> ## Documentation Index > Fetch the complete documentation index at: https://docs.mavera.io/llms.txt > Use this file to discover all available pages before exploring further. # Persona Interview Marathon > Use Speak to have voice conversations with 5 different personas about your positioning — rapid persona interviewing without recruitment or scheduling ## The Scenario You need qualitative positioning feedback from five distinct audience segments — but you don't have three weeks for recruitment, incentive budgets, or a moderator. With Mavera's Speak surface, you have live voice conversations with AI personas, probe their reactions in real time, and extract structured insights. Five personas, five conversations, one afternoon. **Mavera-only workflow.** No recruitment platforms, no incentive payments, no scheduling tools. Just Mavera's Personas, Speak, and Chat surfaces. *** ## When to Use This * Early-stage positioning validation with 2–3 hypotheses that need gut-checks. * Founder-led discovery — practice your pitch with synthetic personas before real prospects. * Message hierarchy testing — which proof points resonate first with each segment? * Pre-launch readiness — talk to 5 ICP segments and capture objections. *** ## Architecture ```mermaid theme={"dark"} flowchart LR A["Create Personas"] --> B["Start Speak Sessions"] B --> C["Conduct Interviews"] C --> D["Extract Insights"] D --> E["Report"] ``` | Mavera Surface | Role in Pipeline | | ------------------------------- | ---------------------------------------------- | | **Personas** (`POST /personas`) | Create 5 ICP personas with distinct priorities | | **Speak** (`POST /speak`) | Open voice conversation sessions | | **Chat** (OpenAI-compatible) | Extract structured insights from transcripts | *** ## What You Need | Requirement | Details | | ---------------------------------- | ---------------------------------------------------------------------------------------------------- | | **Mavera API key** | Starts with `mvra_live_`. Get one at [Developer Settings](https://app.mavera.io/settings/developer). | | **Python 3.8+** or **Node.js 18+** | `requests`/`openai` for Python; native `fetch` for Node. | | **Credits** | \~115–265 total. See [Credits Estimate](#credits-estimate). | ``` MAVERA_API_KEY=mvra_live_your_key_here ``` *** ## Step 1 — Create 5 ICP Personas ```python Python theme={"dark"} import os, time, json, requests API_KEY = os.environ["MAVERA_API_KEY"] BASE = "https://app.mavera.io/api/v1" HEADERS = {"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"} PERSONAS = [ {"name": "Elena — SaaS Marketing Director", "role": "Marketing Director, 200-person B2B SaaS", "description": "Manages team of 8. Spending $40K/quarter on research agencies. Frustrated by slow turnaround. Needs HubSpot integration and 90-day ROI.", "traits": ["ROI-focused", "time-pressed", "integration-demanding", "data-driven"]}, {"name": "Raj — E-commerce Founder", "role": "Founder & CEO, DTC brand ($5M ARR)", "description": "Bootstrapped, writes ad copy personally. Audience is millennial women 25–34. Relies on gut instinct and A/B testing. Budget-conscious but invests for ROAS.", "traits": ["scrappy", "gut-instinct-driven", "ROAS-obsessed", "hands-on"]}, {"name": "Dr. Amara — Healthcare Comms Lead", "role": "Director of Communications, hospital network", "description": "Navigates HIPAA compliance and health literacy. Content for diverse populations with varying reading levels. Multiple compliance review cycles.", "traits": ["compliance-aware", "patient-first", "health-literacy-focused", "risk-averse"]}, {"name": "Marcus — Agency Creative Director", "role": "Creative Director, mid-size ad agency", "description": "Oversees 12 clients in CPG, fintech, retail. Team of 15. Constantly pitching new business. AI-curious but skeptical about quality vs human creativity.", "traits": ["quality-obsessed", "client-facing", "AI-curious-but-skeptical", "pitch-driven"]}, {"name": "Keiko — Enterprise Product Manager", "role": "Senior PM, Fortune 500 tech company", "description": "Platform with 10K+ enterprise customers. Validates feature positioning before engineering invests. Currently uses slow customer advisory boards and lagging NPS surveys.", "traits": ["analytical", "consensus-builder", "evidence-driven", "launch-velocity-focused"]}, ] def create_persona(p): resp = requests.post(f"{BASE}/personas", headers=HEADERS, json={ "name": p["name"], "role": p["role"], "description": p["description"], "traits": p["traits"]}) resp.raise_for_status() data = resp.json() print(f" Created: {data['name']} → {data['id']}") return data["id"] persona_ids = {} for p in PERSONAS: persona_ids[p["name"].split(" — ")[0]] = create_persona(p) ``` ```javascript JavaScript theme={"dark"} const API_KEY = process.env.MAVERA_API_KEY; const BASE = "https://app.mavera.io/api/v1"; const HEADERS = { Authorization: `Bearer ${API_KEY}`, "Content-Type": "application/json" }; const PERSONAS = [ { name: "Elena — SaaS Marketing Director", role: "Marketing Director, 200-person B2B SaaS", description: "Team of 8. $40K/quarter on agencies. Needs HubSpot integration and 90-day ROI.", traits: ["ROI-focused", "time-pressed", "integration-demanding", "data-driven"] }, { name: "Raj — E-commerce Founder", role: "Founder & CEO, DTC brand ($5M ARR)", description: "Bootstrapped, writes ad copy. Audience: millennial women 25–34. ROAS-obsessed.", traits: ["scrappy", "gut-instinct-driven", "ROAS-obsessed", "hands-on"] }, { name: "Dr. Amara — Healthcare Comms Lead", role: "Director of Communications, hospital network", description: "HIPAA compliance, health literacy. Diverse populations, multiple review cycles.", traits: ["compliance-aware", "patient-first", "health-literacy-focused", "risk-averse"] }, { name: "Marcus — Agency Creative Director", role: "Creative Director, mid-size agency", description: "12 clients, team of 15. Constantly pitching. AI-curious but quality-skeptical.", traits: ["quality-obsessed", "client-facing", "AI-curious-but-skeptical", "pitch-driven"] }, { name: "Keiko — Enterprise Product Manager", role: "Senior PM, Fortune 500 tech", description: "10K+ customers. Validates positioning before eng invests. Slow advisory boards.", traits: ["analytical", "consensus-builder", "evidence-driven", "launch-velocity-focused"] }, ]; async function createPersona(p) { const resp = await fetch(`${BASE}/personas`, { method: "POST", headers: HEADERS, body: JSON.stringify({ name: p.name, role: p.role, description: p.description, traits: p.traits }) }); const data = await resp.json(); if (data.error) throw new Error(data.error.message); return data.id; } const personaIds = {}; for (const p of PERSONAS) personaIds[p.name.split(" — ")[0]] = await createPersona(p); ``` *** ## Step 2 — Define Interview Script Seven phases from opening reaction through pricing to a closing suggestion. Consistent across all 5 interviews. ```python Python theme={"dark"} POSITIONING = """ Mavera is an AI-powered audience research platform that lets marketing teams test messaging, positioning, and creative with synthetic personas in minutes instead of weeks. No recruitment, no scheduling, no incentive budgets. """ INTERVIEW_SCRIPT = [ {"phase": "Opening", "prompt": f"React to this positioning:\n\n{POSITIONING}\n\nGut reaction?"}, {"phase": "Comprehension", "prompt": "In your own words, what does this product do? Who is it for?"}, {"phase": "Relevance", "prompt": "How relevant is this to your daily work? Give a specific recent example."}, {"phase": "Differentiation", "prompt": "What alternatives do you use today? How does this compare?"}, {"phase": "Objections", "prompt": "If someone pitched this to you, what's your first concern?"}, {"phase": "Pricing", "prompt": "What would you expect to pay? What feels fair vs too expensive?"