How We Grew Mystica's Revenue 60% With AI Personalization

TL;DR

• Mystica (formerly “Relationship Psychics”) is a spiritual wellness app that needed AI personalization to retain users and grow revenue
• We delivered the engagement in 6 weeks, integrating Clarity’s self-model API to give Mystica’s AI a persistent understanding of each user
• Revenue increased 60% within 90 days of launch
• The core insight: users were not leaving because the AI gave bad answers — they were leaving because the AI did not remember who they were

---

This is the story of how Clarity helped Mystica, a spiritual wellness app, turn a retention problem into a 60% revenue increase — not by changing the AI model, but by changing what the AI knew about its users.

0%

revenue increase within 90 days

0 weeks

from kickoff to production launch

0 days

timeline to measured revenue impact

The Client: Mystica

Mystica is a spiritual wellness app — formerly known as “Relationship Psychics” — that connects users with AI-powered guidance for personal growth, relationships, and spiritual exploration. Their users come back repeatedly, often daily, seeking continuity in their spiritual practice.

The app had a working AI product. Users could have meaningful conversations. The model was competent. The UX was polished. On the surface, everything looked fine.

But under the surface, there was a problem that no amount of prompt tuning could fix.

The Challenge: Conversations Without Memory

Mystica’s core problem was one that plagues most AI products: the AI treated every conversation as if it were the first.

A user would spend 30 minutes in a deep conversation about a relationship challenge. They would share context, explore their feelings, receive guidance. The experience felt personal and meaningful. Then they would come back the next day, and the AI would greet them like a stranger.

The AI had no persistent understanding of who it was talking to. No memory of past conversations. No model of the user’s ongoing journey. Every session started from zero.

For a spiritual wellness app, this was devastating. The entire value proposition depended on continuity — the sense that the AI was a guide who knew your story, not a stranger who asked the same questions every time.

Before: Stateless Conversations

• ×Every session started from scratch
• ×Users had to re-explain their situation
• ×No continuity in guidance or recommendations
• ×AI could not reference past conversations
• ×Experience felt transactional, not relational

After: Self-Model Integration

• ✓AI remembered user context across sessions
• ✓Guidance built on previous conversations
• ✓Recommendations reflected the user's journey
• ✓AI referenced past insights and progress
• ✓Experience felt like an ongoing relationship

Users were not churning because the AI was bad at answering questions. They were churning because the AI did not know who they were. There is a meaningful difference between an AI that gives good generic answers and an AI that understands you specifically.

The Solution: Self-Model API Integration

We ran a Sprint Zero to diagnose the full scope of the problem. The investigation confirmed what the engagement and retention data suggested: the missing piece was not model quality — it was user context.

The solution was integrating Clarity’s self-model API into Mystica’s existing AI stack. A self-model is a persistent, evolving representation of an individual user — their beliefs, preferences, goals, history, and context. Instead of starting each conversation from scratch, the AI could now access a rich understanding of who it was talking to before generating a response.

Here is what the integration looked like in practice:

What the Self-Model Captured

The self-model API tracked several dimensions of each user:

• Conversation history synthesis — Not raw logs, but extracted themes, patterns, and progress over time
• Stated goals and values — What the user said they cared about and were working toward
• Interaction preferences — How the user preferred to receive guidance (direct, gentle, exploratory)
• Journey markers — Significant moments, breakthroughs, and recurring themes across sessions

How It Changed the AI’s Behavior

With the self-model informing each response, the AI could:

• Continue conversations instead of restarting them
• Reference past interactions naturally (“Last week you mentioned feeling stuck with your career decision — how has that been evolving?”)
• Adapt its communication style to match each user’s preferred approach
• Track patterns that the user might not see themselves (“You have brought up this theme three times now — would you like to explore it more deeply?”)

The technical integration was straightforward. Clarity’s self-model API sits alongside the existing LLM pipeline. Before the AI generates a response, it queries the self-model for the current user’s context. That context is injected into the prompt, giving the model the information it needs to respond as someone who knows the user — because it does.

The Delivery: 6 Weeks from Kickoff to Production

The engagement ran 6 weeks from kickoff to production launch:

Weeks 1-2: Sprint Zero and Architecture

We audited Mystica’s existing stack, mapped user journey patterns, and designed the integration architecture. The Sprint Zero identified the self-model as the highest-impact intervention and scoped the implementation work.

Weeks 3-4: Integration and Testing

We integrated the self-model API into Mystica’s conversation pipeline. The core work was designing the observation contexts — deciding what the self-model should track for each user and how that information should surface in the AI’s responses.

Weeks 5-6: Rollout and Validation

We launched the self-model-powered experience to users and monitored the initial results. The early signals were strong: users were having longer conversations, returning more frequently, and engaging with the AI in ways that suggested the continuity was working.

The Results: 60% Revenue Increase in 90 Days

Within 90 days of launching the self-model integration, Mystica’s revenue increased 60%.

0%

revenue increase after 90 days

0 wk

delivery timeline

0

API integration (self-model)

The revenue increase was driven by improved retention. Users who experienced the self-model-powered AI were more likely to stay subscribed, more likely to engage deeply, and more likely to expand their usage of the platform.

This is not a story about a fancier model or a larger training dataset. The underlying LLM did not change. The prompts did not get dramatically more complex. What changed was the context available to the AI when it generated a response. That context — the self-model — turned a stateless chatbot into a product that actually knew its users.

What This Means for Your AI Product

Mystica’s challenge is not unique to spiritual wellness apps. Any AI product that serves returning users faces the same fundamental question: does your AI understand who it is talking to?

If the answer is no, you are likely experiencing the same symptoms Mystica had — declining engagement, flattening retention, users who try the product and drift away. The model is probably fine. The missing piece is user context.

Here is how to evaluate whether your product has this gap:

1. Ask your AI a question, then come back tomorrow and ask a follow-up. Does it pick up where you left off, or start over?
2. Compare responses for a new user and a 30-day user. Are they meaningfully different? If not, your AI is not learning from interactions.
3. Check your retention curves. If engagement drops sharply after the first few sessions, your users are experiencing the same context amnesia that Mystica’s users experienced.

If any of these resonate, the fix is architectural — not more prompt engineering.

Next Steps

Read the full Mystica case study for additional details on the engagement and results.

If your AI product is experiencing the same patterns — declining retention, generic responses, context amnesia — we can help. Clarity’s Sprint Zero assessment diagnoses whether a self-model integration is the right intervention for your specific situation, and our services cover the full implementation lifecycle.

The question is not whether your AI needs user context. It is how long you can afford to operate without it.