01 Case file 02 · all work

A multi-market publishing operation. A one-person week.

A US roofing brand ran a real local-search operation across several metros — Google Business Profile posts, photos, review responses, social, a monthly leadership report — all by hand, all owned by one person, all competing with actual marketing for the same few hours. The work was good. The math wasn't.

SectorHome services · roofing (multi-location)

SizeMarketing team of one

Lines engagedAI Systems & Agents · Demand

DurationMulti-month · 2-week sprints

02 The situation

What the diagnostic found

THE WEEKLY LOAD — AS MAPPED, WEEK ONEDETAIL REDACTED FOR ANONYMITY · SHAPE PRESERVED

The tell was repetition. Every hour mapped in the first three bars followed the same rules week after week — the same source idea turned into a Google Business Profile post for each metro, the same platform-specific social cutdowns, the same monthly report assembled from the same dashboards. The outlined bar is the actual marketing; it was getting the leftovers.

One person owned all of it — GBP posts and photos for every market, review responses, social across three platforms, and a monthly leadership report. Everything had to clear one inbox for approval before it went live, so the queue backed up and the cadence slipped: some channels saw a handful of posts in a month, others hadn't been touched in weeks.

Rule-bound work done by hand isn't a workload. It's a system that hasn't been built yet.

I didn't start with a tool; I started by mapping where the hours went. Most sat in work that never varied — same inputs, same formats, same checks, repeated per location — plus a bottleneck where every draft waited on the founder. What made this worth publishing: it proves the systems line works outside tech, for a local-market brand with no technical staff and no interest in becoming one. I built for the person they already had, not the hire they couldn't justify.

03 What got installed

Two units, one operating layer

Scoped deliberately small: the two systems with the highest hours-returned, built and handed over before anything else was discussed. Ambition is cheap; a running system is not.

INSTALL RECORDALL SYSTEMS IN CLIENT ACCOUNTS · OWNED BY THE CLIENT

I-01

Content repurposing system

One idea in, a compliant channel set out. A brand-trained assistant — loaded with the platform guide, the voice-and-rules doc, and a seasonal marketing-intelligence brief — turns a single idea into platform-specific copy for Facebook, Instagram and LinkedIn plus a GBP post per metro. A compliance pass catches the language Google quietly removes (competitor call-outs, unverifiable claims) before anything is scheduled. Each item lands in a review queue as two or three options; a person picks and approves.

Line AI Systems & Agents

I-02

Reporting engine & leadership dashboard

The monthly report, compiled before anyone asked. A single dashboard pulling budget, marketing spend, local-rank tracking and sprint status into one plain-language view — with grid coverage maps per metro — on a weekly Friday cadence. The founders got direct access; the number they used to wait for was just there.

Line AI Systems & Agents + Demand

I-03

Operating layer & handoff

Built for the person they already had. An SOP for every recurring task, scheduling rights handed to the operator, and a draft → review → approve workflow that removed the founder from the critical path — all in the client's own accounts. One person now runs the whole operation solo, no new technical hire required.

Line AI Systems & Agents

The handoff is the deliverable. Both systems passed the ownership test: the client's own person running them, documentation standing in for me, my access gone. What stayed is theirs.

04 What it did

The numbers, with their timeframes

Real operational figures from the engagement, anonymized. Fixed-width type is reserved for numbers that happened — these did.

Posting cadence · per metro

3–5 / wk▲ steady

from ≈0–1 / wk, backed up in approvals

GBP posts, every active metro, on schedule

Content production

1 wk / session▲ batched

from post-by-post, by hand

A full week of platform-specific, brand-compliant copy for three channels — drafted in one working session

Approval path

self-servebottleneck cleared

from founder-gated, every post

Draft → AI review → 2–3 options → approve. Nothing waits on one inbox; nothing auto-publishes

Output · one 2-week sprint

12 posts · 18 photos

one operator, three metros, all reviewed

Plus every review responded to — the volume a small team used to aspire to

Reviews · flagship metro

109▲ +20 net

from 89 at the start · peaked at 116

Google stripped a batch off the peak — the net gain held; see the honest line

Metros on the system

3 live · 4th building▲ scaling

from 1 tracked at the start

Same system replicated per location as the brand expands

THE HONEST LINE — WHAT DIDN'T GO TO PLAN

The first drafts went out too fast: AI-written posts published without fact-checking (an unverified storm claim among them), and photos got paired with the wrong copy — both fixed by adding a mandatory verification step before anything schedules. The compliance pre-check threw false positives early and needed tuning against real GBP takedowns. Review growth was real but not durable — Google stripped a batch back from 116 to 109, and the incentivized-review program needed constant babysitting to stay honest. And the honest limit: I can prove the time returned and the cadence, but attributing downstream revenue to the publishing system specifically is still maturing — though the engagement's lead-intake system did convert at least one job in the $15–16K range.

05 Next step

Which process eats your week?

If you can name it — the report, the repurposing, the upkeep — I can probably measure it, and if it runs on rules, I can probably build it. The diagnostic maps where the hours go and ranks what's worth automating. Tell me what's eating the time.

Get in touch

CAPACITY — [N] BUILD SLOTS OPEN · [QUARTER]