}, {"phase": "Closing", "prompt": "One thing you'd change about how this product is described?"}, ] ``` ```javascript JavaScript theme={"dark"} const POSITIONING = `Mavera is an AI-powered audience research platform that lets marketing teams test messaging with synthetic personas in minutes instead of weeks. No recruitment, no scheduling.`; const INTERVIEW_SCRIPT = [ { phase: "Opening", prompt: `React to this:\n\n${POSITIONING}\n\nGut reaction?` }, { phase: "Comprehension", prompt: "In your own words, what does this do? Who is it for?" }, { phase: "Relevance", prompt: "How relevant to your daily work? Give a specific example." }, { phase: "Differentiation", prompt: "What alternatives do you use? How does this compare?" }, { phase: "Objections", prompt: "If someone pitched this, what's your first concern?" }, { phase: "Pricing", prompt: "What would you expect to pay? Fair vs too expensive?" }, { phase: "Closing", prompt: "One thing you'd change about the description?" }, ]; ``` *** ## Step 3 — Run Speak Sessions Start a Speak session for each persona and run through the interview script turn by turn. ```python Python theme={"dark"} def run_interview(name, pid): print(f"\n{'─' * 50}\nINTERVIEW: {name}\n{'─' * 50}") session = requests.post(f"{BASE}/speak", headers=HEADERS, json={ "persona_id": pid, "context": "You're in a product positioning interview. Answer honestly from your professional perspective.", }).json() transcript = [] for step in INTERVIEW_SCRIPT: response = requests.post(f"{BASE}/speak/{session['id']}/messages", headers=HEADERS, json={"message": step["prompt"]}).json() answer = response.get("transcript", response.get("content", "")) print(f" [{step['phase']}] {answer[:100]}...") transcript.append({"phase": step["phase"], "question": step["prompt"], "answer": answer, "persona": name}) return transcript all_transcripts = {} for name, pid in persona_ids.items(): all_transcripts[name] = run_interview(name, pid) ``` ```javascript JavaScript theme={"dark"} async function runInterview(name, pid) { console.log(`\n${"─".repeat(50)}\nINTERVIEW: ${name}\n${"─".repeat(50)}`); const session = await fetch(`${BASE}/speak`, { method: "POST", headers: HEADERS, body: JSON.stringify({ persona_id: pid, context: "You're in a positioning interview. Answer honestly from your professional perspective." }), }).then((r) => r.json()); const transcript = []; for (const step of INTERVIEW_SCRIPT) { const response = await fetch(`${BASE}/speak/${session.id}/messages`, { method: "POST", headers: HEADERS, body: JSON.stringify({ message: step.prompt }), }).then((r) => r.json()); const answer = response.transcript || response.content || ""; console.log(` [${step.phase}] ${answer.slice(0, 100)}...`); transcript.push({ phase: step.phase, question: step.prompt, answer, persona: name }); } return transcript; } const allTranscripts = {}; for (const [name, pid] of Object.entries(personaIds)) allTranscripts[name] = await runInterview(name, pid); ``` Speak sessions are sequential — each turn waits for the response. Budget \~2–3 minutes per interview, \~10–15 minutes for all 5. *** ## Step 4 — Extract Structured Insights Feed each transcript into Chat with a structured output schema. ```python Python theme={"dark"} from openai import OpenAI mavera = OpenAI(api_key=API_KEY, base_url="https://app.mavera.io/api/v1") INSIGHT_SCHEMA = {"type": "json_schema", "json_schema": {"name": "interview_insights", "strict": True, "schema": {"type": "object", "properties": { "positioning_clarity": {"type": "number"}, "relevance_score": {"type": "number"}, "primary_objection": {"type": "string"}, "price_expectation_low": {"type": "number"}, "price_expectation_high": {"type": "number"}, "current_alternatives": {"type": "array", "items": {"type": "string"}}, "key_quote": {"type": "string"}, "suggested_change": {"type": "string"}, "would_buy": {"type": "string", "enum": ["yes", "maybe", "no"]}, "top_3_insights": {"type": "array", "items": {"type": "string"}}, }, "required": ["positioning_clarity", "relevance_score", "primary_objection", "price_expectation_low", "price_expectation_high", "current_alternatives", "key_quote", "suggested_change", "would_buy", "top_3_insights"]}}} def extract_insights(name, transcript): formatted = "\n\n".join(f"[{t['phase']}]\nQ: {t['question']}\nA: {t['answer']}" for t in transcript) resp = mavera.responses.create(model="mavera-1", input=[ {"role": "system", "content": "Extract structured insights from this positioning interview. Use exact quotes."}, {"role": "user", "content": f"Interview with {name}:\n\n{formatted}"}, ], extra_body={"response_format": INSIGHT_SCHEMA}) return json.loads(resp.output[0].content[0].text) all_insights = {} for name, transcript in all_transcripts.items(): ins = extract_insights(name, transcript) all_insights[name] = ins print(f" {name}: clarity={ins['positioning_clarity']}/10, " f"relevance={ins['relevance_score']}/10, would_buy={ins['would_buy']}") ``` ```javascript JavaScript theme={"dark"} const OpenAI = require("openai").default; const mavera = new OpenAI({ apiKey: API_KEY, baseURL: "https://app.mavera.io/api/v1" }); const INSIGHT_SCHEMA = { type: "json_schema", json_schema: { name: "interview_insights", strict: true, schema: { type: "object", properties: { positioning_clarity: { type: "number" }, relevance_score: { type: "number" }, primary_objection: { type: "string" }, price_expectation_low: { type: "number" }, price_expectation_high: { type: "number" }, current_alternatives: { type: "array", items: { type: "string" } }, key_quote: { type: "string" }, suggested_change: { type: "string" }, would_buy: { type: "string", enum: ["yes", "maybe", "no"] }, top_3_insights: { type: "array", items: { type: "string" } }, }, required: ["positioning_clarity", "relevance_score", "primary_objection", "price_expectation_low", "price_expectation_high", "current_alternatives", "key_quote", "suggested_change", "would_buy", "top_3_insights"] } } }; async function extractInsights(name, transcript) { const formatted = transcript.map((t) => `[${t.phase}]\nQ: ${t.question}\nA: ${t.answer}`).join("\n\n"); const resp = await mavera.responses.create({ model: "mavera-1", input: [ { role: "system", content: "Extract structured insights from this interview. Use exact quotes." }, { role: "user", content: `Interview with ${name}:\n\n${formatted}` }, ], extra_body: { response_format: INSIGHT_SCHEMA } }); return JSON.parse(resp.output[0].content[0].text); } const allInsights = {}; for (const [name, transcript] of Object.entries(allTranscripts)) allInsights[name] = await extractInsights(name, transcript); ``` *** ## Step 5 — Marathon Summary Report ```python Python theme={"dark"} def print_marathon_report(insights): print("\n" + "=" * 70) print("PERSONA INTERVIEW MARATHON — SUMMARY") print("=" * 70) print(f"\n{'Persona':<20} {'Clarity':>8} {'Relevance':>10} {'Buy':>6} {'Price Range':>15}") print("─" * 60) for name, d in insights.items(): pr = f"${d['price_expectation_low']}–${d['price_expectation_high']}" print(f"{name:<20} {d['positioning_clarity']:>7}/10 {d['relevance_score']:>9}/10 " f"{d['would_buy']:>6} {pr:>15}") print(f"\n{'─' * 60}\nOBJECTIONS\n{'─' * 60}") for name, d in insights.items(): print(f" {name}: {d['primary_objection']}") print(f"\n{'─' * 60}\nKEY QUOTES\n{'─' * 60}") for name, d in insights.items(): print(f" {name}: \"{d['key_quote']}\"") # Alternatives frequency alts = {} for d in insights.values(): for a in d.get("current_alternatives", []): alts[a] = alts.get(a, 0) + 1 print(f"\n{'─' * 60}\nALTERNATIVES MENTIONED\n{'─' * 60}") for a, c in sorted(alts.items(), key=lambda x: -x[1]): print(f" {a}: {c} persona(s)") print_marathon_report(all_insights) with open("interview_marathon.json", "w") as f: json.dump(all_insights, f, indent=2) ``` ```javascript JavaScript theme={"dark"} function