agentleFS
search-atlas-groupsearch-atlas-group/amm-founding-circle/skills/client-health-score/SKILL.md

client-health-score

Skill0 starsChanged 5 months agoMITView on GitHub

Rolls engagement, deliverable timeliness, and sentiment signals for each client into one internal early-warning score, so a churn risk gets caught months before the renewal conversation instead of showing up as a surprise cancellation email. Use when a client cancels and it "came out of nowhere," when you want an honest read on which accounts are actually shaky before your next renewal push, or as the internal feed behind your client-facing dashboard's status.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Evidence-based, not vibes-based: Every score change traces to an observable signal, not a hunch — if there's no evidence, the score doesn't …

    line 118

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

client-health-score

The problem this solves: the first sign most agencies get that a client is unhappy is the cancellation email. By then it's over — the decision was made weeks or months earlier, quietly, in a client who stopped replying fast, stopped opening reports, or started asking pointed questions about value. Those signals were sitting in your inbox, your ClickUp thread, your report open-rates the whole time. This skill turns them into one internal number per client, tracked over time, so a cooling relationship shows up as a trend while there's still a renewal conversation left to have — not as a surprise.

A churn risk you catch in month 2 of a slide is a save-able conversation. The same risk caught the week of the renewal invoice is a lost account you're now writing a post-mortem about. This is the difference between those two outcomes.

This is an internal tool — the score itself is never client-facing (see client-dashboard for what a client is allowed to see).


Say this to your agent

"Set up a health score for each client. Pull from: response time to our messages, whether they open/engage with our weekly reports, on-time payment, how many deliverables we shipped vs. planned this month, and any explicit sentiment (complaints, praise, 'checking in' calls they initiate). Score each client weekly, show me the trend, and flag anyone trending down two weeks in a row — even if the current score still looks fine. Never show this score to the client — it's for us."


The signals that actually predict churn (and the ones that don't)

Predictive signalWhy it mattersWhere it usually lives
Reply time slowing downThe first sign someone's checked outEmail, Slack/ClickUp thread
Report opens/engagement droppingThey've stopped believing the work mattersYour reporting tool's open data, or ask directly
On-time payment slippingBudget scrutiny often precedes a cancellationInvoicing/CRM
Deliverables shipped vs. promisedYour own execution gaps create the risk, not just their moodProject board
Unprompted "just checking in" callsCan go either way — flag for a human read, don't auto-score it negativeCalendar, call notes
Explicit complaint, even a small oneWeight this heavily — it's the rare direct signalAny channel

Vanity numbers to leave out: contract length remaining, total lifetime spend, or "how much we like them" — none of those predict whether they are about to leave.


The pattern

  1. Pick the signals once, per client type (retainer vs. project-based clients often need slightly different weights — a project client without a live report to open isn't automatically "disengaged").
  2. Score weekly, on a simple scale (e.g., 1–5 or green/yellow/red) — resist building an elaborate weighted formula before you have a few months of real trend data to tune it against.
  3. Track the trend, not just the snapshot. A client sitting at "yellow" for six straight weeks is a different problem than one that dropped from green to yellow this week — the second is more urgent even though the raw number looks the same or better.
  4. Flag two-consecutive-weeks-down before the score itself looks bad. The whole point is catching the slide early — waiting until the score is already red defeats the purpose.
  5. Route flags to a human conversation, never an automated save-attempt email. The score's job is to tell you who to check in on personally — it should never trigger an autopilot "we miss you" sequence to a paying client.

Weekly health read format

## Client Health — Week of <date>

| Client | Score | Trend | Flag |
|---|---|---|---|
| Acme Co | 4/5 | steady | — |
| Bolt Industries | 3/5 | down 2wk | Reply time up from 4hr to 2 days; report opens dropped to 0 last 2 weeks |
| Crest Partners | 2/5 | down 3wk | Missed 2 of 4 planned deliverables; invoice 12 days late |

Recommend a personal check-in this week: Bolt Industries, Crest Partners

What goes into the score vs. what stays out

IncludeLeave out
Behavioral signals you can actually observe (reply time, opens, payment)Guesses about mood with no evidence behind them
A trend line, not just today's numberOne bad week treated as a permanent downgrade — let it recover before re-flagging
A specific "why" next to every flagA bare red/yellow with no explanation the owner has to go dig for
Explicit complaints, weighted heavilyInternal team gossip about a client ("they seem difficult")

What a good result looks like

  • A cooling client shows up as a two-week downward trend, with a concrete reason, before renewal season — giving you weeks to have the save conversation instead of reading a cancellation notice.
  • The score is boring to check every week precisely because most weeks nothing's wrong — which is exactly why the weeks something IS wrong stand out.
  • Nobody outside your team ever sees this score — it's a compass for where you personally spend attention, not a client-facing artifact.
  • Renewal conversations stop being a surprise either direction — you already know, weeks out, who's solid and who needs a real conversation.

The rules it runs under

  1. Internal-only, always. This score never appears on a client dashboard, in a client email, or in anything the client can see — pairs with, but is the mirror image of, client-dashboard's "never leak internal machinery to the client" rule.
  2. Evidence-based, not vibes-based. Every score change traces to an observable signal, not a hunch — if there's no evidence, the score doesn't move.
  3. Flag early, don't wait for red. A two-week downward trend is worth a look even while the absolute score still reads "fine."
  4. A flag routes to a human conversation, never an automated retention sequence — the fix for a cooling client relationship is a person reaching out, not a drip campaign.
  5. Composes with agency-scorecard (aggregate "clients below threshold" can be one Scorecard row) and feeds the client-safe summary in client-dashboard — but the two are never the same document.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.