Simulated Focus Groups AI-Powered Ideation

Welcome back to another edition of AI Insiders. To start things off, we have a special announcement from our guest Jennifer Fuentes. Interestingly, both of our guests today came from conversations at CES this year—truly the gift that keeps on giving.

### (00:00:35)

We’ll introduce our featured speaker in just a minute, but first, Jenna Fuentes is doing some great work at the nonprofit OpenClassrooms and has a few quick updates. Jenna, please take it away.

### (00:00:56)

Thank you, David. Hello, everyone—it's great to meet you. I’d love to connect with each of you.

We also have an upcoming event in New York, and I encourage anyone interested in networking or discussing the future of work and team building to join. I’m sharing details in the chat and, if possible, I can share my screen as well. The event is on April 30th at Impact Hub, and I’d love to see you there.

To share a bit about myself, my background is in higher ed and edtech. I’ve lived in Spain and Argentina, and at my previous edtech startup, we scaled from 100 to 1,000 employees in three years and were acquired for $750 million.

I’ve also consulted for clients in Germany, Mexico, the Philippines, and the US. Today’s topic is AI in marketing and sales, and our company is based in France with expansions in the UK. I manage operations in California and New York, and we are a B Corporation that measures our carbon footprint.

We provide free training and apprenticeships to help people grow. We also offer free marketing training and act as a recruitment source, providing marketing talent at $20 per hour.

Our programs are open for discussion, and I’m happy to go into more detail, but I’ll keep this short. If you’re interested in hiring remote, onsite, or hybrid marketing apprentices—or want to learn data analytics and marketing for free—we’re sponsoring several events, including this webinar, and would love to collaborate further.

### (00:03:11)

Thank you, Jennifer. It’s inspiring to see your work, and we appreciate OpenClassrooms' support for AI Marketers and Serial Marketers. Everyone, please check out the link Jennifer shared and connect if you can attend the event next week—there’s much more to explore.

### (00:03:29)

Now, I’d like to introduce a principal from another European company, whom I also met at CES, thanks to Chris Path. Chris introduced me to Céline from Largo AI. As soon as I learned what Largo AI was doing, I knew it was something special. Watching Céline and her team explain their work was a revelation, especially their unique approach to simulated focus groups compared to synthetic data or synthetic audiences.

### (00:04:47)

Rather than continue talking, I’d like to welcome Céline and hear directly from her. These sessions are participatory, so please ask questions—Céline is eager to engage. Welcome, Céline.

### (00:05:09)

Thank you, David. Hello everyone, and thank you for joining. I’m calling from Switzerland, hence my French accent. We’re based on the French side of Switzerland. I want to keep this interactive, so please feel free to ask questions or share comments at any point.

Let me give you some background on what we do at Largo AI, our approach to AI, and I’ll share a concrete example of a project advertisement campaign, showing you real results directly on our platform.

### (00:05:56)

I’m Céline, the Chief Operating Officer at Largo. We were founded six years ago in Switzerland as a spin-off from EPFL, one of the world’s top tech universities for computer science. We have a large R&D department and conduct extensive research to identify the most effective AI methods. Over the years, we’ve also discovered many methods that don’t work, especially in the area of simulated focus groups, and I’ll share those insights as well.

### (00:06:46)

Largo originally started in the movie industry, and we now have over 600 production companies as clients, ranging from independent filmmakers to large studios. Over the past year, we’ve entered the advertising industry, working directly with brands and agencies.

### (00:07:21)

Let me share my screen to demonstrate how the platform works. We are a post-Series A company, having closed our latest funding round this year. Our platform operates on a yearly subscription model, allowing brands and agencies to access it hands-on. You can log in, test your advertising ideas or new concepts, and get rapid feedback.

### (00:07:58)

We use simulated focus groups. Instead of waiting weeks and incurring high costs for traditional focus groups, our platform allows you to test a wide range of content quickly and affordably. Here’s how it works: We replicate real people to test ideas, enabling faster, more scalable insights than traditional methods.

### (00:09:08)

It’s important to clarify: simulated personas are different from synthetic personas. Synthetic people are created entirely by AI, while simulated personas are based on real individuals. For example, we take someone like David, have them complete an in-depth survey developed with EPFL researchers, and then simulate their behavior. We check the quality by comparing responses from the real person and their digital twin; if the correlation isn’t over 90%, we eliminate that persona.

### (00:10:47)

To use the platform, you select the country you want to target—say, the United States. The AI will not only test your campaign but also benchmark it by generating and comparing similar ideas, either reflecting your brand identity or those of competitors. You can upload a brief (PDF or text), and the AI will analyze it, summarize the key details, and generate a key image.

### (00:12:24)

You’ll then see AI-generated ideas for benchmarking, alongside your own. The results show audience predictions, such as the demographic breakdown in your selected country. If you want to target a specific group (e.g., only women or a certain age), you can set those parameters. Simulated people—anonymized for privacy—act as your focus group. Your idea is presented to them online, and you see results including emotional responses, likability, and buying interest, compared to the AI-generated benchmarks.

