Flowscale
FOR AGENCY OWNERS WHO WANT TO KEEP CLIENTS · MODULE #7 OF FLOWSCALE · RUNS IN YOUR OWN CLAUDE

// THE RETENTION MODULE · THE WHOLE RETAIN ROOM

Clients leave quietly.
You find out at the cancel.

The account that churns next month is the one you went quiet on three weeks ago. This module scores every client's health out of 100, raises the churn alarm before the client feels it, flags the work that's gone stale, and surfaces the upsells that are ready, each with the angle. So you catch the slip early and pitch at the right moment. $97 USD once, in your own Claude.

See what's inside ↓

Runs entirely in YOUR Claude account · get-you-running-or-refund guarantee · keep the files either way

// THE PART NOBODY RUNS ON PURPOSE

Keeping a client is cheaper than winning one. So why does retention never get done?

Because retention is invisible work. Landing a new logo feels like progress. Checking on a quiet account feels like nothing, until the account is gone. The math says keeping the client you have beats chasing a new one, but the calendar never makes room for it, so it doesn't happen.

And clients don't churn with a warning. They churn quietly: the replies that get slower, the call that gets skipped, the results dip that nobody flagged, the invoice that slides a week late. Each one alone looks like a busy week. Together they're a client who has already decided.

By the time you feel it, the decision is made. The whole game is seeing the slip while it's still three small signals and not yet a cancellation email. That's the room this module gives you.

// WHAT IT ACTUALLY IS

Two real skills and a brain that gets smarter, in your own AI.

A "skill" is a file of operating instructions your Claude reads and follows. This module installs two, plus a starter brain you grow over time. You type a sentence; the right skill loads and reads the client files you already keep.

SKILL 1 · HEALTH-CHECK

every client → a score out of 100

Walks every client, recomputes a 0-100 score from real signals (contact recency, deliverable cadence, results trend, payment), and sorts the book worst-first. One command rescores everything and tells you why each score moved.

SKILL 2 · UPSELL-FINDER

the book → who's ready to expand

Scans health, contract, and delivery to surface who's in the window for an upsell, with the specific angle for each. It deliberately skips anyone with an open problem: fix first, expand later. Never pitch more work to an unhappy client.

THE PART THAT MAKES IT TRUSTWORTHY

The score is computed, not guessed. It starts at 100 and drops by fixed amounts for each bad signal: silence, missed deliverables, a declining campaign, a late payment. You can read the exact rubric in the skill. And every move is explained: "Lakeline 86 to 71: 18 days no contact (-15)". A reported number with no reason is a vibe. A reported number with the signal that moved it is a decision you can act on.

PLUS · A BRAIN THAT COMPOUNDS

Two starter knowledge files, churn-signals.md and upsell-triggers.md, hold generic warning signs and timing patterns. Every time you lose a client or land an expansion, you add what you learned. Over a year this becomes your own churn-and-upsell playbook, in your own words.

// THE OUTPUT, NOT A MOCKUP

This is the Monday-morning loop.

You type plain English into a terminal; the right skill loads and reads the client folders you already keep. No dashboard to log into, no report to build by hand. This is the gut-feel review you keep meaning to do, run for you in a minute.

your-agency/ · THE RETENTION LOOP
You: How healthy are my accounts?
  -> reads every clients/{slug}/health.yaml, recomputes each score
  -> prints worst-first: score, change since last run, churn risk, days since contact

You: Any churn risks?
  -> filters to high + critical, with the reason each is at risk

You: Review health for Lakeline Roofing
  -> deep-reads that one client, rescores, explains the move (86 to 71)

You: What upsells are ready?
  -> surfaces the clients in the right window, each with a specific angle

Nothing sends. Every save-the-account email and upsell pitch is a draft written to a file, for you to send from your own inbox in your own voice. The system never contacts a client on its own.

// WHAT SHIPS

Every file, annotated. No mystery box.

my-agency/ · THE RETENTION MODULE
├── clients/
│   ├── README.md <- the health.yaml format + scoring rubric
│   ├── _templates/ <- blank health.yaml + upsells.md you copy per client
│   └── lakeline-roofing/ <- a full worked example, health + upsells
│      ├── health.yaml <- score 0-100, signals, churn risk
│      └── upsells.md <- running expansion list, each with an angle
├── brain/
│   ├── churn-signals.md <- starter library of warning signs you grow
│   └── upsell-triggers.md <- starter library of timing patterns you grow
└── .claude/skills/
    ├── health-check/ <- the scorer + rubric: book in, worst-first out
    └── upsell-finder/ <- who's ready to expand, and the angle

Plus the install walkthrough on the delivery page, written for someone who has never opened a terminal. You get the source in copy-paste blocks AND a zip. Edit anything; it's yours.

