and it spends it on the target's side of the relationship rather than yours. Ask both parties before connecting. The double opt-in introduction protects the only asset in the transaction, which is your contact's willingness to do this again. ## Step 8. After a live meeting, follow up the same night Send it the same night, not the following week. Carry something of substance forward: the one thing they said they were stuck on, plus a concrete piece of work, a document chosen for that specific problem, or an answer to the question they could not answer in the room. Name the next date in the same message. A pleasure-to-meet-you note is not a follow-up. It is a receipt, and it puts the next action on their side. ## Step 9. Press and guest-post outreach The same sequence with the direction of value reversed. Lead with what is newsworthy to **their** audience, not with what is new at your end. A launch is not news outside the company. A number, a reversal, a pattern across many cases, or a public failure can be. Offer an exclusive angle only where you can honour it, since an exclusive offered twice ends both relationships at once. Then make the amplification act trivial: draft the piece, supply the assets, the two-sentence version and the caption. The gate still applies, to their publication: name the specific piece of theirs your angle extends, and the gap in it you are filling. ## The decision rule - **You can name a quantified specific.** Send touch 1 today. - **You cannot yet, but the material exists and you have not looked.** Do the research, 30 to 60 minutes, then send. This is the most common branch and it is the one people skip. - **You cannot, nothing is findable publicly, and the recipient is high value.** Change the channel. Ask for a warm route, go where they are, or wait for an event that creates a specific. Do not spend the one first impression on a generic message. - **You cannot, and the recipient is not high value.** Cut them from the list. This is a result, not a failure. - **You cannot tell whether what you have is specific enough.** Run the substitution test: take the draft, replace the recipient's name and organisation with another plausible one in the same category, and read it again. If it still reads as true and sensible, it is generic and will be received as generic, so do not send it. Use this branch by default, because the feeling that a message is personal is not evidence that it is. The sequence itself lives wherever you can actually see it: a spreadsheet, a CRM record, or one text file per recipient. Nothing here depends on a particular mail client or operating system, and a sequence tracked in your head produces the double-send that ends the conversation. ## Worked example, compressed A two-person team builds scheduling software for independent clinics and wants a paid pilot with a mid-sized clinic group. The target is the operations lead, found by name on the group's own site. **Research, 40 minutes.** The group's booking page lists five locations. Three of them have no online booking at all, and those bookings go through a phone line whose published hours are 09:00 to 15:00. **Gate: passed.** The quantified specific is three of five locations taking bookings by phone only, for six hours a day. **Subject line.** "Three of your five sites, phone-only bookings", 43 characters, their terms, no product name. **Touch 1, day 0, 104 words.** Two lines on the specific, one line on why it is them asking, and a directive close: a nine-minute call at 15:15 on Thursday, with one fallback line. The ask is the call, not the pilot, and the pilot is not mentioned. **Touch 2, day 3, 61 words.** One new fact: the phone line's hours end before the two busiest booking hours shown in the group's own published opening times. Same close, different day. **Touch 3, day 7, 38 words.** The why-not question. No claims. **Reply, day 8.** A rejection: they are eighteen months into a contract with another vendor and are not moving. **Rejection branch.** Three sentences, zero product claims, one question about what would have won the business, and a named date to check back in February. The reply arrives the same afternoon: the contract renews in March, the operations lead is not the decision maker for it, and the problem that actually costs them money is not booking at all, it is no-show follow-up. **Verdict: the pilot did not happen this quarter, touch 4 is cancelled, and the file closes until the named February date.** What the sequence produced instead is a dated re-entry point, the name of the person who does own the renewal, and a corrected understanding of which problem is worth building for. That is the rejection branch doing the job it exists for, and it took one message that sold nothing. ## Failure modes **The generic send.** No quantified specific, so the message is filed as a broadcast in the first line. From the outside it looks like a channel that does not work, and the response is more volume, which buys more of the same result. **Selling past the no.** A rejection is answered with three further reasons and a discount. The recipient stops reading permanently, and you forfeit both the answer to the one question and the February date you could have had. **The open-question close.** The message ends by asking how they feel about talking soon. The recipient now owns a scheduling task, which loses to everything else on their list, and the silence that follows gets recorded as disinterest. **Escalation inversion.** Touch 1 asks for the contract, the pilot or the partnership. Every smaller ask afterwards now looks like a retreat, and the sequence has nowhere left to go but down. **The forwarding tax.** An introduction is requested without a forwardable block, so the person doing the favour has to compose one. It sits in their drafts for a fortnight and then does not happen, and nobody involved reads that as a refusal because it was not one. **Length as persistence.** Follow-ups get longer instead of shorter, on the theory that the case was not made clearly enough. It reads as pressure, and the growing word count is the clearest possible signal that nobody is replying. **The unbounded sequence.** No terminal touch, so messages continue until somebody complains. The file is never closed, the recipient is never re-approachable, and the sender believes they are being diligent. **Automation that ignores replies.** The sequencer keeps sending touch 3 after a human answered touch 2. It is the fastest way to destroy a conversation you had already won, and it is invisible from the sender's side until the reply arrives asking whether anyone reads these. **The odd duration you overrun.** You asked for nine minutes and took forty. The precision that made the ask credible now works against you, and the second meeting is harder to get than the first was. ## What this skill does not do - It cannot do the research, and the gate depends entirely on what the research finds. Given nothing, it will invent something plausible, which is worse than not sending. - It does not decide whether you are permitted to send. Consent and unsolicited commercial mail are regulated differently across jurisdictions, and that question is answered before any of this applies. - It does not get the message delivered. Cold sending to strangers is the riskiest sending pattern there is, and deliverability is a separate discipline with its own failure modes. - The touch count and interval are our default, chosen and labelled as such, not a measured result. Adjust them on the value of a yes and on evidence the address is live. - It does not find, verify or deduplicate addresses, and it does not stop a sequencer when a reply arrives. Those are tooling problems with good tooling answers. - It is built for small lists worked by hand. It scales badly on purpose, because the gate is the thing that makes it work.
These skills all ask your assistant to check things against your actual codebase, your actual schema, your actual design system. Locul keeps that context current on its own, from the files you already have, on your machine. Mac and Windows, free to start.