Skip to content
BranchesTags
 
 

Repository files navigation

Outcome Kit

An open-source AI outcome analysis system that finds fake winners, real winners, and conversion leaks, even when your tracking is messy.

Built for OpenClaw and Claude Cowork style agent workflows.

License: MIT OpenClaw Claude Cowork

What It Does

Ads + Pages + Outcome Signals → Real Winners → Fake Winners → Leaks → Next Moves

Outcome Kit reads your ad data, landing page data, and whatever business outcome signal you actually have, leads, bookings, signups, activations, purchases, revenue.

Then it tells you:

  1. What’s actually working — the angles, creatives, and pages driving outcomes
  2. What only looks like it’s working — high CTR, cheap leads, weak downstream quality
  3. Where the leak is — ad, page, or follow-through
  4. What to do next — scale, cut, rewrite, rebuild, or test

The result: a daily operator brief that gives you decision-grade truth, not dashboard theater.

Why This Exists

Most teams don’t have a data problem.

They have a fake winner problem.

The ad with the best CTR often brings the worst buyers. The cheapest lead often books the fewest calls. The campaign report that looks great in-platform often hides the weakest downstream outcomes.

And most businesses do not have perfect attribution. No clean CRM. Broken UTMs. Partial tracking. Calendly bookings in one place, ad spend in another, landing page data somewhere else.

That’s normal.

Outcome Kit exists because you still need to make decisions anyway.

This system doesn’t wait for perfect infrastructure. It uses the signals you do have and turns them into a clear read on:

  • real winners
  • fake winners
  • leaks
  • underfed opportunities

So you can stop trusting vanity metrics and start acting on what’s actually producing outcomes.

What a Real Morning Looks Like

7:08am. Telegram from the agent:

OUTCOME KIT — DAILY BRIEF
Primary outcome: Demo Bookings
Secondary signal: Show Rate
Confidence: MEDIUM

REAL WINNER
• "Stop guessing" angle
  - 18% lower cost per booking
  - 24% higher show rate
  - strongest blended outcome score

FAKE WINNER
• "Save time" creative family
  - highest CTR in Meta
  - weakest booking rate from page visit
  - likely curiosity traffic

LEAK
• Founder-led psychology ads → generic SaaS page
  - ads pull strong clicks
  - page converts 34% below account average
  - message/page mismatch

UNDERFED WINNER
• Competitive teardown angle
  - low spend
  - highest booking efficiency
  - only 9% of budget

NEXT 3 MOVES
1. Shift 15% budget from "Save time" to teardown angle
2. Build a new page for founder-led psychology traffic
3. Generate 4 new creatives from the "Stop guessing" family

That’s the product.

Not another dashboard. Not 14 tabs open across Meta, GA4, and your booking tool. Not "it depends."

Just:

  • what’s real
  • what’s fake
  • what’s leaking
  • what to do next

Who This Is For

Outcome Kit is for operators with enough data to be confused, but not enough infrastructure to be certain.

That usually means:

  • SaaS founders
  • agencies
  • coaches / consultants
  • lean in-house marketing teams
  • service businesses
  • e-commerce brands with partial post-click tracking

If you’re spending money on traffic and trying to figure out which message actually produces the outcome you care about, this is for you.

Supported Outcome Signals

You choose the business outcome that matters most.

Primary outcome

  • leads
  • bookings
  • qualified leads
  • signups
  • trial starts
  • activated users
  • purchases
  • revenue

Optional quality signal

  • show rate
  • activation rate
  • purchase rate
  • close rate
  • AOV
  • LTV proxy
  • qualification score

The more downstream data you connect, the smarter the agent gets.

But you do not need a perfect CRM to get value from this.

The Four Outcome Truths

Every run classifies your funnel into four buckets:

1. Real Winner

Strong upstream response and strong downstream outcome quality.

2. Fake Winner

Looks good in-platform, weak on actual outcomes.

3. Leak

The interest is real, but something after the click is breaking.

4. Underfed Winner

Low volume, strong efficiency, deserves more budget or more testing.

This framework is the core of the kit.

How It Works

The Pipeline

INGEST → MAP → SCORE → DIAGNOSE → RECOMMEND → LEARN

1. Ingest

Pull data from:

  • Meta Ads
  • Google Ads (optional)
  • landing pages / site analytics
  • booking or signup tools
  • CRM or revenue systems if available

2. Map

Normalize everything into a shared structure:

  • angle
  • asset
  • page
  • audience
  • source
  • outcome event

3. Score

Evaluate each angle, creative family, and page against:

  • cost per outcome
  • conversion efficiency
  • quality signal
  • data confidence

4. Diagnose

Find:

  • real winners
  • fake winners
  • leaks
  • underfed opportunities

5. Recommend

Turn that into action:

  • scale
  • cut
  • rewrite
  • rebuild
  • clone
  • test

6. Learn

Store patterns over time:

  • which hooks drive low-intent traffic
  • which pages break conversion
  • which creative styles produce better outcomes
  • which angles are over-credited upstream

This is not a snapshot. It compounds.

