How Synilly Works & Where We're Going

Why this matters: Traditional surveys know your age and income. We know that Maya adopted her dog Mochi after a breakup, switched to premium organic food after a health scare, and secretly buys things she publicly calls overpriced. That depth is what makes conversations feel real.

Why this matters: The same pricing question produces authentically different responses because each persona's background creates distinct reasoning patterns — just like real people.

Example Match Reasons

• Maya: "Brand overlap — uses Farmer's Dog" + "High pet spending" + "Target age range"
• Derek: "Research-driven buyer" + "Premium brand experience" + "Prior pet nutrition study"

The experience: Our AI moderator guides the conversation in real time. Messages appear naturally, personas react to each other, and an insights sidebar updates as themes emerge. It feels like watching a real focus group unfold.

Maya's History

The pattern: Maya brought up her pet food frustration in a completely different study — unprompted. Her price sensitivity was consistent across both conversations.

Derek's History

Across 5 studies, Derek has always asked for data export and clinical evidence. That kind of consistency is what makes these personas believable.

4 of 5 panelists flagged $49/mo as too high. Acceptable range: $29–39/mo.

Every panelist demanded concrete evidence. The type of proof varies by segment — Maya wants social proof, Derek wants clinical data.

Recommendation (High Priority)

Introduce a "Guidance Only" tier at $15–20/mo — separating the nutritionist advice from the food delivery.

The Problem

Our personas are plausible but not verified. Maya's $95k income and Portland location were created by a human making educated guesses. We have no way to prove this profile reflects real people in meaningful numbers.

The Solution

Feed real data sources — census records, salary databases, consumer surveys — into an AI persona builder. Each generated persona comes with citations pointing back to the real data. Demographics are statistically representative, not imagined.

Data We'd Use

• US Census — demographics, income, geography
• Bureau of Labor Statistics — job titles, salary ranges by region
• APPA Pet Owners Survey — pet spending data
• Pew Research — media habits, tech adoption rates

Briefing for the AI persona builder:

Request: Create a 30-something UX designer in Portland who owns a dog.

Real data provided:

• Portland, OR median income: $73,340 (Census 2024)
• UX Designers in Portland earn $85k–$110k (Labor statistics)
• 67% of millennials own pets (APPA 2024)
• Pet owners age 25–34 spend an average of $1,480/year (APPA)

Rules:

• Income must fall within the real salary range for that job
• Pet spending must match what the surveys actually show
• Every demographic fact must cite its source
• Personality can be creative, but can't contradict the data

The key gain: When a client asks "Is this persona representative?" we can point to real data instead of saying "we think so."

What We Solved

Traditional synthetic research uses scripted conversations — same question always produces the same answer. We built something fundamentally different:

• An AI Moderator guides the conversation, decides who speaks next, and follows up on surprises
• Each AI Persona responds in character, drawing on their full background profile
• Personas react to each other — agreeing, disagreeing, or building on each other's points

How It Works

You share the participant personas and set the research brief. The AI moderator runs the session. Same brief, different run = different conversation. The moderator probes deeper when something unexpected comes up. Insights emerge naturally.

The insight: Price isn't actually the issue — brand trust is. This kind of discovery happens because our personas react to each other in real time.

Our core advantage: This is what makes Synilly an actual research tool, not a demo. Every improvement ahead builds on top of this live AI foundation.

The Problem

We have no way to measure if our synthetic research is accurate. Our confidence scores (80%, 90%) are educated guesses. A client has no reason to trust them over their own intuition.

The Solution

Partner with clients who are also running real focus groups. Run the same brief through Synilly. Compare the outputs. Build a growing dataset that shows where synthetic research agrees with reality — and where it diverges.

What This Unlocks

• Confidence scores backed by real evidence
• Known blind spots (e.g., "We tend to underestimate price sensitivity by 15%")
• Correction factors we can apply to future results
• A credibility story that wins enterprise clients

The trust equation: A tool that says "we're 74% confident and here's why" is more trustworthy than one that claims "90% confident" with nothing behind it.

The Problem

Our personas exist in a frozen world. Maya's opinions about The Farmer's Dog were written months ago. She can't reference a recent price increase, a viral social media controversy, or a competitor's new product.

The Solution

Before each conversation, gather live context relevant to the research topic:

• Pricing: What competitors are actually charging right now
• Reviews: What real people are saying on Reddit, Amazon, Trustpilot
• Social buzz: Trending conversations on TikTok and Instagram
• News: Relevant industry developments

Each persona gets news filtered through their own media diet. Maya sees Instagram and TikTok content. Derek sees Reddit threads and YouTube reviews. Just like real people consume different media.

The insight upgrade: From "people think $49 is a lot" (obvious) to "$49 is competitive vs Farmer's Dog but weak vs Spot & Tango" (actionable).

The Problem

We present findings as if 5 personas is a valid sample. "4 out of 5 agree" sounds compelling, but a sample of 5 has almost no statistical significance. Any experienced researcher will immediately question this.

The Solution

• Run multiple sessions: Same brief, 10 different panels of 5 = 50 total opinions
• Calculate real confidence: "Price concern appeared in 78% of sessions (range: 65–88%)"
• Warn when sample is too small: Flag claims that need more data
• Detect bias: Check if our persona demographics over- or under-represent certain populations
• Test sensitivity: Would the finding change with different personas?

Our advantage: Running 10 synthetic sessions costs minutes and pennies. Real research at that scale costs tens of thousands of dollars. We should exploit this.

The paradox: Showing less certainty (ranges, bias warnings) actually increases trust with sophisticated research buyers.

1. The persona foundation is strong

Rich persona profiles with built-in psychological contradictions, memory across studies, and nuanced brand relationships. Our persona depth is genuinely deeper than most competitors.

2. Live AI conversations are already working

Our AI backend already powers moderated discussions between persona agents. Every session produces unique, dynamic conversations with real emergent insights. The core engine is shipped.

3. Credibility requires honesty

Statistical rigor and real-world calibration are what turn "interesting tool" into "tool a researcher would actually trust." Showing uncertainty earns trust.

4. Our long-term advantage is the data loop

Anyone can get AI to role-play a persona. Real-data grounding + calibration against actual studies + statistical validation creates a system that gets better with every use. That's the moat.