ChangeSignal — Product Design Spec

Author: Daniel Jonas, Campana & Schott Date: April 8, 2026 Status: Ready for Build Day Build Day Team: TBD (1 builder + 4 non-builder roles)

1. Problem Statement

Change management teams in pharma transformations have a data problem — but not the kind they think. The data exists. It's just trapped in conversations, meeting debriefs, and hallway check-ins that never make it into a system anyone can act on.

1. The reporting tax. Every two weeks, someone spends 8–12 hours pulling adoption data from surveys, attendance logs, and their own memory to build a SteerCo report. At consulting rates, that's $4,000–$6,000 per cycle — north of $100K per year — on copy-paste work.

2. People fall through the cracks. A 30-stakeholder transformation means 30 relationships to maintain. The loud ones get attention. The quiet ones get forgotten. The workstream lead who stopped showing up three weeks ago? Nobody noticed until the SteerCo asked why that workstream stalled.

3. No early warning system. By the time resistance shows up in a status report, it's already metastasized. Change teams react to problems after they surface because there's no system watching for the signals in between.

4. Zero behavior change required — but no tool captures it. Change teams already debrief after every meeting. They already post updates in Teams. They already know who's bought in and who isn't. The problem isn't data collection — it's that none of this gets captured, aggregated, or acted on systematically.

The structural problem: the profession that exists to manage human transitions through change has the worst tooling for managing the humans. Sales has Salesforce. Customer success has Gainsight. Change management has spreadsheets and memory.

2. Product Concept

One-sentence pitch

ChangeSignal turns your team's natural post-meeting debriefs into two things: an executive adoption report for leadership, and a daily action board for your change managers — from a single Teams channel.

How it works

The change team posts in a dedicated Teams channel after meetings, the way they already do:

"Kickoff with Field Force SE went well. Kevin Walsh seemed checked out, left early. Rachel asked good questions about workflow integration. Soo was skeptical about timeline but didn't push back hard. Rest of the group seemed bought in."

ChangeSignal's AI already has the stakeholder map (loaded once at setup). It parses that message and extracts: Kevin Walsh → negative, Rachel Martinez → positive, Soo Son → neutral, 8 others → positive. Each person gets an interaction record logged automatically.

That same data feeds two views:

• Report view (for leadership): Aggregates all channel posts over a sprint cycle. Generates a written executive adoption report with heatmaps by stakeholder group, resistance flags, and recommended interventions. Biweekly. Sent to the SteerCo.
• Action board (for the change team): Tracks individual stakeholder relationships in real time. Flags people going cold. Generates personalized re-engagement messages. Used by the change manager every morning.

One channel. Two views. Same data, different altitude.

The strategic play

ChangeSignal is deployed on every C&S change engagement from day one. The client's team uses the channel naturally. When C&S leaves, the client has two choices: lose the accumulated stakeholder intelligence and go back to spreadsheets — or license ChangeSignal and keep the institutional memory. Every engagement becomes a land-and-expand motion.

3. Target Buyer

• Primary buyer: The VP or SVP who owns a transformation program and is accountable for adoption outcomes. They're spending $5–50M on a transformation and their only visibility is a biweekly deck that arrives two days late.
• What they do today: Ask the change team to compile adoption reports manually (8–12 hours per cycle, ~$5K per report). Track stakeholder engagement in spreadsheets nobody trusts by month three.
• Why they would pay: Real-time adoption visibility replaces biweekly guesswork. Resistance gets flagged weeks earlier. Reporting labor drops 80%. Nudge generation saves 3–5 hours per week on outreach. Total labor replaced: ~$8K/month.
• Secondary beneficiary: The change management team. They stop building spreadsheets and start having conversations.

4. Architecture

Input Layer

One primary input: a Teams channel (or Slack). Change managers post natural language debriefs after meetings, touchpoints, and informal conversations. The AI parses names, sentiment, and topics against the pre-loaded stakeholder map. One channel post = multiple interaction records logged automatically.

Additional inputs that layer in over time (not required at launch): Copilot meeting summaries forwarded to the channel, email threads forwarded in, and eventually direct Graph API integration for passive capture.

The MVP input is "post in the channel like you already do." Zero behavior change.

Transform Layer (Core IP)

A Claude-powered pipeline with a carefully engineered system prompt (the Skill) that performs three functions:

1. Natural language parsing: Extracts participant names from free-text posts, maps them against the stakeholder registry, classifies sentiment per person (positive/neutral/negative), and pulls out key topics and concerns. This is the hardest AI problem and the highest-leverage prompt engineering work.