Why This Works When Most “Attribution” Tools Don’t

Most tools show you numbers.

Outcome Kit makes decisions.

Most tools assume:

  • pristine UTMs
  • perfect CRM hygiene
  • fully connected revenue data

Outcome Kit assumes reality:

  • messy tracking
  • incomplete downstream data
  • disconnected tools
  • still needing to make better calls anyway

It doesn’t pretend to know more than it knows. Every recommendation is confidence-aware.

That means the agent can say:

  • High confidence: this angle is the strongest by booked-call efficiency
  • Medium confidence: this creative family is likely bringing low-intent traffic
  • Low confidence: revenue linkage is weak, recommendation withheld

That honesty is the point.

The 12 Skills

Skill What It Does
meta-source-reader Pulls Meta campaign, ad set, and ad performance data
page-signal-reader Reads landing page sessions, CVR, and engagement data
outcome-event-reader Pulls leads, bookings, signups, purchases, or other outcome events
metadata-loader Loads angle tags, creative tags, audience tags, and page mappings
angle-mapper Groups ads, pages, and outcomes into message families
outcome-scorer Builds blended scoring around your chosen outcome
fake-winner-detector Finds assets that look good upstream but fail downstream
leak-diagnoser Identifies whether the break is in the ad, page, or follow-through
opportunity-finder Finds low-volume, high-quality angles worth scaling
decision-writer Converts analysis into concrete operator moves
brief-sender Delivers daily or weekly briefs to Telegram, Slack, or WhatsApp
pattern-memory Stores recurring patterns and learns over time

Each skill can run standalone or as part of the full pipeline.

The Real Unit of Analysis

Most marketing reports are built around:

  • campaigns
  • ad sets
  • ads
  • channels

That’s not enough.

Outcome Kit is built around angles.

Because “Ad 27” is not the strategic question.

The real question is:

  • Is the time-savings angle working?
  • Is founder-led proof driving better bookings?
  • Is the competitive teardown message producing stronger buyers?
  • Is the page failing the message?

That’s the level that actually matters.

Example Use Cases

SaaS with demo bookings

  • Find which ad angle produces the best cost per booking
  • Spot high-CTR ads that bring weak show rates
  • Identify page-message mismatch on demo pages

Agency with lead forms

  • Compare cheap lead angles vs qualified lead angles
  • Find the creative family that drives the best booked calls
  • Detect landing pages suppressing form completion

Product-led SaaS with signups + activation

  • Find which campaigns bring signups that actually activate
  • Catch curiosity hooks that inflate top-of-funnel volume
  • Recommend more creatives around the best activation-producing message

E-commerce brand

  • Find which creative angles drive actual purchases, not just clicks
  • Compare page variants by conversion efficiency
  • Spot underfed winners worth scaling

Example Daily Brief

OUTCOME KIT — DAILY BRIEF

PRIMARY OUTCOME: Activated Signups
SECONDARY SIGNAL: Purchase Conversion
CONFIDENCE: HIGH

REAL WINNER
• "Stop guessing" angle
  - lowest cost per activated signup
  - strongest downstream purchase rate
  - stable across 14-day and 30-day windows

FAKE WINNER
• "Save hours" angle
  - best CTR
  - weakest activation rate
  - high curiosity, low buyer intent

LEAK
• Founder-led ads → generic product page
  - strong clickthrough
  - weak signup CVR
  - message-page mismatch

UNDERFED WINNER
• Competitive teardown angle
  - low spend
  - high activation efficiency
  - deserves more budget

NEXT 3 MOVES
1. Shift 20% spend from "Save hours" to teardown angle
2. Create a new page matching founder-led proof messaging
3. Generate 3 new creatives from the "Stop guessing" family

V1 Scope

This kit is intentionally focused.

V1 includes

  • Meta as the primary paid source
  • landing page / site signal ingestion
  • one outcome source
  • angle tagging
  • confidence-aware scoring
  • daily or weekly operator brief
  • persistent learning / memory

V1 does not include

  • full multi-touch attribution
  • perfect revenue reconciliation
  • automatic budget changes
  • enterprise dashboards
  • every ad platform on earth

The point of V1 is not “complete marketing measurement.” It’s better decisions, fast.

Running with OpenClaw or Claude Cowork

This repo now supports both:

  • OpenClaw for full agent runtime, memory, cron, and chat workflows
  • Claude Cowork for teams who prefer Claude's desktop agent with sub-agent coordination

The underlying pipeline is the same. The difference is the agent shell around it.

