Product Overview

How Theazo turns conversations into understanding

A deep dive into every capability, from real-time persona building to predictive intelligence. Written for people who decide what goes into the stack.

Overview

Every conversation a user has with your product contains more signal than a thousand page views. Theazo is a Voice and Chat SDK that captures this signal, structures it, and builds AI personas of your users in real time.

Companies embed Theazo to get two things at once: a concierge that remembers every user and personalizes their experience, and continuous qualitative research extracted automatically from every conversation. No surveys. No manual tagging. No data science team required.

The result: users stay longer because the app actually knows them. Companies ship better products because they finally hear what users are saying, wanting, and avoiding.

Chat SDK

Text-based conversations with streaming AI responses

Voice SDK

Real-time speech with Deepgram STT and ElevenLabs TTS

Persona Engine

AI extracts and structures preferences from every signal

The Persona Engine

Traditional analytics tells you what users did. Theazo tells you who they are. The persona engine combines three layers of signal into a living profile that updates with every interaction.

SSAID — Explicit Preferences

What users directly tell you in conversation. These are the highest-confidence signals because the user chose to express them.

"I prefer minimalist, dark-colored things"
Style preference
"Nothing over $200"
Budget constraint
"It's for my wife's birthday next month"
Life event + intent
"I'm training for a marathon"
Activity context

DDID — Behavioral Signals

Actions speak louder than words. Theazo tracks hover duration, click patterns, scroll behavior, and time-in-viewport to extract implicit preferences users never articulate.

Hover duration

12s on Leather Jacket = strong interest

Viewport time

Tracks how long each item stays visible

Scroll speed

Fast scroll-past = disinterest signal

Click patterns

Which categories get repeated clicks

IINFERRED — AI Synthesis

Claude combines explicit and implicit signals to infer an archetype, predict behavior, and recommend how to engage the user. This layer gets smarter with every interaction.

Example inference

Archetype: The Understated Creative

Prefers quality over quantity. Drawn to minimalist aesthetics, natural materials, and independent brands. Price-aware but willing to pay more for craftsmanship. Avoids anything flashy or mass-produced.

Approach: Lead with material quality and design intent. Avoid hard sells. Reference their past browsing to show memory.

Negative Preference Modeling

What users avoid is 3x more reliable than what they say they like. People are inconsistent about preferences but remarkably consistent about aversions. Theazo captures and classifies negative signals at four severity levels.

Mild disinterest
-0.3
Mismatch
-0.5
Rejected
-0.7
Permanent aversion
-1.0

Behavioral aversions

Fast scroll past an item (<0.5s in viewport), hover then quick abandon, clicking away from a category repeatedly. All captured without the user saying a word.

Verbal aversions

"Nothing flashy," "I hate bright colors," "Not that brand." Explicit negative statements extracted from conversation and scored at -0.7 or higher.

Why this matters: Recommending something a user has already rejected destroys trust instantly. A single aversion violation undoes ten correct recommendations. Theazo ensures your concierge never shows a user something they have already told you they do not want.

Predictive Personas

Once Theazo has enough data points (typically after 2-3 minutes of interaction), the persona engine generates predictions that feed directly into business decisions.

Purchase likelihood

78% likely to convert

Based on engagement depth, expressed intent, and archetype match with historical buyers.

Lifetime value estimate

$340-$520 / year

Projected from price sensitivity signals, category affinity breadth, and return visit probability.

Churn risk

Low (12%)

Derived from frustration signals, unresolved complaints, and engagement decay patterns.

Price trigger

$150 threshold

The price point where this user hesitates. Detected from stated budget and browsing patterns around price ranges.

Memory Moments

Your concierge remembers users across sessions. Not just what they bought, but what they talked about, what life events they mentioned, and what they were looking for. This creates the feeling of a relationship, not a transaction.

Visit 1Today

"I'm training for a marathon next month."

The concierge notes: running, endurance athlete, event deadline.

Visit 22 weeks later

User returns to browse shoes.

"How's the marathon training going? I remember you were preparing for that. We just got new stability shoes that a lot of runners have been gravitating toward."

Visit 36 weeks later

User returns again.

"Did you end up running it? If you're in recovery mode, we have compression gear on sale this week. Your knees will thank you."

Memory persists across sessions, devices, and channels. A conversation started on voice continues seamlessly on chat.