printMarathonReport(insights) { console.log("\n" + "=".repeat(70)); console.log("PERSONA INTERVIEW MARATHON — SUMMARY"); console.log("=".repeat(70)); for (const [name, d] of Object.entries(insights)) { console.log(` ${name.padEnd(20)} clarity=${d.positioning_clarity}/10 relevance=${d.relevance_score}/10 ` + `buy=${d.would_buy} $${d.price_expectation_low}–$${d.price_expectation_high}`); } console.log("\nOBJECTIONS:"); for (const [n, d] of Object.entries(insights)) console.log(` ${n}: ${d.primary_objection}`); console.log("\nKEY QUOTES:"); for (const [n, d] of Object.entries(insights)) console.log(` ${n}: "${d.key_quote}"`); } printMarathonReport(allInsights); const fs = require("fs"); fs.writeFileSync("interview_marathon.json", JSON.stringify(allInsights, null, 2)); ``` *** ## Example Output ``` PERSONA INTERVIEW MARATHON — SUMMARY ══════════════════════════════════════════════════════════════════════════ Persona Clarity Relevance Buy Price Range ──────────────────────────────────────────────────────────────── Elena 8/10 9/10 yes $300–$800/mo Raj 7/10 6/10 maybe $100–$300/mo Dr. Amara 6/10 5/10 maybe $200–$500/mo Marcus 9/10 8/10 yes $500–$1500/mo Keiko 8/10 7/10 maybe $400–$1000/mo KEY QUOTES ──────────────────────────────────────────────────────────────── Elena: "If this replaces one agency cycle per quarter, it pays for itself." Raj: "Cool tech, but I need to see it beat my gut on actual ROAS." Dr. Amara: "Can I trust synthetic personas for health messaging?" Marcus: "I'd use this for pitch prep in a heartbeat." Keiko: "Show me a case study where this predicted real user behavior." ``` *** ## Variations Generate a follow-up question based on the persona's answer: ```python theme={"dark"} def adaptive_probe(session_id, previous_answer, phase): probe = mavera.responses.create(model="mavera-1", input=[ {"role": "system", "content": "Generate one follow-up question probing deeper."}, {"role": "user", "content": f"Phase: {phase}\nAnswer: {previous_answer}"}, ]) return requests.post(f"{BASE}/speak/{session_id}/messages", headers=HEADERS, json={"message": probe.output[0].content[0].text}).json() ``` ```python theme={"dark"} for i, pos in enumerate([POSITIONING_A, POSITIONING_B]): INTERVIEW_SCRIPT[0]["prompt"] = f"React to this:\n\n{pos}\n\nGut reaction?" for name, pid in persona_ids.items(): run_interview(f"{name}_v{i}", pid) ``` ```python theme={"dark"} objections = [ins["primary_objection"] for ins in all_insights.values()] fg_questions = [{"question": f"How concerning is: '{obj}'", "type": "NPS", "order": i+1} for i, obj in enumerate(objections)] fg = requests.post(f"{BASE}/focus-groups", headers=HEADERS, json={ "name": "Objection Validation", "persona_ids": list(persona_ids.values()), "sample_size": 25, "questions": fg_questions}).json() ``` ```python theme={"dark"} with open("interviews.md", "w") as f: for name, transcript in all_transcripts.items(): f.write(f"# {name}\n\n") for t in transcript: f.write(f"## {t['phase']}\n**Q:** {t['question']}\n**A:** {t['answer']}\n\n---\n\n") ``` *** ## Credits Estimate | Operation | Typical Cost | Notes | | --------------------------------- | ------------- | -------------------------------- | | Persona creation (×5) | 0–25 | One-time; reuse across runs | | Speak sessions (×5, 7 turns each) | 100–200 | Primary cost driver | | Chat insight extraction (×5) | 15–40 | Structured output per transcript | | **Total** | **\~115–265** | | Keep interviews to 5–7 questions for the best depth/cost balance. Add follow-up probes selectively rather than for every question. *** ## What's Next Switch from interviews to a structured Focus Group with 10 buying-committee personas Quantify interview findings with a systematic message × persona grid Pit opposing buyer types against each other for pricing insights Test across age demographics instead of role-based personas Choose the right persona types for your research goal Pre-flight checks and usage tracking