People Intelligence

At its core, People Intelligence answers a question no traditional CRM can: "What do I need to know about this person right now, and what should I do about it?" Enterprise CRMs model people as functions of business transactions — they know what someone does at work but almost nothing about who they are as a person. Personal CRMs like Monica track birthdays and notes but can't extract behavioral insights from actual conversations. AI-native tools like Clay and Attio push enrichment further, pulling data from 150+ sources, but they optimize for "who should I sell to?" not "how should I relate to this human being across all dimensions of life?" People Intelligence bridges this gap. It is a personal relationship intelligence system grounded in relationship science — Big Five personality research, DISC communication theory, Dunbar's social circles, Social Penetration Theory, and Gottman's interaction research — that models people the way a deeply attentive friend would: knowing their communication style, remembering what matters to them, understanding their decision patterns, respecting their boundaries, and growing this knowledge organically from every interaction.

Every person is represented by a structured 10-section profile, each section justified by specific use cases and grounded in research. The first section, Identity & Contact, covers the foundation — full name, preferred name, primary channels (phone, email, WhatsApp), location, timezone, and languages. The second, Relationship Context, captures how you know each other, the relationship type (a person can be friend, colleague, and co-investor simultaneously, as Attio's relational model demonstrates), Dunbar layer classification (5 intimate / 15 sympathy / 50 close / 150 active), and relationship depth mapped to Social Penetration Theory stages from orientation through stable exchange.

The Communication Profile section is where DISC theory becomes actionable. It captures directness preference, formality level, detail appetite, preferred channel, and response timing expectations — the dimensions that directly drive message drafting. When the system knows someone is a D-style communicator (direct, results-oriented, decisive), it suggests concise, bottom-line-first messages. For an S-style (patient, reliable, team-oriented), warmer and more relational. Personality & Psychology goes deeper — approximate Big Five positioning (openness, conscientiousness, extraversion, agreeableness, neuroticism), attachment tendency (secure, anxious, avoidant), decision-making style, risk tolerance, conflict style, and motivational drivers. These dimensions predict behavior in negotiations, conflict, and relationship dynamics. Research shows GPT models can classify personality types from text with 73–76% accuracy — the system can infer these over time from message history.

Personal Life captures what makes you a good friend — birthday, family context (partner, children), interests, important dates, current life priorities. This is what Monica CRM got right: the details that show you remember and care. Professional Context covers current role, career trajectory, industry, expertise, and professional aspirations — the substance behind business relationships. Network & Connections maps who-knows-who: mutual contacts, organizational relationships, introduction pathways, and community memberships.

The final three sections are where People Intelligence diverges from any existing system. Behavioral Patterns is entirely auto-computed from interaction data — response times, frequency trends, conversation initiation balance, sentiment trajectory, message length reciprocity. No manual entry required. Interaction History is the unified conversation timeline — every exchange across every channel, chronologically ordered, with extracted topics, commitments, and conversation nuggets. And Working Notes & Observations is the structured catch-all: freeform notes, things to remember, sensitive topics to handle carefully, and open questions to explore next time you connect.

The system draws from every available source — and each source feeds different profile sections. Private conversations reveal how someone relates to you. Public presence reveals how they present to the world. Both matter, for different reasons.

Sources fall into three categories. Conversation data — WhatsApp messages, email threads, meeting transcripts, Telegram chats, and other direct messaging channels — captures how someone communicates with you: their style, formality, directness, response patterns, topics they raise, commitments they make, personal details they share naturally. This is the richest source for Communication Profile, Behavioral Patterns, Relationship Context, and Interaction History. Public profile data — LinkedIn, Instagram, Facebook, X/Twitter, TikTok, YouTube channels, personal websites, blogs, and professional bios — reveals how someone presents to the world: their interests, lifestyle, professional positioning, opinions, creative output, and the image they curate. This feeds Personal Life, Professional Context, Personality & Psychology, and Identity & Contact. Structured contact data — Google Contacts, phone contacts, email signatures, company directories — provides the factual foundation: phone numbers, email addresses, job titles, company affiliations, physical addresses.

Each source contributes to multiple profile sections, and no single source is comprehensive. WhatsApp messages reveal communication style but not career trajectory. LinkedIn provides professional context but not communication preferences. Instagram surfaces interests and lifestyle but not how someone handles conflict. A waterfall enrichment strategy queries all available sources — WhatsApp contacts + Google Contacts + email signatures + transcript mentions + LinkedIn + Instagram + Facebook + X + personal websites + public web data — and each adds facets the others miss.