// WHAT YOU ACTUALLY NEED · READ THIS BEFORE BUYING

The honest requirements.

Why it works this way: I never resell Claude access, and your client list never touches my servers. The module runs entirely inside YOUR account. That's the design, not a disclaimer. You wanted in on AI. This is what in looks like: the real stack, in your hands.

// HOW IT WORKS

Install, point it at your clients, then run it every Monday.

01

Install

Unzip into a folder, install Claude Code, log in. About 5 minutes with the walkthrough.

02

Point it at your book

Copy a health.yaml into each client folder you already keep. The template makes it a fill-in. 10 minutes, once.

03

Score, then act

"How healthy are my accounts?" every Monday. Save the at-risk ones, pitch the ready ones. Both as drafts in your repo.

THE SYSTEM BEHIND IT

This module is the retention discipline I run my own agency on. The same worst-first review, pointed at my own book, is how a quiet account gets caught at week three instead of at the cancel. It takes the place of the gut-feel review you keep meaning to do and never do, and it pays for itself the first time it flags an account you would have lost.

// THE MATH

$97 USD. Once.

One saved account pays for this many times over. Losing a $2,000/mo USD client costs you $24,000 USD a year; this is $97 USD to see that client going cold while you can still fix it. Unlimited clients, no monthly fee to me. It's one of seven rooms; when you want them connected, OS Pro is $997 USD.

The guarantee: if you can't get it running, I will personally get you running, or refund the $97. And you keep the files either way. The only way you lose money here is by letting an account churn quietly.

// STRAIGHT ANSWERS

Questions a smart buyer asks.

How does it score health without my data?+

It reads the client files YOU keep: the contact dates, the deliverables, the results. Nothing leaves your account. And the score is transparent, not a black box: it starts at 100 and drops by fixed amounts for each bad signal, so you always see exactly why it moved. "Cedar Park 100 to 92: one deliverable a week late (-5), payment on time (no change)." You can read the full rubric in the skill and change it.

I've never opened a terminal in my life.+

Then the walkthrough was written for you. It's click-by-click: install one app, paste one command, log in with the Claude account you already have. The whole point is that the machine does the technical part. About 25 minutes, once, most of it copying a health.yaml into each client folder. Stuck? Email me and I'll get you running, that's the guarantee.

Is this just a prompt pack?+

No. A prompt pack is text you paste and hope. These are skills: structured procedures your Claude loads and executes, with a scoring rubric, output formats they follow every time, and rules they can't break (worst-first always, explain every score move, never upsell into a problem, nothing sends). The difference shows up on client #30, when every score is still computed the same way and you can still trust the worst-first list.

Does it ever email my clients?+

No, and that's a hard rule. The module writes assessments and drafts to files. Any save-the-account email or upsell pitch is a draft you read, edit, and send from your own inbox, in your own voice. The system never contacts a client on its own. It tells you what's at risk and hands you the draft; the send is always yours.

What does running it actually cost me?+

The Claude plan, paid to Anthropic, not me. A weekly health check across a normal book is light; if you run a large agency, Anthropic's higher tiers exist. For real client data, use a commercial tier (Team, Enterprise, or API), not the free plan. Either way you pay Anthropic. There is no per-file fee and no subscription to me, ever.

Where does this fit in the bigger system?+

This is module #7 (RETAIN), the last room of the agency operating system. It reads the clients/ folder the other modules fill: the onboarding module creates the health.yaml this one reads (new clients start at 100), and the delivery module fills the deliverables/ that feed the campaign and on-time signals. It stands alone, point it at the client folders you already keep. When you want all seven wired into one system with a shared brain, OS Pro ($997 USD) merges them.

// FOR THE SCROLLERS

The short version: the Retention Module is the whole RETAIN stage of a real agency, installed in your own Claude. It scores every client's health out of 100 from real signals, sorts your book worst-first, raises the churn alarm before the client feels it (14 days quiet is AT RISK, 21 is CRITICAL), flags stale work, and surfaces every ready upsell with the angle. The score is computed and every move is explained, no vibes. Nothing sends; every save and pitch is a draft in your voice. You need Claude Pro ($20/mo USD, from Anthropic) and about 25 minutes of setup with a hand-holding walkthrough. $97 USD once, unlimited clients, keep-the-files guarantee.

// THE LAST ROOM

That was the last room. Now connect all eight.

Retention closes the loop: HUNT, QUALIFY, APPROACH, CAPTURE, CLOSE, ONBOARD, DELIVER, and now RETAIN. Each room is a sibling module that shares these exact files, the health.yaml this one reads is the same one the onboarding module writes the night a client signs. They release one at a time, and you don't need all of them to use this one. But when you have the rooms, the next move is to wire them into one system. OS Pro ($997 USD) merges all eight, HUNT through RETAIN, with a brain that routes the whole loop.