Friday afternoon at most trades shops in 2026 looks the same. The marketing manager finishes the lead-source attribution report (Lesson 1's deliverable) at 10:25 a.m. The owner reads the Friday recap at 4:30 p.m. Somewhere between those two moments, next week's marketing plan is supposed to get drafted. Pre-AI, the drafting was Tuesday-and-Wednesday work — 3-5 hours of spreadsheet edits, vendor portal logins, copy revisions, and email threads with the GLSA agency, the NiceJob admin, and the Hatch sequence editor. The plan landed in the owner's inbox Thursday afternoon and was already 4 days stale by Monday's standup. The 2026 AI-built version compresses the drafting to 25 minutes Friday afternoon, ships before the owner reads the recap, and lands on Monday's standup as a current document grounded in last week's actual attribution data. The trick is not faster typing — it is a system prompt that reads like the marketing manager wrote it because the prompts are theirs. The Hatch nurture sequence rewrite, the GLSA bid adjustment memo, the NiceJob campaign refresh, and the direct mail piece reorder all draft off the prior week's attribution table with the manager's voice and the manager's judgment baked into the prompt structure. This lesson is the named workflow for the second hour of every Friday — between attribution and recap, the 25 minutes that produce next week's marketing plan.

Why the Plan Must Draft in 25 Minutes

The plan that takes 3-5 hours to draft is the plan that doesn't get drafted. The marketing manager has 11 other things on Friday afternoon — review responses queued, the GLSA agency's weekly check-in, an NiceJob photoshoot the truck did Tuesday, a Hatch sequence A/B test report. The 3-5 hour plan slips to Monday, Monday slips to Tuesday, and by Wednesday afternoon next week's plan is half-stale and the spend is already running on autopilot. Most shops in 2026 are still operating in this pattern. They are running the same marketing plan every week with manual tweaks because the planning cadence is broken by lag.

The L3 manager's contract with the owner is weekly reallocation grounded in last week's actual data. That contract requires the plan to draft fast enough that it lands before the owner reads the recap on Friday. 25 minutes is the working number — 10 minutes drafting the channel allocation memo, 10 minutes drafting the Hatch nurture sequence revision, 5 minutes drafting the GLSA bid adjustment memo and any direct mail or NiceJob updates. Manager edits during drafting; AI produces; manager reviews and ships. The 25-minute window fits between the 10:25 a.m. attribution finish and the 4:30 p.m. recap delivery without breaking the manager's other Friday work.

The compression is not magic. It is system-prompt design. The drafting AI knows last week's attribution table (it just produced it), the shop's voice (loaded from prior plans and Friday recaps), the trade-specific journey patterns (in the system prompt from Lesson 1), and the constraints the owner has on spend (total weekly budget, channel min/max, no-go channels, must-have channels). With those four inputs locked in the prompt, the drafting AI produces a channel allocation memo, a Hatch sequence revision, and the supporting memos in the time it takes to read them.

The Channel Allocation Memo — The First 10 Minutes

The channel allocation memo is the centerpiece. One page, six rows, six columns. Rows: each channel from the attribution table. Columns: last week's spend, last week's RPL, last week's confidence, next week's proposed spend, delta, rationale. The bottom of the page has the next week's total spend (must match the owner's weekly budget cap), a 3-sentence narrative summary, and any flagged decisions requiring owner sign-off (above-threshold reallocations, new channel tests, paused channels).

The drafting flow. Marketing manager opens the channel allocation prompt. Pastes last week's attribution table as input. Adds the owner's spend constraints (typically: weekly cap $X, GLSA floor $Y to keep the auction position, direct mail prepaid for the quarter, NiceJob review nudges always on, after-hours service flagged for replacement). Hits run. AI produces the six-row table with proposed spend per channel. Manager edits in 6-8 minutes — adjusting any line where the AI's recommendation conflicts with shop context the AI doesn't see (the new tech starting Monday who needs lower-margin calls to ramp, the seasonal furnace tune-up campaign that needs front-loading, the warranty recall backlog that requires CSR bandwidth above what next week's booking volume would consume).

A representative week's output at the 7-truck HVAC shop from Lesson 1. GLSA: last week spend $4,400 at $814 RPL with 88% confidence; next week proposed spend $4,400 (hold, confidence stable, RPL within trades band). Hatch: last week $450 at $900 RPL with 78% confidence; next week proposed $600 (+$150 reallocation from after-hours service, RPL above blended median). NiceJob: last week $450 at $1,164 RPL with 64% confidence; next week proposed $450 (hold spend; dedicate one campaign to a tracked-only test to tighten attribution). Direct mail: prepaid through quarter, no spend change; flag for tracked-only campaign next quarter. After-hours answering service: last week $1,140 at $600 RPL with 91% confidence; next week proposed $990 (-$150 reallocation to Hatch; pending AI receptionist pilot decision). Total: $7,440 vs. $6,890 prior week, within the owner's $7,500 weekly cap.

The narrative summary at the bottom of the page reads in three sentences. "We reallocated $150 from after-hours answering service to Hatch nurture based on the $900 vs. $600 RPL delta at acceptable confidence. NiceJob spend held with a dedicated tracking-number test next week to tighten the attribution band from 64% toward 80%+. Direct mail flagged for a tracked-only campaign next quarter to defend the 42% confidence band." Owner reads in 30 seconds, approves or pushes back, the conversation is grounded in numbers. No vendor portals were opened in the drafting; that is Monday morning work.

The Hatch Nurture Sequence Rewrite — Minutes 11 Through 20

Hatch's nurture sequences are the second-most-edited artifact in the marketing manager's weekly work. The sequences run on segments — dormant-by-call-type, dormant-by-last-touch-date, dormant-by-quote-amount, dormant-by-equipment-type. Each segment has its own text-and-email cadence. The cadences need rewriting every 4-6 weeks as conversion patterns shift, seasons change, and the segments drift. Pre-AI, a sequence rewrite was a 90-minute Tuesday-afternoon block of staring at prior conversion stats and rewriting copy by hand. Post-AI, the rewrite is a 10-minute Friday flow in the same drafting session as the channel allocation memo.

The mechanic. Marketing manager opens the Hatch sequence prompt. Pastes the prior 4 weeks of Hatch conversion data by segment (which segment is converting, which is flat, which is dropping). Adds the shop voice context (prior text examples that worked, the offer language the owner approved, seasonal triggers — heating season approaching, AC tune-up window opening, panel-upgrade rebate ending). Names the segment being rewritten and the conversion pattern that triggered the rewrite. Hits run. AI produces the revised sequence — 3 text messages and 2 emails for that segment, scheduled across a 14-day cadence, with branching logic based on customer response (replied / didn't reply / opened / unopened).

The voice match is the part that has to be right. Generic SaaS-AI sequence copy reads exactly like SaaS-AI sequence copy — the homeowner's spam filter catches half of it, the open rate drops, and the conversion rate falls. The marketing manager's voice — built into the system prompt from 6-12 months of prior Hatch sequences that performed — preserves the shop's tone, the regional phrasing, the way the manager addresses the homeowner. When the AI-produced sequence reads like a continuation of last quarter's voice, conversion holds. When it reads like generic AI, conversion falls 30-50% within two weeks.

The manager edits the AI-produced sequence in 4-5 minutes. Fixes one or two lines where the AI used a phrase the shop doesn't use ("partner with us" — the shop uses "work with us"). Adjusts a CTA timing where the AI front-loaded too aggressively. Validates the financing-language line against Reg Z/FCRA compliance (any payment amount or APR mentioned must match the actual portal disclosure; AI never invents these). Approves the sequence. Uploads to Hatch's sequence builder via copy-paste in 2-3 minutes. Sequence is live by 11:00 a.m. Friday for Monday morning activation.

The GLSA Bid Adjustment Memo and Supporting Updates

The remaining 5 minutes covers the GLSA bid adjustment memo, any direct mail piece updates, and the NiceJob campaign refresh. Each of these is a short-form document — 100-200 words, structured as a directive to the executing system (GLSA AI bidding, the direct mail printer, NiceJob's campaign editor).

The GLSA bid adjustment memo names the bid floor and ceiling for the week, the zip-code prioritization changes based on RPL by zip (the attribution model surfaces this on demand), the equipment-keyword tuning (last week's HVAC-replacement RPL was higher than HVAC-repair RPL — bid higher on replacement intent keywords next week), and the negative-lead-dispute pattern for the prior week (which leads to dispute as non-service inquiries). The memo's audience is the GLSA AI bidding tool (Ryze AI or the equivalent) and the GLSA agency rep if the shop uses one. Length: 5-8 lines. Drafted in 90 seconds; manager edits in 60 seconds; sent to the GLSA workflow tooling.

Direct mail updates run on a quarterly cadence, not weekly. The 5-minute window only includes direct mail when a piece needs reorder approval, a tracking number swap, or a campaign extension decision. Most weeks the direct mail line is "no change, quarterly campaign running per plan." When a piece needs action, the AI drafts the printer brief with the tracking number, mailing date, target zip cluster, offer language, and the prior piece's RPL with confidence band so the printer reps see the context.

NiceJob campaign refreshes are weekly when a new campaign launches, otherwise monthly. The AI drafts the campaign copy — review request text variants by job type, follow-up sequence for non-responders, escalation script for negative-trending sentiment — in 90 seconds. Manager edits in 60-90 seconds, uploads to NiceJob in 2 minutes. The dedicated tracking-number tests for the NiceJob attribution tightening (called out in the channel allocation memo) get their own CallRail number provisioning, which the manager handles after the 25-minute drafting block.

"Reads Like the Manager Wrote It" — The Voice Discipline

The defining feature of this workflow is the voice match. The plan must read like the marketing manager wrote it because the owner reads it as the manager's plan. If the owner reads the plan and detects generic SaaS-AI phrasing — synergies, leverage, value props, optimization opportunities — the plan is discounted. The owner stops trusting that the marketing manager actually has a plan, and starts assuming the plan is automated boilerplate. Trust collapses faster than it took to build.

The voice discipline lives in the system prompt. The prompt is loaded with 6-12 months of prior plans, prior Friday recaps, prior Hatch sequences, prior NiceJob campaigns — all written by the actual marketing manager. The AI reads the voice from these examples: phrasing patterns, sentence length, transition phrases, the way the manager refers to customers (names, segments, equipment types), the way the manager frames recommendations to the owner ("I recommend" vs. "we should" vs. "considering"). The prompt explicitly forbids the SaaS-AI lexicon — no "synergies," no "leverage," no "value-add," no "optimize for." Trades phrasing only — "the GLSA line moved $X," "cost per closed dropped to $Y," "the NiceJob channel needs a tighter band."

The discipline scales across manager turnover. When the marketing manager leaves, the next manager inherits the voice prompt with 6-12 months of accumulated voice data and edits it forward as their own voice develops. The shop preserves the conversation pattern with the owner across the transition — the owner doesn't relearn how to read marketing reports every time the role changes. Voice is the institutional knowledge the L3 manager builds and the next manager inherits, the same way the attribution-model system prompt from Lesson 1 carries forward.

Flagged Decisions and the Owner Sign-Off Protocol

Three categories of decisions in the weekly plan require owner sign-off before Monday execution. The manager flags these in the plan's narrative summary so the owner reads them in the Friday 4:30 p.m. recap consumption window and can approve or push back before Monday. The categories are documented in the system prompt so the AI surfaces them consistently every week.

