Reply-handling playbook
DRAFT for Brad's review2026-09-06
When they write back
Every reply lands in exactly one bucket, every bucket has one move, and the moves never improvise past the doctrine: the AI drafts, Brad signs, no means no forever, and correspondence is private.
The privacy correction, first
The front page said prospects stay "a number until they reply," which read as if replying earns you a public name. Wrong, and fixed: replying is a private act. Correspondence never appears on the ledger or the map with a name attached. A prospect is named publicly only under the public-fact rule (their own posting, their own press), by their consent, or in a jointly agreed announcement after a win. The ledger logs that A reply arrived and its bucket, never who.
The buckets
1. No. "Not interested," "remove me," or any recognizable no.
Move: stage closed, forever, same hour. Send NOTHING back: the P.S. promised
"you will never hear from us again," and a confirmation email is hearing from
us again. Exception: a no that contains a direct question gets the question
answered once, ending with confirmation that they are closed out.
Ledger: "a no arrived; honored, logged, closed forever."
2. Not now. Two different animals, split on their words:
- Explicit deferral with a date shape ("try us in Q2", "after the ERP
go-live"): that is consent to ONE future contact at that time. Log
defer_untilon the row, stage staysreplied, calendar it. The future email opens by quoting their own invitation. - Bare "not now" with no invitation: that is a polite no. Closed forever.
3. A question. "How would this work?", "What do you charge?" (they get the same numbers that are posted), "Is this real?" Move: draft the answer within 4 working hours, plain voice, no pressure, one question back at most. Goal is a conversation, not a close. Brad signs and sends same day.
4. A yes. Meeting request or "call me." Move: fastest bucket in the book. Draft a scheduling reply offering two concrete times same-day; Brad sends. The AI preps a one-page brief for Brad: the signal, the letter sent, the public facts on file, the likely {ONE_BORING_JOB}. Brad takes the meeting personally; the AI never attends, never speaks for him live.
5. A referral. "Talk to Jane in IT."
Move: thank the referrer AND ask permission to use their name in one line
("May I tell Jane you pointed me her way?"). With permission: warm letter to
Jane naming the referrer. Without: Jane gets a normal signal-based letter, no
name dropped. The referrer's thread stays open at replied.
6. Anger or a spam accusation. Move: one apology, no argument, no defense of the model. Closed forever, same as a no. AND logged honestly on the public ledger (anonymized): "someone told us off; they were right that unsolicited email is unsolicited; closed forever." The open-hunt brand only works if the misses are logged too.
7. Autoreply / out-of-office. Not a reply. Stage does not flip. Note the return date; nothing re-sends before it, and nothing re-sends after it either (one email is one email); it only delays a bucket-2 deferral we already had.
8. Bounce. Not a reply. Hard bounce: the address is wrong, back to the contact-discovery queue (the one case where a "second email" is really the first). Soft bounce twice: same. Mailbox hygiene matters for the sending domain's reputation, so bounces get processed the day they happen.
Mechanics
- Stage flips: buckets 1-6 flip
contactedtoreplied(then toclosedor onward as the thread resolves). 7 and 8 flip nothing. - SLA: reply drafted within 4 working hours of arrival; Brad sends same working day. Speed is respect, and it is also the best conversion lever we own.
- Disclosure continues: every reply carries the one-line footer that an AI drafted it and a human signed it, same as the letters. The disclosure is the brand; it does not stop at message one.
- SMS: only if they hand over a number and invite it, and then only for logistics ("running 5 late"), never for pitching. (TCPA, and taste.)
- Detection: manual for now (Brad's inbox). When cbemail polling lands, the pipeline flags new replies and queues draft work; the buckets and moves above do not change, only the plumbing that notices.
- Every bucket writes a ledger entry, anonymized, same day.
Reply skeletons (voice-matched, for speed; Brad edits every send)
Bucket 3 (question):
{FIRST_NAME}, fair question. {ANSWER, two paragraphs max, numbers included.} If it is easier to talk it through, I will take 20 minutes whenever suits; if not, the whole operation is documented at housecalls.bradley.io. (Drafted by the AI I operate; reviewed and signed by me.)
Bucket 4 (yes):
{FIRST_NAME}, glad to. {TIME_1} or {TIME_2} this week both work; if neither does, name one and I will make it fit. Nothing to prepare on your end. (Drafted by the AI I operate; reviewed and signed by me.)
Bucket 5 (referral ask):
Thank you, that is genuinely helpful. May I tell {REFERRAL_NAME} you pointed me their way, or would you rather I leave your name out of it? (Drafted by the AI I operate; reviewed and signed by me.)
Out of scope here
What happens after won (kickoff, contracting, the jointly-agreed public
announcement) is its own document when we need it. This playbook ends where
the engagement begins.