Enrichment is continuous, not a one-time event. Three triggers drive it. On creation — when a new Person profile is created, the system discovers all channels where this person exists (find their Instagram handle, X account, LinkedIn, Facebook, personal website), pulls what's available from each, and builds the initial profile. On demand — at any point, you can say "enrich this profile from recent conversations" or "check their latest LinkedIn posts" to pull fresh data from specific sources. Passive and automatic — new WhatsApp messages, new transcripts, new emails gradually feed behavioral patterns and update the profile in the background without any manual trigger.

The profile deepens over time through a natural accumulation process. Early profiles are thin — perhaps just a name, phone number, and how you met. After a few WhatsApp exchanges, the system detects communication style preferences. After a transcript of a meeting together, professional context fills in. A LinkedIn check adds career trajectory and mutual connections. Instagram reveals interests and lifestyle. After months of interaction, behavioral patterns emerge — response time averages, conversation initiation balance, sentiment trends. The system actively surfaces opportunities to learn more, recognizing that knowledge deepens through Social Penetration Theory's progression from surface-level orientation through exploratory exchange to genuine depth. Inference transparency is critical throughout: the profile clearly marks what was directly observed ("said they prefer morning calls") versus what was inferred ("likely high conscientiousness based on communication patterns").

Communication patterns are the most honest signals of relationship dynamics. Research from PNAS (2022) shows that response time is "an honest signal of connection because it precedes conscious control." But individual data points are noise — a single late reply means nothing. What matters are trends. The Behavioral Pattern Extractor analyzes communication data to surface meaningful signals: response time trends over months (improving, stable, or declining), conversation initiation balance (who starts conversations more often, indicating investment level), message length reciprocity (roughly balanced lengths over time indicate healthy engagement), and sentiment trajectory (is the emotional tone trending positive, negative, or neutral?).

The system adapts Gottman's interaction ratio — the finding that stable, healthy relationships maintain a 5:1 ratio of positive to negative interactions — to digital communication. Declining sentiment ratios trigger proactive alerts. The Communication Style Analyzer uses LLM analysis of message history to infer DISC-approximate preferences: directness, formality, detail level, and pace. These aren't one-time classifications but evolving assessments that update as more data accumulates. The focus is always on what's meaningful versus noise: response time changes are more revealing than absolute times, sentiment trends matter more than individual message sentiment, and topic breadth over time indicates relationship depth progression.

The Relationship Health Monitor evolves the existing Pulse system from simple frequency tracking to multi-dimensional health monitoring. It tracks four key dimensions. Frequency stability — consistent interaction frequency over time, where sudden drops are red flags. Reciprocity balance — roughly equal initiation, response effort, and emotional investment from both sides. Sentiment trajectory — whether the emotional tone of interactions is trending positive, negative, or neutral over time. And response reliability — consistent response times, where erratic patterns suggest ambivalence.

Each person is automatically classified into Dunbar layers based on interaction frequency and depth — the 5 closest relationships, the 15 sympathy group, the 50 close network, and the 150 active network. Different layers warrant different levels of profile detail and different outreach cadences. When relationship health indicators decline — a three-week gap with someone you normally talk to every week, a steady drop in sentiment ratio, a shift from mutual initiation to one-sided contact — the system surfaces proactive alerts. It suggests positive-gesture outreach when the Gottman ratio drops, flags contacts drifting out of their usual Dunbar layer, and reminds you of upcoming birthdays, anniversaries, and commitments. The goal is never to let a valued relationship decay through simple neglect.

Building detailed profiles about people raises genuine ethical questions. People Intelligence is governed by four principles. The Friend Principle: a good friend remembers things about you because they care, not because they're surveilling you. The system should model what an attentive, caring friend would naturally remember — not what a surveillance state would collect. If it would feel uncomfortable to tell a contact "my AI keeps detailed notes on your personality type," the data point probably shouldn't exist in that form. The Reciprocity Principle: profile only what the person has willingly shared, plus publicly available information. Don't infer sensitive information — political views, health conditions, sexual orientation — from indirect signals. The person should, in principle, be comfortable with everything in their profile. The Purpose Limitation Principle: every data point must serve a relationship-enhancing purpose. No data collection for its own sake. Every field answers "how does knowing this help me be a better friend, colleague, or partner to this person?" The Dignity Principle: never reduce a person to a manipulation target. There's a line between knowing someone prefers direct communication so you'll be straightforward (good) and knowing someone has anxious attachment so you'll exploit that in negotiations (bad). The profiles help you understand people, never manipulate them.

Legally, the system operates under the GDPR household exemption (Article 2(2)(c)) — personal data processing by a natural person for purely personal or household activity — and the Singapore PDPA personal/domestic exemption. Both require the data to stay personal and never be shared with third parties. The system is designed for the strictest standard: purpose limitation, data minimization, inference transparency, temporal decay (old data loses accuracy and is weighted accordingly), and secure encrypted storage. Profile data never leaves the system — it informs your behavior but is never sent to anyone.