Proposed Engagement

An AI ecosystem
for Win Without
Pitching

We've spent three years building an AI-powered operating system for our own agency — 35+ client accounts, 24 custom skills, six data connectors, autonomous agents monitoring around the clock. We rebuilt how we write, report, analyze, and execute.

This proposal outlines what we'd build for WWP. Not a single tool — a layered system where each layer makes the next one more powerful. You start with a configured Claude environment and end with an intelligent platform that knows your business, connects to your data, and works while you sleep.

How it's structured

Four layers, each building on the last

Each layer is independently valuable — you get real capability at every step. But each one also makes the next layer more powerful. Foundations give you a platform. The Creator Package gives you production tools. Pipedrive Intelligence gives you live data. The Intelligence Portal brings it all together.

1
Foundations
Claude Teams environment, custom INSTRUCTIONS, baseline skills,
anti-AI writing, team training, ongoing maintenance
Learn more
Recommended
2
Creator Package
Presentation decks, dashboards & interactive tools,
report generator, deploy to web — branded production
Learn more
Recommended
3
Pipedrive Intelligence
Custom MCP connector, Claude.ai skill, autonomous
weekly briefing agent with persistent memory
Learn more
Recommended
4
Customer Intelligence Portal
Unified contact graph, lead scoring, web portal,
automated alerts across 5 data sources
Learn more
Proposed project
Layer 1

Foundations

The groundwork. We configure Claude as a unified platform for your team, teach it how your business works, and build the habits that make everything else possible.

Claude environment setup

We configure a Claude Teams environment for WWP with dedicated project spaces, organized by function. Each project gets a custom INSTRUCTIONS file that grounds Claude in the right context — who the client is, how to behave, what tools are available.

This is the same pattern we use internally across 35+ client accounts. It's the difference between "using ChatGPT" and having an AI that knows your business.

What gets configured

Claude Teams workspace for WWP
Project folders organized by function
Custom INSTRUCTIONS per project (voice, context, rules)
MCP connections for tools and data
Ongoing updates as Claude evolves (new features, new models)

Custom baseline skills

We write skills that encode WWP's specific knowledge — not generic AI instructions, but your actual business context, terminology, workflows, and decision-making frameworks.

These aren't documents Claude reads once. They're active skill files that change how Claude works in every conversation. When someone asks Claude to help with positioning, it uses Blair's frameworks, not generic advice.

To build these effectively, we need your team to provide current documentation — SOPs, process descriptions, frameworks, terminology guides, sales playbooks. The more clearly your workflows are documented, the more precisely we can encode them into skills. We'll work with you to identify what matters most, but we need the raw material to come from you.

What we'd build for WWP

Business context skill — who WWP is, the products (Masterclass, Pricing Workshop, Positioning), the regions, the RTD structure, the sales process
Workflow skills — SOPs and processes we identify that Claude should follow when helping with specific tasks
Anti-AI writing skill — our production-grade content filter that catches and removes AI-sounding patterns. 50+ banned phrases, banned structures, mandatory post-draft audit.
Ongoing refinement — skills improve as we learn what works. New skills get added as we identify opportunities.

Training

We train your team on how to use Claude effectively — not generic prompt tips, but specific workflows for your actual work. How to query data. How to produce content that doesn't sound like AI. How to use Cowork mode with your existing tools.

Ongoing maintenance

Claude evolves constantly — new models, new features, new capabilities. We keep your environment updated, your skills tuned, and your team current. This isn't a one-time setup.

The platform effect

Once Foundations is in place, Claude.ai becomes the single interface where your team goes to get things done. Not a tool you switch to — the place you already are when you need to write, analyze, query, build, or decide.

The difference between "using AI" and having AI that knows your business is Foundations.

Layer 2

Creator Package

We teach Claude the WWP brand, opening the door to producing polished deliverables — presentations, dashboards, reports, interactive tools — as living web elements that can be deployed, shared, and updated.

Presentation decks

Interactive HTML decks built from a conversation. Navigation, build animations, responsive. Give Claude a document or walk through it — the output is a polished, browser-based presentation deployed to a live URL.

Dashboards & tools

KPI dashboards, data visualizations, calculators, intake forms, interactive reports. Built with a design system that enforces consistent output. Deployed to a live URL instantly.

Report generator

Structured report production that pulls from connected tools. Reports persist — build new, load saved, update incrementally. Three modes, one workflow, branded output.

Deploy to web

Say "deploy this" and Claude pushes any artifact to a live URL. Password-protected if needed. SSL automatic. Share a link instead of a file. Update later without re-sending.

These combine in a single conversation

"Write a summary of our Q3 pipeline, put it in a presentation deck, and deploy it so I can share the link with Blair."

Four skills — data query, writing, presentation, deploy — chained in one session. The output is a live URL Blair can open in a browser.

"Take last week's report, update the pipeline numbers, add a section about the new LATAM leads, and re-deploy."

Load, edit, update, re-deploy. The report is a living document at a stable URL, not a file attachment that goes stale.

