The Hypothesis

Can a standardised MCP-style protocol make customer cohort data queryable by any AI agent, tool, or business user?

The Concept

Just as Model Context Protocol provides a standardised way for AI models to access tools and data sources, this experiment builds a protocol layer specifically for customer cohort data. Any AI agent, automation, or business user can connect and make structured requests: "Get persona definition for segment X," "What are the top pain points for enterprise evaluators?," "Which cohort responds best to urgency-driven CTAs?" The protocol turns customer intelligence from tribal knowledge into machine-queryable infrastructure.

MCP protocol for customer cohorts

The hypothesis

Can a standardised MCP-style protocol make customer cohort data queryable by any AI agent, tool, or business user?

The concept

Just as Model Context Protocol provides a standardised way for AI models to access tools and data sources, this experiment builds a protocol layer specifically for customer cohort data. Any AI agent, automation, or business user can connect and make structured requests: “Get persona definition for segment X,” “What are the top pain points for enterprise evaluators?,” “Which cohort responds best to urgency-driven CTAs?” The protocol turns customer intelligence from tribal knowledge into machine-queryable infrastructure.

How it works

1. Define persona schema — structured format: demographics, motivations, pain points, objections, tone, triggers
2. Populate from data sources — CRM, analytics, behavioural data → mapped to persona segments
3. Expose via MCP-style protocol — standardised query interface for agents and tools
4. Enable natural language access — business users ask questions conversationally
5. Feed back outcomes — campaign results refine cohort definitions automatically

A single protocol that makes customer intelligence accessible to machines and humans alike.

What it explores

• Can a standardised protocol for cohort data reduce integration complexity for downstream tools?
• Do AI agents that query cohort data via protocol produce better output than agents given static persona docs?
• Is natural language access to cohort data useful for non-technical business users?
• How does the protocol handle conflicting data from multiple sources?
• Can the protocol scale to support real-time queries from multiple agents simultaneously?

What we found

• AI agents querying the protocol produced 34% more targeted content than agents given static persona documents
• Natural language access drove 4x more persona queries from marketing managers than the previous dashboard
• Integration time for new downstream tools dropped from days to hours
• The protocol handled concurrent queries from up to 12 agents without performance degradation
• Pain points and objections were the most-queried fields — demographics were rarely used

Learnings

• An opinionated schema outperforms a flexible one — agents need structure to know how to use the data, not freedom to interpret it.

• Natural language access serves different users differently — ad-hoc strategic questions from humans, structured queries from agents.

• Outcome data feeding back into cohort definitions creates a virtuous cycle but needs a human review gate to prevent drift.

• Design persona schemas around the fields agents actually query — weight pain points and objections over demographics.

Where it goes next

This protocol is becoming a core layer in DataSpec. The question now: can the same protocol pattern work for other types of institutional knowledge — not just personas, but product context, competitive intelligence, and brand guidelines?