Cohort Intelligence

Theazo automatically discovers user segments from behavioral and preference data. No manual tagging, no predefined categories. Cohorts emerge organically from how users actually behave.

Each cohort powers cross-pollination: "Users like you also loved..." recommendations within the same company's data, without any data ever leaving the company's boundary.

Urban Creatives

142 users (17%)

Minimalist aesthetics
Dark color palette
Independent brands

Budget Hunters

89 users (11%)

Price-first decisions
Sale alerts
Bulk buying

Gift Givers

67 users (8%)

Event-driven purchases
Personalization requests
Multiple recipients

Actionable intelligence

Each cohort comes with cross-sell opportunities, retention levers, and growth signals. Example: Urban Creatives have 2.4x higher AOV when shown curated collections vs. category browsing. That insight is only possible because Theazo understands preferences, not just behavior.

Ask the Audience

Query your entire customer base in natural language. No SQL. No dashboards. Just ask a question and get structured findings with evidence.

What do churned customers have in common?

Finding 1

Price sensitivity is the strongest churn predictor

73% of churned users expressed budget constraints in their first conversation. Users who mention price in their first 3 messages churn at 4.2x the baseline rate.

Finding 2

Unresolved frustrations within 48 hours

Users who reported a frustration and didn't receive a follow-up within 48 hours churned at 2.8x the normal rate. The most common unaddressed frustration was checkout speed.

Recommendation

Implement proactive price-match messaging for users who mention budget in their first session. Set up automated frustration follow-ups within 24 hours.

Ask the Audience works over your full persona dataset. Every response includes sourced evidence and actionable next steps, not generic summaries.

Ghost Mode

Session replay with intelligence. Ghost Mode records mouse movements, scroll behavior, and hover patterns, then overlays the behavioral signals Theazo captured at each moment.

Session replay

Leather Jacket
Vinyl Record
Bright Hoodie
Art Print
Black Sneakers
Coffee Maker
Neon Shirt
Wireless Buds
High engagement
Moderate
Avoidance

Every hotspot is annotated: which signals fired, what confidence changed, and what the persona engine concluded. Ghost Mode turns raw session recordings into structured behavioral insight.

The Dashboard

Everything Theazo captures flows into a dashboard built for both product teams and customer success managers.

Conversations

Every conversation transcribed, searchable, and tagged with extracted insights. Filter by sentiment, channel, or topic. Click to expand any transcript with color-coded preference highlighting.

Personas

Card view of every user persona with archetype, confidence score, top preferences, aversions, and predicted metrics. Drill into the full SAID/DID/INFERRED breakdown.

Insights

Aggregated intelligence across your user base. Theme trends, sentiment over time, frustration alerts with spike detection, and AI-discovered correlations.

Live Feed

Real-time event stream of every signal being captured. Preferences, frustrations, aversions, and feature requests, as they happen across all active sessions.

Security & Privacy

Theazo is built on a hard privacy principle: your customers walked into your store, not ours. Their data belongs to you. It never leaves your boundary.

Per-app data siloing

Every company gets a completely isolated data environment. Company A's persona data never touches Company B's data. There is no shared model, no shared embedding space, no cross-company training.

No cross-company data sharing. Ever.

Facebook should not know your DoorDash history. Spotify's persona stays with Spotify. This is not a deferred feature or a future roadmap item. It is a foundational architecture decision.

GDPR-ready

Right to deletion, data export, consent management. All persona data linked to anonymized user IDs. PII handling follows strict extraction rules with automatic redaction.

SOC 2 Type II roadmap

Infrastructure runs on Supabase (PostgreSQL) with row-level security. Encryption at rest and in transit. SOC 2 certification in progress for enterprise customers.

"Your customers walked into YOUR store, not ours. Their conversations, their preferences, their personas — they belong to you. We are infrastructure, not a data broker."

Integration

One SDK. One install. Chat and Voice in the same package, features gated by your plan, not by which package you import.

App.tsx

import { Theazo } from '@theazo/sdk'

// That's it. Drop it in your app.

<Theazo appId="your-app-id" />

React

Supported

Vue

Supported

Angular

Supported

Flutter

Supported

Starter plan includes Chat. Grow and above includes Voice. Enterprise adds custom model tuning, dedicated infrastructure, and premium SLAs. No package switching required when you upgrade.