### (00:14:30)

For example, we worked with a sports equipment company in Mexico preparing to launch a soccer concept store in the US. Their initial idea performed poorly on our platform, but the qualitative feedback showed that their concept—a "temple of soccer"—didn’t resonate with US audiences. The AI suggested adding arcade games, and after making those changes, they created a successful new concept.

### (00:16:08)

A common question is how our simulations can accurately reflect someone’s preferences when we haven’t asked every possible specific question. We’ve invested seven years of research to refine our AI, and we found that using only synthetic (AI-created) personas resulted in average, generic feedback that missed key personality differences. By basing simulations on real people and continually updating our models, we achieve much more nuanced and accurate results.

### (00:18:22)

Regarding audience panels, we already have a large database of simulated people, and you can specify highly targeted "micro-audiences" by interests, demographics, and more. If we lack enough of a particular segment, we can run new surveys to create additional simulated personas. For extremely niche groups or those difficult to recruit (e.g., young children or elderly), we use AI-generated synthetic personas to supplement the data.

### (00:20:00)

Our platform can handle a wide variety of products and industries, including software, consumer goods, airlines, sports, and skincare. You can define micro-audiences with detailed interests or behaviors—such as cat owners who prefer real meat and natural ingredients—and we’ll either use our existing panel or create new simulations as needed. Analyses typically take 15 minutes, but if a new audience is required, the process may take about two weeks.

### (00:22:00)

We’re active in over 20 countries, with thousands of simulated personas in each. For even more specificity—such as targeting Gen Z or unique psychographic groups—we supplement with synthetic personas when real ones are not available, though the accuracy is somewhat lower.

### (00:23:27)

We offer yearly subscriptions only, starting at $12,000 and scaling up based on the number of analyses needed. With a $20,000 package, you can run about 20 projects—about ten times more cost-effective than traditional focus groups.

### (00:24:25)

Regarding the intake process for recruiting panelists, we ask a comprehensive set of questions to understand individual behaviors and preferences, similar to classic research methodologies (like MRI or GWI panels). While the full list is proprietary, we use hundreds of questions to build a reliable digital profile, ensuring accurate simulations for diverse and specific needs.

---

### (00:28:57)

Cultural nuance is addressed by designing specific surveys for each country—there’s no one-size-fits-all approach. For example, responses to horror films differ dramatically between countries like Finland and Italy, so we tailor our process to each region. Our R&D team constantly updates and expands these country-specific surveys.

---

### (00:40:14)

We can accommodate sensitive audiences, such as those in healthcare. If a required micro-audience is too niche, we combine real and synthetic personas, always ensuring privacy and compliance. Simulated focus groups are also highly secure, as confidential concepts aren’t exposed to real people who might leak information.

---

### (00:41:41)

When launching in a new market, we typically start with a few thousand respondents per country and constantly expand. If you need to test a specific group that’s underrepresented, we can quickly scale up through additional recruitment and automation.

---

### (00:44:35)

Simulated focus groups reduce bias found in traditional groups. By analyzing massive historical data (especially from the film industry), we detect authentic emotional responses, using a kind of “cinematographic DNA” to predict how people actually feel, even if they can’t fully articulate it themselves.

---

### (00:46:02)

The data is primarily survey-based, supplemented with extensive analysis of content and media. Our AI models break down text, visuals, and audio to connect specific creative elements to emotional outcomes.

---

### (00:47:18)

There are some limitations: Simulated personas cannot currently test physical products or interact with web pages in real time. However, you can upload screenshots or design concepts to gather feedback. Full user experience testing is not yet possible, but may be in the future.

---

### (00:49:04)

Our system accepts text, video, or audio. At an early stage, you can simply describe your idea, or provide documents and visuals for more advanced analysis. The system summarizes the content and generates feedback from the simulated audience.

---

### (00:50:28)

We understand there’s skepticism about whether a solution that’s faster and ten times cheaper can be accurate. That’s why we offer a free trial: you can test a past campaign and compare the simulated focus group’s feedback to your actual results. So far, our results have matched over 90% of the time.

---

### (00:53:00)

The platform was originally developed for the movie industry, but can be used for email responses, business concepts, and other formats. Any idea or creative concept that could be tested in a traditional focus group can be pretested using our simulation.

---

### (00:54:09)

Thank you all for your questions and participation. Céline, thank you for sharing these insights with us, and Chris, thank you for the introduction. We appreciate everyone’s engagement today and look forward to seeing what Largo AI does next.

---

### (00:54:34)

Thank you very much for inviting me and for all your thoughtful questions. It was a pleasure to present today. [Music]

---

**Let me know if you need further editing, want this divided into even shorter sections, or need a downloadable format!**