iMario 2.0: The Human Voice Behind Every Decision

Which price. Which message. Which market first. Behind every one of these decisions sits the same question: what do real people actually think?

For a hundred years the honest answer has been "in six weeks, if the budget allows". A focus group of eight. A survey that lands after the launch. Insight reserved for the brands that could afford to wait. Everyone else guessed.

I started iMario because that was never good enough, and it no longer has to be. iMario turns real human data into synthetic audiences you can ask anything, in minutes, at any scale, and then check.

The generating part was never the hard part. Any model will give you a confident answer. The hard part is earning the right to be believed. So every audience is calibrated against real human data, the benchmark is public, and we do not ship a number we cannot stand behind.

iMario 2.0 is the release where that promise becomes something you use on a Tuesday. Ask 200 physicians a question and read their answers before your coffee cools, for twenty dollars. Ask the customers already in your CRM why they left, and hear it in their own words. Ask from inside ChatGPT, Claude or Codex, without leaving the page you are working on.

The ambition is simple and large: put the human voice at the heart of every business decision, everywhere on earth. Here is what changed, in seven parts.

1. The customer data you already own can answer new questions

Every company has a graveyard of customer data. Interview recordings nobody replays. Survey exports from March. A CRM with fifty thousand contacts. Ten thousand support tickets. Each one cost real money to collect, answered one question, and went quiet.

iMario 2.0 wakes it up.

Connect Zoom, HubSpot, Salesforce, Intercom, Qualtrics, Typeform, Google Drive and the rest of your stack, or just drop in the files. iMario reads the records, the answers, the transcripts and the tickets, and builds an audience of Synthetic Individuals anchored to the real people behind them. The customer who complained about onboarding on a call in May and answered your NPS survey in June is one person in the audience, with both on file. When the next batch of calls lands, bring it in. The audience grows with your business.

Then ask what you never got to ask. Would these customers pay for the annual plan? Which objection would the new pricing trigger first? What did the people who churned in Q2 have in common? The interview you ran in March can answer today's question.

Every connector is read-only. You approve each permission in plain language before anything is read, and iMario never writes back. Before the audience is created you check the match: which column is the customer ID, which transcript belongs to whom. Every answer you get later says whether it came from that person's own data or from a modelled profile, and you can open the individual and read the real quotes behind the summary.

2. Why the answers hold

Making a synthetic person sound convincing is easy. Making a thousand of them answer the way the real population would, question after question, is the whole problem. 2.0 rebuilds every audience on three layers.

Foundation is who is in the room and in what proportions, sampled from real distributions instead of invented by a prompt. Mind is what they believe and how they weigh things, fitted to real survey data from the same population, so the splits look like the real splits. Behavior is what they have actually done, so a stated intent is tied to a track record.

Calibrated separately, fused into one person. On our public benchmark, synthetic answers agree with real survey distributions 89% of the time at group level across 11 populations, and the agreement holds when the question is repeated or rephrased.

That is the difference between a demo and a decision. When the question is "which direction" rather than "exactly how much", you can trust the direction.

3. 5.4 billion people, ready before you upload anything

You do not need your own data to start.

Market audiences. iMario 2.0 opens our own population: 5.4 billion simulated people across 59 markets, built from official statistics and kept calibrated against real survey data. Pick Germany, describe "working parents under 40 who shop online weekly", and the audience exists.

Official audiences. Some people are nearly impossible to interview. Physicians. High-income households. New parents. Professionals who use AI at work. We built these as certified audiences, anchored on real surveys of that exact group and calibrated on the way its members actually talk. Try booking twelve doctors for Thursday. Here, two hundred are waiting.

Audience creation is no longer a step. Open a task, pick who should answer, and ask.

4. Ask the way you think

The canvas is gone. In its place is one box. Type what you want to learn, attach what you want tested, choose who answers.

• Survey a thousand people on five questions. Done in minutes.
• Interview fifty of them with follow-up probes, or seat eight in a focus group and watch them react to each other.
• A/B two pricing pages, three headlines, or two offers.
• Show an ad or a design and hear what they see first and what stops them.
• Paste a URL and let them read the live page.
• Agent tests: put your AI agent in front of a room of realistic customers before it meets a real one. Coming soon.

Close the tab. The run keeps going.

5. Do not stop at the chart

A result in 2.0 is where the interesting part starts.

Click the 31% who said "too expensive" and see who they are. Cross-tab them by income, by market, by what they answered on question three. Save them as their own audience and ask the next question only to them. Ask one of them why, in a conversation, and get an answer in their own voice.

Not sure the synthetic answer is right? Send the same question to real people from inside the result and read the two side by side.

When you need to take it to the room, generate the report. iMario writes the full document over the runs you choose, every claim tied back to a respondent quote, in the shape a stakeholder expects. Focus groups are read as a group, not as a pile of separate answers.

6. iMario inside ChatGPT, Claude and Codex

From 2.0, iMario is an MCP server. Connect it once at https://mcp.imario.ai/mcp, and the tools you already work in can ask your audiences.

You are in Claude Code, halfway through a new onboarding flow. Ask the audience which version they would finish. You are in ChatGPT writing the launch email. Test three subject lines with the segment you sell to before you pick one. You are in Claude with a mockup open. Show it to the audience and ask what they would click. The answer lands in the same conversation, and the full study waits in your iMario workspace for the drill-down.

Approve the connection in ChatGPT, the Claude apps, Claude Code or Codex, or use a workspace API key from your own code. Every study is estimated before it runs, one daily limit covers every connected app, and a run that fails costs nothing.

7. One price

We dropped the tiers. iMario 2.0 is usage only: $0.05 per credit, packs from $0.04, and 1,000 free credits for every new account.

Creating audiences and individuals is free. Reports are priced on the runs they cover. No seats, nothing expires.

The 50-person interview study you commission today takes three to four weeks and costs tens of thousands of dollars. Here it takes minutes and costs under a hundred. A tenth of the price, a hundred times the speed, for the decisions you make every week: which market, which message, which price, which feature first.

Try it this afternoon

Your 1,000 free credits are enough to survey a thousand people, or interview forty. Pick a market, type the question you have been guessing at, and read the answer before your next meeting.

Start with 1,000 free credits →

Then connect your first data source in the docs, or bring iMario into your agent with the MCP guide.

And if you care about consumer insight, AI you can trust, or the future of market research, let us talk.