2. Engagement health scoring: Computes a health score per stakeholder based on recency of last interaction, frequency trend, sentiment trend, and influence level. A high-influence stakeholder trending cold gets flagged faster. Rule-based for transparency, AI-interpreted for nuance.

3. Dual output generation:

   • Report view: Aggregates interaction data at the stakeholder group level over a sprint cycle. Generates a written executive summary with adoption heatmaps, resistance flags with severity, and recommended interventions. Low temperature for consistency. Output reads like a senior analyst wrote it.
   • Action board: Scores individual stakeholder health, sorts into engaged/watch/at-risk categories, and generates personalized nudge messages referencing the stakeholder's role, last interaction, known concerns, and workstream status.

Display Layer

Two views:

• Action board: A kanban-style board showing all stakeholders in Engaged (green), Watch (amber), At Risk (red) columns. Each card shows name, role, last touchpoint, sentiment. Click to see engagement timeline and generate a nudge message.
• Report view: A clean, readable adoption report — executive summary, adoption heatmap by stakeholder group, resistance flags, recommended interventions, trend vs. prior sprint. Review, edit, export.

Tech Stack (Hackathon)

• Frontend: HTML/Tailwind or React — the board view is the hero UI
• Backend: Lightweight Node or Python API — handles channel post parsing, health scoring, Claude API calls for nudge and report generation
• AI Layer: Claude API with system prompt defining the stakeholder parsing, scoring, and output generation framework
• Storage: JSON files — stakeholder map + interaction log
• Sample Data: Pre-built map of 15 stakeholders across 3 workstreams, with 8–10 sample channel posts representing 2 months of history
• Deployment: Vercel or Replit — shareable URL for judges

5. Hackathon Execution Plan

How the live demo works

During the pitch, the Narrator opens the input interface, types a real debrief from a meeting that happened during the hackathon, and the audience watches ChangeSignal parse it live — names light up on the board, health scores shift, a nudge generates in real time. No slides. The product IS the presentation.

Timeline

Critical risk and fallback

Biggest risk: The natural language parsing is inconsistent — misidentifies names or misreads sentiment. The live demo becomes a live failure.

Fallback: If parsing isn't reliable by the halfway mark, lock input to the pre-loaded sample data. Focus all effort on making the two views beautiful. A polished board + report from pre-processed data beats a janky live pipeline. The pitch shifts from "watch it parse live" to "here's what the output looks like, and here's the input interface that feeds it."

Role allocation

6. Defensibility & Competitive Moat

Why a client would pay — and keep paying

• The engagement history is the moat. After 6 months, ChangeSignal contains every stakeholder relationship — who championed, who resisted, what worked. That history carries across transformations. Walking away means losing institutional memory.
• The AI improves with use. More interactions = more context = better nudges. Early nudges are good. Month-6 nudges are significantly better.
• Cross-program intelligence. A VP in three programs has one unified profile. The change team on Program B sees they were a late adopter on Program A.
• Zero behavior change to adopt. The input is a Teams post. Your team already does this. Adoption barrier is effectively zero.

Pricing model

Competitive positioning

Closest analogue: "Gainsight for change management — but fed by a Teams channel instead of product telemetry." Nothing like this exists.

7. Success Criteria

Hackathon demo (April 17)

• Text input accepts a natural language debrief and parses names + sentiment correctly
• Action board renders stakeholders in red / amber / green based on health scoring
• Clicking a stakeholder generates a personalized, context-aware nudge message
• Report view generates an executive adoption summary with heatmap and resistance flags
• Demo uses realistic pharma transformation data with identifiable patterns
• At least one nudge is good enough that a judge says "I'd actually send that"

"Would a client pay for this?" test

• The action board answers "who needs my attention today?" in one glance
• The report replaces a $5K manual reporting cycle with a 30-second generation
• The nudges feel human — specific enough to send, not generic enough to ignore
• The pricing is anchored to concrete labor the client already pays for
• The "no CM tool exists for this" gap lands clearly

8. Open Questions

1. How good can the NLP parsing get in 4 hours? The product depends on reliably extracting names and sentiment from casual Teams posts. 90%+ accuracy = demo hit. 70% = risky live demo.

2. Should the system handle ambiguity explicitly? "Soo was skeptical but didn't push back" — neutral or negative? Showing confidence levels adds trust but also complexity.

3. Can we use other hackathon teams' actual updates as input? If all teams post in a shared channel during Build Day, ChangeSignal demos with real data from the event itself.

4. What's the minimum viable stakeholder map? 15 people across 3 workstreams for the demo. Can setup take under 10 minutes?

5. Post-hackathon: how fast to Copilot integration? Passive capture of meeting notes is the unlock from "easier than a spreadsheet" to "it just works without me doing anything."