Shared setup

# Clone the repo
git clone https://github.com/TheMattBerman/outcome-kit.git
cd outcome-kit
cp .env.example .env
cp config.example.json config.json

# Add your keys, data sources, and primary outcome

Check readiness

npm run doctor

Run the sample pipeline

npm run run:sample

Run the live pipeline

npm run run

OpenClaw usage

OpenClaw is the best fit if you want:

  • Telegram / chat-based operation
  • cron jobs and recurring briefs
  • memory and longer-lived agent behavior
  • tool routing across a bigger agent system
npm install -g openclaw
openclaw start

Then message it naturally:

  • “Analyze which ad angles are driving booked calls”
  • “Find fake winners in my Meta funnel”
  • “What’s leaking between the ads and the page?”
  • “Which angles deserve more budget?”
  • “Give me the daily outcome brief”

Claude Cowork usage

Claude Cowork is a strong fit if you want:

  • desktop-native agent workflows with sub-agent coordination
  • agent instructions that live in CLAUDE.md and .claude/
  • explicit skill, rule, and hook structure via plugins

This repo includes:

  • CLAUDE.md
  • .claude/agents/
  • .claude/skills/
  • .claude/rules/
  • .claude/hooks/

That means Claude Cowork or Claude Code can understand:

  • pipeline order
  • what each stage does
  • which artifacts matter
  • what good diagnosis looks like
  • how confidence should constrain recommendations

Current agent guardrails

The repo now includes:

  • pipeline-order hook checks so later stages do not run on missing or invalid artifacts
  • confidence-aware diagnosis rules
  • per-angle attribution quality surfaced in briefs
  • validate-first suppression when evidence or attribution quality is too weak to justify action

If you use Claude Cowork, this gives the agents much better rails than a generic repo prompt.

Example Config

{
  "primary_outcome": "bookings",
  "secondary_signal": "show_rate",
  "sources": {
    "meta_ads": true,
    "google_ads": false,
    "page_signals": true,
    "crm": false,
    "booking_tool": true
  },
  "confidence_rules": {
    "min_sample_size": 3,
    "allow_medium_confidence_recommendations": true,
    "withhold_low_confidence_actions": true
  },
  "source_enrichment": {
    "rules": [
      {
        "match": "competitor breakdown",
        "angle": "competitive-teardown",
        "page": "/teardown"
      }
    ]
  },
  "angles": [
    {
      "name": "time-savings",
      "audience": "heads-of-growth",
      "creative_family": "founder-direct",
      "page": "/time",
      "match_tokens": ["save time", "time savings", "hours back"]
    }
  ]
}

File Structure

outcome-kit/
├── README.md
├── SETUP.md
├── SPEC.md
├── AGENTS.md
├── SOUL.md
├── .env.example
├── config.example.json
├── package.json
├── scripts/
│   ├── lib.js
│   ├── run.js
│   ├── meta-source-reader.js
│   ├── page-signal-reader.js
│   ├── outcome-event-reader.js
│   ├── map-angles.js
│   ├── score.js
│   ├── diagnose.js
│   ├── recommendations.js
│   ├── write-memory.js
│   ├── brief.js
│   └── validate-config.js
├── CLAUDE.md
├── .claude/
│   ├── agents/
│   │   ├── data-reader.md
│   │   ├── diagnostician.md
│   │   └── brief-writer.md
│   ├── skills/
│   │   ├── meta-source-reader/
│   │   ├── page-signal-reader/
│   │   ├── outcome-event-reader/
│   │   ├── angle-mapper/
│   │   ├── outcome-scorer/
│   │   ├── fake-winner-detector/
│   │   ├── leak-diagnoser/
│   │   ├── opportunity-finder/
│   │   ├── decision-writer/
│   │   ├── brief-sender/
│   │   ├── pattern-memory/
│   │   └── run-analysis/
│   ├── rules/
│   │   ├── data-quality.md
│   │   ├── scoring.md
│   │   └── diagnosis.md
│   └── hooks/
│       └── check-pipeline-order.sh
└── workspace/
    └── outcomes/

Anti-Patterns This Kit Defends Against

By default, AI and dashboards both tend to do dumb shit here.

Outcome Kit is built specifically to avoid:

  • declaring winners from CTR alone
  • treating cheap leads as high-quality outcomes
  • hiding broken tracking instead of surfacing it
  • overclaiming certainty when confidence is low
  • flattening different message angles into one blended average
  • reporting numbers without decisions
  • optimizing for platform metrics that don’t match business outcomes

If the data is messy, the agent says so. If confidence is low, it says so. If the platform winner is actually garbage, it says that too.

About

Built by Matt Berman.

This is for operators who need truth before they need another chart.

Star the repo if this helps. It tells me to keep building.

About

Outcome analysis system with enrichment, attribution quality, and confidence-aware recommendations

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages