Synthetic Audiences: Focus Groups, AI and Simulation Engines

Just how good are your product pages, really?

One of the frustrating truths of creating for an audience is: however much work is involved, however much expertise is at your command, however spotless your past performance may be, there will never be certainty about how something new will be received.

This is as true for theatre as it is for selling toasters. Creating something new, or even presenting something old in a new light, is an inherently risky business. For any number of reasons beyond your knowledge or factors beyond your control, you can underperform. Or fail.

Given the stakes involved, it stands to reason that there are tools and strategies available to guard against this and provide insights ahead of any costly launch activity. Traditionally this was primarily human-focused: Consultant and agencies, market research, focus groups and A/B testing. But recently, a company called Socialtrait claims to have built “the world’s first Audience Behaviour Simulation Engine”.

Put simply, it uses GenAI to create AI agents that reflect the behaviour of the audience segments that interact with your product, or those you are hoping to target.

It then presents these agents with your product content or proposed branding changes, and subjects them to several tests, surveys, and virtual focus groups to establish preferences, insights, and actionable advice. All at scale, speed, and a lower associated cost than the equivalent traditional research methods.

The problem with focus groups

Why might an AI audience simulator be beneficial? Anyone who has interacted with market research knows the pitfalls and unreliability that are inherent in these methods. Humans can be fickle and contrary beings at the best of times, and attempting to get feedback that is honest and accurate from a focus group can be challenging. Finding participants that align with your targeted audience segments is often difficult. Participants can often be few and self-selecting due to the circumstances of the role. Even with expert moderation, it is easy for insights to become skewed by any number of social factors.

In comparison, something like A/B testing on a platform like Amazon may provide more quantitative data rather than qualitive, but this too comes with drawbacks. A threshold of views is required for meaningful results, and there is a lack of control over the process. And, at the end of it, you are left with data that is still in need of interpretation and conclusions drawn.

Neither approach is particularly fast or cheap, especially when any degree of reliability requires them to be repeated at scale.

What answers does Socialtraits look to provide?

It’s precisely these drawbacks that the Socialtraits Audience Behaviour Simulation Engine is hoping to target. By generating AI Agents, they’re confident they can provide accurate, actionable insights at speed, at scale, and at a cost lower than traditional methods.

To facilitate this, they’re offering a broad range of services their technology can accommodate.

• Custom AI community, to your audience specifications
• Image and video ranking
• Image attention heatmapping
• Focus group discussions
• Surveys
• Social media engagement strategy

All of these promise the opportunity to intensively test your designs and content before launch, providing feedback with actions to improve the likelihood of success.

What questions remain?

As with any generative AI system, there will always be questions over the dataset underlying it: how was it obtained, how has it utilised, and how has it been interpreted to allow the engine to predict future human interactions? Something based on historical data will always carry the suspicion of being backwards-facing, and therefore may be less-likely to anticipate a future shift in trends.

No system is truly unbiased, and the audiences generated by this tool will conform to the experience, knowledge and biases of the people creating it and controlling it. But this is much the same as acting on the advice of a consultant. The hope is that the expertise of the creators, backed up by robust data, will help you make effective decisions in a format that is more convenient and cost effective than the alternatives.

If nothing else, an engine like this could be calibrated against your own existing testing and survey data to more closely match your anticipated audience. Being able to do this easily and repeatably may allow you fine tune your recommendations and more easily test this tool’s guidance against actual performance. You may also be able to use syndicator data or Amazon metrics to highlight areas for deeper insights.

What are your feelings about AI focus groups?

Is the prospect of large-scale synthetic audiences appealing? Or do too many questions exist about its reliability? We’d love to hear your take on this and the other big question in online retail. Drop us a line at info@digishare.eu or join in the discussion on the LinkedIn Product Detail Page Group.