Layer 3

Pipedrive Intelligence

A custom MCP connector that gives Claude direct, live access to your Pipedrive CRM — plus a skill that teaches Claude how WWP specifically uses Pipedrive. This isn't a generic CRM integration. It's trained on your pipeline, your stages, your data.

Live CRM data in Claude

Claude queries your Pipedrive data in plain English. Contacts, deals, activities, pipeline stages — all live, all current. Seven dedicated tools for common queries (fast, no AI cost) plus a code-generation engine for open-ended questions that cross-reference data in ways the standard tools can't.

Built on the same architecture as our Harvest connector — battle-tested across thousands of queries.

Trained on how WWP uses Pipedrive

The connector comes with a custom skill that teaches Claude your deal stages, your products, how to disambiguate contacts and organizations, what "stale" means in your pipeline, how to present results.

A generic Pipedrive integration doesn't know that when Blair says "how's the pipeline?" he means open deals grouped by product, or that the LATAM RTD needs different context than the NA RTD.

B
Which deals haven't had any activity in the last 30 days?
C
I found 3 open deals with no logged calls, meetings, or emails in the last 30 days:

[Each deal shown with title, value, current stage, last activity date, last activity type, and a note about what happened]

The query cross-references your deals with the activities table and filters for open deals where the most recent activity is older than 30 days. It pulls the last activity detail for each so you can see exactly where things left off.

Next steps: Want to see the full activity history for any of these? Or compare to deals that are actively moving?
What you can ask: Any question about your contacts, deals, activities, or pipeline. "Who has the most open deals?" "What's our win rate this year?" "Show me everything for [person]." "Compare this quarter to last." The connector queries your live Pipedrive data — the answers reflect your actual CRM, not a snapshot or export.
Also included in Layer 3

Pipedrive Agent

Alongside the connector that you use in Claude.ai, we deploy the Pipedrive Agent — an autonomous AI system that runs on a schedule, connects to your live CRM data, and gets smarter every time it runs.

What it does

Every Monday morning, the agent reviews your Pipedrive pipeline on its own — no prompt, no human trigger. It identifies deal status by stage, flags stale deals, tracks what moved, and writes a briefing with specific follow-up recommendations. Delivered via email or Slack before your week starts.

This uses the same Pipedrive connector Claude.ai uses, the same live data. The agent just runs it autonomously.

How it improves over time

Managed Agents aren't stateless. They have three mechanisms that make them genuinely better with each run:

Persistent memory
The agent retains what it learns across sessions. If you correct it ("that deal closed last week, stop flagging it") or it discovers something useful ("this contact always goes quiet in Q3"), that knowledge persists to the next run. It doesn't start from zero each Monday.
Self-grading (Outcomes)
The agent evaluates its own output against quality criteria before delivering it. Did it actually answer the right questions? Did it miss any stale deals? Are the recommendations specific enough? If not, it takes another pass — automatically, no human review needed.
Dreaming
Between sessions, a background process reviews the agent's accumulated memory — merging duplicates, resolving contradictions, surfacing cross-session patterns it couldn't see from any single run. The agent's understanding of your pipeline compounds over weeks and months.
The result: The first briefing is useful. The tenth briefing is significantly better — it knows your pipeline's patterns, your team's follow-up habits, which deals tend to stall and why. This isn't a report template. It's an analyst that learns your business.

Layers 1–3 give your team a connected, intelligent workspace. Layer 4 builds on that to create something new.

Layer 4 — Proposed Project

Customer Intelligence Portal

The most ambitious layer. A dedicated application that goes beyond Pipedrive — aggregating contact data from five sources (Pipedrive, Kit, Circle, Fathom, Clay) into a unified contact graph with lead scoring, a web portal, and automated alerts.

This is a separate project that builds on everything in Layers 1–3. The same Pipedrive connector, the same agent infrastructure, the same Claude.ai skills. It adds a scoring engine, additional data connectors, a dedicated web portal, and intelligent automation that surfaces your highest-value leads to the right person with a recommended next step.

Unified profiles

One view of every contact across five sources. Email engagement, Academy activity, call summaries, deal status, LinkedIn data — merged into a single profile matched by email.

Lead scoring

A configurable scoring engine that weighs engagement signals, applies recency decay, and ranks every contact. Tuned by your team based on what actually predicts conversions.

Automated alerts

When a lead crosses the engagement threshold, an AI agent reviews their full profile and sends the right RTD a contextual recommendation — not a template.

The intelligence portal has its own detailed product specification.

View the full spec
How it's operated

You own Claude. We run the rest.

WWP owns the Claude environment directly — your subscription, your users, your spend limits. We advise on the configuration as part of Foundations (workspace structure, member roles, budget controls), but the account stays in your name and under your control.

Everything else — the connectors, the agents, the deployed tools, the infrastructure that makes all of this work — we take care of.

Isolated
Dedicated network and compute. Your data never mixes with another client's.
Encrypted
All API keys in Secrets Manager. All connections over HTTPS. No anonymous access.
Monitored
Automatic alerts on failures. Cost tracking per client. Health dashboard.
Maintained
We handle updates, deployments, and scaling. You don't touch infrastructure.
Prepared by Newfangled · May 2026