Marketing agency

Business

Owner-led marketing agency. Strategy, paid ads, social content, and websites for local businesses. Every client starts with the same basics: a website, a Google Business Profile, and social media accounts.

Clients

A lot of local businesses, mostly in hospitality and services, and lately some larger businesses with several locations.

Team

The owner, with an outside video partner who cuts reels and someone who usually handles his social content. [Headcount to confirm]

Systems

Claude, Google Workspace, Apple Calendar, Meta Business Suite and Ads Manager, Google Ads, Squarespace, Final Cut Pro, and Photoshop

Before the build

He used Claude mostly to build strategy decks and plan campaigns in long chats, with almost nothing custom set up, and expected it could do more without knowing how yet.

Engagement

Our first engagement, run by Phin. An audit of the owner's client work came first. The build follows one workflow at a time, starting with editing, and works through his own Claude account and the Google and Meta tools he already uses.

Brought in by

The owner

Status

In progress. Updated Oct 28, 2026.

For each client, the owner works out a strategy, then makes the creative, launches it, and reports on it. Strategy and onboarding were already his strength. By his account, most of his hours went into production: getting footage, editing it, building each client's content calendar, posting, and watching how the posts did. When a campaign wrapped, clients wanted a report on what had worked, and he put it together himself with little energy left. He kept most of this in his head or on paper. On one of our calls he stepped away to shoot photos for a client's social media himself. That, he said, was the kind of task he needed help with. His client list grew to include larger businesses, and the work stopped having a slow season. He did not want to hire, because each client needed something different. He had tried AI marketing tools that promised to build every ad from a folder of assets. One zoomed in on a real photo of a client's dish and left it looking artificial. So the work had to start from real footage of real people. Quick everyday phone content mattered more to him than cinematic polish, which he saves for premium jobs and events. The quality still had to hold, and each client's brand had to stay its own.

In October we had a short intro call with the owner, then a screen-share. On his own screen he took us through Final Cut Pro and Photoshop, the Meta and Google ad accounts he runs for clients, the Squarespace sites he built, and the Claude chats where he works out personas and channels. We went in with questions on onboarding time, creative distribution, client communication, and turnaround. A later call, where we walked him through the plan, added how his footage reaches him. We found that:

• 01the bottleneck is volume: every client wants a steady stream of content, each in its own brand
• 02by his account, producing and posting that content takes most of his hours and client reports come second, so the build starts with content
• 03creative arrives in pieces: a friend texts him ads that he saves to a folder and loads by hand, and his video partner gets footage one reel at a time
• 04strategy and onboarding already work, so they stay with him, as do day-to-day ad decisions, client relationships, and in-person work such as helping a restaurant client launch a loyalty program
• 05the people running his client businesses can film on their phones but can't edit or caption, so phone clips are the raw material
• 06his footage is spread across his phone's camera roll, Apple Photos, project folders, and Google Drive
• 07the Squarespace sites he built give Claude no way to edit pages, so website changes stay manual for now, with Claude drafting copy for him to paste
• 08he works across a pile of tools and browser tabs and wants one place to work from
• 09he wants to see what stage each client is at, which clients might leave, and which accounts take the most time for what they pay

Each workflow is sorted step by step into delete, plain code, agent, and human, then built inside the systems already in use.

• Editing and captioning

  Live

  The owner gives Claude a phone clip. A small speech model transcribes it offline, and plain code drops the dead air. Claude cuts false starts and repeats. Then it plans what goes on screen: the words that get the client's accent color, a hook if the opening needs one, and a graphic only when the speaker walks through steps or a before-and-after. Often the answer is captions alone. For most clips he gets one review that lists each cut, the final script as captioned, and what appears on screen when. A client's first reel, or a clip he says matters, gets a short back-and-forth before planning. Once he approves, Claude checks still frames to keep graphics off faces and captions clear of Instagram's buttons, and plain code renders the vertical reel. His notes on taste go into that client's preferences file, which Claude reads before the next clip. Captions match what was said, apart from fixes to misheard words like names, and Claude never generates or alters faces, food, or places.

• Client project setup

  Live

  A skill (a set of instructions saved in his Claude account) creates a project and folders for a new client, ready to hold that client's brief, brand voice, personas, and files. Claude's connectors (its links to outside tools such as Meta Ads) are shared across every project, so each client's project notes which ad account is theirs. Phin does his own building in a separate project in that account, so the owner's existing work stays untouched.

• Ad reporting

  In progress

  The owner is connecting the ad accounts he runs for clients to his Claude account, following Phin's guide and approving each sign-in prompt himself. Meta comes first, through the official Meta Ads connector, and goes live once he finishes its setup. Then he can ask how any client's ads are doing, or have Claude build a campaign for him to check. Claude changes no live ad's budget or status without his yes. Uploading a client's customer list to build an audience stays out of any automation until consent is checked. Google Ads comes next, through a read-only connection. For any change there, Claude gives him the exact steps and he makes it in Google Ads himself. Monthly client reports that Claude drafts from his ad results are an add-on he can choose after that.