Above-threshold reallocations. Any single-channel spend shift above a documented threshold (typical: $500 weekly or 15% of the channel's prior-week spend). The owner sees the proposal, the RPL delta justifying it, and the confidence band. Approves or counters. Most weeks the reallocations are below threshold and run on the manager's signature. The threshold exists so the owner sees material changes without being copied on every $50 shift.

New channel tests. Any campaign in a channel the shop wasn't running last week. A tracked-only direct mail test, a new GLSA campaign type (commercial-keyword set vs. residential, a new geographic ring), a NiceJob influencer-partnership pilot. New tests have higher attribution uncertainty by definition and the owner approves the experiment design and budget cap. The 30/60/90 kill rules from L5 Chapter 2 framing apply.

Paused or killed channels. The decision to pause or end a channel — the after-hours answering service replacement, a poorly-performing Hatch segment, a NiceJob campaign that won't tighten. Pausing has downstream effects (existing scheduled sends, vendor contract terms, attribution-trend disruption) that the owner sees before the manager executes. Most weeks the pause/kill decisions are flagged but not executed — they get queued for the quarterly campaign review unless urgent.

Weekly Cadence and the Monday 8:00 a.m. Handoff

The named workflow runs Friday from approximately 10:30 a.m. to 11:00 a.m. — picked up directly from where Lesson 1's attribution model finished at 10:25 a.m. The 25-minute drafting block produces the channel allocation memo (10 min), the Hatch sequence rewrite (10 min), and the GLSA/direct mail/NiceJob updates (5 min). The plan is in the shop's shared Notion or Google Doc by 11:00 a.m.

The plan is referenced in the Friday 4:30 p.m. recap that the owner reads at the end of Friday. The recap's narrative pulls from the plan's narrative — the "what to do next week" line on the recap is the plan's reallocation summary in one sentence. The owner approves or counters the flagged decisions over the weekend or by Monday 7:30 a.m.

The Monday 8:00 a.m. standup opens with the marketing manager presenting one slide pulled from the channel allocation memo. The owner has already read the recap and the flagged decisions; the standup conversation is short — confirm the reallocation, confirm any new test launch, confirm any pause. By 8:15 a.m. the plan is in execution. GLSA bid adjustments live by 9:00 a.m. Hatch sequence live by 9:30 a.m. NiceJob campaign refreshes by 10:00 a.m. The Monday morning execution cycle compresses from the historical Tuesday-Wednesday-Thursday plan-then-execute pattern into a single morning.

The cycle compounds. 52 reallocation conversations a year vs. the median shop's 4-12. Each conversation grounded in confidence-banded attribution data, voiced as the manager wrote it, executed Monday morning. The compounding ROI is the moat — five years of weekly reallocation produces a marketing P&L line no quarterly-reallocation shop can catch.

Failure Modes and the Manager's Discipline

Three failure modes degrade this workflow if the manager doesn't guard against them.

Failure mode 1: AI-drafted copy ships without voice editing. The manager pastes the AI output directly to Hatch or NiceJob without the 4-5 minute voice review pass. Two weeks later conversion rates have dropped 30-50% and the manager doesn't know why. Fix: every AI-drafted customer-facing copy gets the 4-5 minute voice pass, period. The 25-minute window is calibrated to include this pass; skipping it to save time costs revenue downstream. The L1 Cardinal Rule applies: nothing AI says about a customer goes to the customer without a human eye on it.

Failure mode 2: Plan reallocations chase weekly noise. The manager reallocates $400 to a channel that spiked one week, then reallocates back the next week when the spike reverts. The plan ships smooth from week to week but the marketing P&L is volatile because the reallocations are noise-driven, not signal-driven. Fix: reallocations above the threshold reference the 12-week trend, not just last week. The system prompt is updated to surface the trend alongside the prior-week table. Manager applies the trend-check before approving the reallocation.

Failure mode 3: Voice prompt drift across managers and seasons. The voice prompt from 12 months ago doesn't reflect the current manager's actual phrasing or the current season's tone. The AI produces output that sounds like a prior era. Fix: the voice prompt gets a documented quarterly refresh — pull the prior 4 weeks of actual manager output, add to the prompt's voice corpus, archive the oldest section. The prompt evolves with the manager and the calendar. Quarterly refresh is a 30-minute Friday block once per quarter, not a weekly burden.

The Manager's Deliverable for This Lesson

By the end of this lesson, the marketing manager builds the named weekly drafting workflow. The deliverable has three parts. The first part is the channel allocation prompt — system prompt with shop voice, owner spend constraints, attribution-table input schema, and output template (6 rows × 6 columns table plus narrative). Saved in shared Notion or Google Doc, version-controlled.

The second part is the Hatch nurture sequence prompt — system prompt with shop voice from 6-12 months of prior sequences, segment definitions, offer language library, branching-logic patterns. Loaded with the financing language compliance lock (no invented APRs, no fabricated payment amounts). Saved alongside the channel allocation prompt.

The third part is the GLSA / direct mail / NiceJob support prompts. Short-form, structured to produce the bid adjustment memo, direct mail printer brief, and NiceJob campaign refresh. Each saved with its own template.

Three months in, the shop has 12 weeks of plan documents, 12 weeks of voice-evolution data in the prompt corpus, and a Monday-execution cadence that the owner trusts. The marketing manager's role has decisively shifted from data-assembler-and-drafter to channel-strategist with weekly defended decisions. The L3 contract is in place. The L4 chapter on marketing strategy at shop scale takes this workflow as a substrate; the L5 chapter on multi-shop AI rollout treats this as a per-location capability that centralizes once 5+ locations standardize.

Key Takeaways

• The 25-minute drafting block runs Friday between the 10:25 a.m. attribution finish and the 4:30 p.m. owner recap. 10 minutes channel allocation memo, 10 minutes Hatch sequence rewrite, 5 minutes GLSA/direct mail/NiceJob updates. Plan ships by 11:00 a.m. and is referenced in the recap the owner reads at end of day.
• The plan reads like the manager wrote it because the prompts are theirs. System prompt loaded with 6-12 months of prior plans, recaps, Hatch sequences, and NiceJob campaigns from the actual marketing manager. AI mirrors voice. SaaS-AI lexicon (synergies, leverage, value-add, optimize for) explicitly forbidden in the prompt.
• The channel allocation memo is one page, six rows, six columns. Last week's spend, last week's RPL, last week's confidence, next week's proposed spend, delta, rationale. Narrative summary in 3 sentences. Flagged decisions for owner sign-off named.
• Three categories require owner sign-off: above-threshold reallocations (typically $500/week or 15% of channel spend), new channel tests (any campaign in a channel the shop wasn't running last week), and paused or killed channels.
• The Hatch nurture sequence rewrite drafts in 10 minutes vs. 90 minutes pre-AI. AI produces text and email cadence by segment; manager edits 4-5 minutes for voice, CTA timing, and financing-language compliance lock (Reg Z/FCRA; no invented APRs or payment amounts).
• Monday 8:00 a.m. standup opens with the channel allocation slide. GLSA bid adjustments live by 9:00 a.m., Hatch sequence live by 9:30 a.m., NiceJob refreshes by 10:00 a.m. Monday morning execution compresses the historical Tuesday-Wednesday-Thursday plan-then-execute cycle into one morning.
• Three failure modes the manager guards against: AI-drafted copy shipping without voice editing (conversion drops 30-50% in two weeks); plan reallocations chasing weekly noise instead of 12-week trend; voice prompt drift across managers and seasons (quarterly refresh, 30 minutes Friday once per quarter).
• The compounding moat is 52 reallocation conversations a year vs. the median shop's 4-12. Each grounded in confidence-banded attribution data, voiced as the manager wrote it, executed Monday morning. Five years of weekly cadence produces a marketing P&L line no quarterly-reallocation shop can catch.