• Posting and scheduling

  Next

  The official Meta Ads connector handles ads but not regular posts, so posting needs a connector of its own. It will run on a small hosted server, because posts have to go out while his Mac is asleep and Instagram pulls each video from a public link. In his Claude account it shows up as one more connector. Each new post waits for his yes, given in Claude, and the server publishes approved posts to Facebook and Instagram on time. One dashboard will show every client's schedule and how each post did. When a post becomes an ad, the ad starts paused and spends nothing until he turns it on. Each client's ad account will also get a Meta spending limit as a backstop. Clients who post for their own business get the captioned clip and suggested copy back instead. Phin tests publishing on his own accounts first. Then they check one client's Page, with nothing posted, to see whether the server needs Meta's approval before it can post for clients.

1. Oct 14

   Intro call

   He said most of how he worked wasn't written down anywhere. Phin floated a record for each client and a home base in Claude, and the owner was excited by the idea. Afterward Phin set a screen-share walkthrough as the next step. He also decided the owner would keep working in tools he already uses, with anything new reaching him through his Claude account, which ruled out Notion.

2. Oct 16

   Screen-share and a call back

   The owner gave Phin a tour of his tools and ad accounts on screen, then picked the conversation back up from his car. Content volume came out as the bottleneck. Phin described a test he had run that week: a clip of himself in a restaurant, shot on a phone, that a Claude skill cut, captioned, and animated. The owner was keen to see it, and they agreed the build would edit and distribute real footage and never generate it. He also floated a login where his clients could do some of the work themselves. Phin dropped it the same day, so everything runs in the owner's own Claude account.

3. Oct 16

   Editing skill, first version

   Phin built the skill that turns a phone clip into a cut, captioned vertical reel, and tested it on his own footage. The first version stopped for approval at the cut list and again at a preview sheet. The same afternoon he reworked it so a typical clip needs one review, since the owner's time is the constraint.

4. Oct 24

   Plan walkthrough

   Phin walked the owner through the plan on a follow-up call. The owner brought up the editing demo without being asked and said he had already shown it to one of his restaurant clients. He showed reels he edits by hand for another client and asked how footage from his camera roll, or a better camera, would get into the system. Meta came first for him, with client campaigns close to launch, and he offered to show Phin how he builds a campaign by hand. He let Phin into his Claude account through an emailed sign-in link, so no password changed hands. Then Phin installed the editing and captioning skill and the client project setup skill there, and wrote guides for connecting Meta Ads and for getting clips off his phone.

5. Oct 24

   Posting plan

   Phin settled how posting would work: a small hosted server that holds each client's posts and publishes nothing until the owner approves it. Meta's own assistant for small businesses can post, but Claude can't reach it, and when Phin asked, the assistant itself recommended building a connector. Buying a scheduler that already connects to Claude is still an option.

6. Oct 28

   Where it stands

   Editing and captioning are live in the owner's Claude account. Ad reporting waits on him finishing the Meta Ads setup, and the posting server is next.

• Held calls with the owner, including a screen-share of his tools and ad accounts
• Wrote the script for the screen-share walkthrough
• Built the editing and captioning skill and tested it on Phin's own footage
• Reworked the skill so a typical clip needs one review from the owner
• Adapted the skill for Claude's Cowork mode and added a check that reports what a setup is missing without installing anything
• Built a deck and a web page for him showing the system map, one clip's path from phone to post, and a day in his Claude account
• Installed the editing and captioning skill and the client project setup skill in his Claude account
• Wrote step-by-step guides for connecting Meta Ads and for getting phone clips into Claude
• Gave him a next-steps document with the plan, the add-ons he can choose, and his checklist
• Wrote up how the posting server will work, including the approval rule and the paused-ad rule

Editing time per video

For each video made with the skill, Phin logs how long the edit took next to how long that kind of video usually takes.

Owner hours a week on content and reports

His own rough estimate of his whole week is the starting point, and it stays labeled as an estimate. Phin then monitors his workflow and splits his week by task, and tracked hours replace the estimate before any figure is published.

Owner hours a week on the pilot client

He notes roughly how many hours a week that client takes him now. The client then runs through the system for a month, and we compare his hours with that baseline.

Time from raw clip to posted reel

Measured once posting is live, from the clip arriving to the post going out.

Days to send a client report

Measured if he adds the monthly reports Claude drafts from his ad results. It counts from the end of the period a report covers, against how long his reports take today.

Posts published per client per month

Counted for each client once posting is live.

The return he is after is capacity: a steady stream of content for each client without hiring, while strategy and client relationships stay with him. It will be measured in his time, starting with his hours each week on content and reports, against a baseline. The figures publish at the end of the proof window, once he agrees which ones can be public.

The owner finishes the Meta Ads setup and sends real footage from his clients, including a long interview to cut into several short videos. Google Ads comes next, then Phin builds the posting server. Before the pilot, the owner picks which client goes first and notes roughly how many hours a week it takes him now. He also decides who owns the Meta app and the hosting account for the posting server. Phin recommends putting both under the agency, with Phin added, so the system belongs to the owner. The editing skill still needs a widescreen version for walkthrough videos and brand settings pulled from each client's project. In his plan, he marks each add-on yes, later, or no. The add-ons are post copy written to draw comments and shares, automatic DMs for people who comment a keyword, an email router, a morning brief, monthly client reports that Claude drafts from his ad results, and a content calendar.