From Vision
To Validation

Our system uses powerful proprietary algorithms to extract critical attributes from your images, such as colors, objects, text, style, mood, lighting, and scenes, providing richer data for analysis. 

The personas then react to these concepts based on their own attributes, using proprietary models to give you an indication of how each persona might respond to each concept. 

Currently, you can upload 10 image concepts per chat. 

If you have more than 10 concepts to test, you can open a new chat with your personas and upload 10 new images to that chat. The personas will not be able to compare concepts between chats, but they will still be able to react to and rate your concepts so you can build out your analysis. 

You can upload up to 10 images to a chat, and these images will be remembered within that chat. If you start a new chat, you will need to upload new images, and the personas will not have context of the images in other chats.

JPEG, PNG, and WebP are all supported.

Sometimes, an image may fail at the processing/upload stage. There are multiple statuses that you can see to know that your images have uploaded successfully: 

1. Processing (loading icon running) – Upload is being processed
2. Failed – The upload has failed. You will be given the option to ‘retry’ or ‘delete’ (On delete , you will need to upload the file again)
3. Complete (clean image, no loading icon) – Your image is ready to chat with. 

If your image/s have failed, you can click the arrow shaped ‘retry’ button to reprocess the image, or you can delete the image and upload a new version. 

Most users receive initial feedback within seconds of submitting their images.

Your uploaded images are stored securely with strict access controls and are never shared outside the platform. Yabble operates as a walled-garden, so everything processed and generated within your Yabble account is used only in your account, and is not used to train any of our own models, or shared with any of our third party providers. 

• Packaging designs: Explore how layout, color, and claims resonate across audience segments before you commit to production.
• Campaign creative: Get early feedback on static ads, hero images, or digital banners to fine-tune your message and look.
• Product concepts: Test different visual approaches for innovation pipelines — from early-stage sketches to final mockups.
• Social & ad mockups: Discover which visuals capture attention and drive intent for your products or services.
• Marketing messaging: Get early feedback on marketing messaging for your upcoming campaigns. 

Vision for Virtual Audiences isn't a replacement for traditional concept testing, but it adds a powerful, data-backed layer to your early-stage strategy. Whether you're exploring product ideas, packaging, or creative campaigns, AI personas built from public and proprietary data sources provide immediate, attribute-driven feedback. That means you can move beyond gut instinct and start validating your ideas with real, probabilistic insight – before investing in full-scale testing. 

The Yabble Virtual Audiences tool is the ultimate end to end AI solution for research.  Making insights in minutes a reality.  

• Specify your topic of interest
• Load from saved persona library, or build new personas and specify your desired audience using demographic and behavioral characteristics or simply approve Yabble's suggestion based on your topic.

• Review, amend or approve the questions built on Yabble’s decades of research expertise.
• Using augmented data, Yabble’s Virtual Audiences tool provides instant answers to your questions using AI-generated data.  
• Interrogate the data for further analysis in Yabble to explore the full power of AI for insights. 

Yabble’s proprietary augmented data model uses the combined knowledge of Large Language Models (LLMs), relevant and recent trend data, social data, and behavioral statistics to create a unique AI-generated audience to answer your research questions. This unique approach via the Yabble platform provides increased relevancy, accuracy, and recency of data above what a large language model alone can provide. 

In each virtual audience project, you’ll have access to: 

• Up to 8 AI-generated Personas
• Up to 20 survey questions (the AI generates 10 for you to start with)
• Theme counting of all questions with theme, sub-theme, and sentiment output with Yabble Count
• Unlimited persona chat
• N=50 individual virtual survey responses per question
• An Overall Summary detailing your key insights from your topic
• Persona Summary providing a detailed breakdown of insights by each persona
• A per-question summary providing key insights across each survey question
• A PDF export report that includes all summaries above, as well as market trends and leading indicators (including sources) related to your topic 

Yes, Yabble's augmented data tools have a very broad range of knowledge and should be reliable for most topics and markets. If you have a highly specialized topic and need a different research approach, contact the Yabble team via hello@yabble.com 

The number of topics you can research is unlimited (2500 credits per Virtual Audiences project). For each topic, there is a limit of 250 completed responses within the Virtual Audiences environment. This gives a good indication of what a market is thinking or feeling on any topic. 

In the Yabble platform, you can: 

• Run additional analysis on your data  
• Ask your data questions
• Download and visualize your insights
   

 

Don’t just take our word for it – we’ve done the data-driven diligence when it comes to validating our augmented data model. Read more about our Virtual Audiences validation work here.

Yabble operates a 'walled garden' approach to data security. When you use the Yabble platform you are covered by Yabble's security and privacy settings which include ensuring that your data does not pass to OpenAI to train their models.

Virtual Audiences has lots of different use cases. It's great for kicking off a project when you want to gather a broad understanding of who the customers are in a market and their behaviors. It's also great at looking at market trends, understanding changing behaviors, tastes, and perceptions, and working out what this means for your future products or communications strategies, or even looking at competitive sets and what the market sees as strengths or weaknesses of other players in your industry.

If you're looking at AI-powered insights solutions, you may have heard the terms 'augmented data' and 'synthetic data' used interchangeably. There are some key reasons why Yabble's proprietary augmented data model is superior when it comes to AI-generated data for insights.

Synthetic data is like creating a digital twin of real-world data. Think of it as a simulated dataset, created to resemble the real thing without using any real details.

Augmented data is like giving your data a boost. It's taking the original data and enhancing it to make it more complete and more useful. Think of it as adding extra details or features to existing data to get a clearer picture. This is why we choose augmented data for market research – it is both grounded in reality and optimized for the most profound insights.

Ultimately, anything built on an LLM will have an inherent bias based on the corpus (training data) and how it was trained, specifically with RLHF. However, we don't solely rely on LLMs to create personas, we search proprietary, public, social...etc... data to find the relevant demographic information for people who would be interested in the topic entered into Virtual Audiences. We use this to generate demographic criteria for personas, and then through subsequent steps we begin to fill those criteria.

This does not mean there is no bias - there will always be bias, as there is likely to be with traditional research, but we mitigate the detrimental aspects of bias to the best of our capabilities, as you do with traditional research.

Yabble's Augmented Data model revolutionizes the process of insight generation by integrating diverse data sources, including proprietary datasets, academic content, and real-time web searches, with advanced machine learning algorithms. This method transcends traditional surveys and human feedback, utilizing generative AI to synthesize data into comprehensive knowledge lakes. In this model, data amalgamation serves to create detailed personas, formulate relevant questions and answers, and distill key insights, offering users a nuanced, multi-dimensional view of their topics of interest and delivering precise, AI-driven insights.

Our current model uses open-ended questions to provide you with rich insights into your topic or business question. This model has gone through extensive testing to ensure insights provided are deep and accurate. All open-text answers can be quantified through the Yabble Count tools. This will be done automatically, and you will be alerted when the data is ready for review. Over time, we will introduce new question types once we are confident in the validity of the data from these new question types.

Yes! You can run simple ideation concept tests through Yabble, both through our new Vision feature (image concepts) and using text descriptions of your ideas.

In cases where you can see duplicates of the same parent URL in your source list, this is because Yabble's augmented data model has visited multiple pages on the same site to get context for your project ie. http://www.transfertravel.com/page1 and http://www.transfertravel.com/page2 etc. We do not currently display the full URLs for sources.

To get the most out of the proprietary data feature of Virtual Audiences, types of data best suited to the tool:

• Internal market research data or reports e.g. survey data, reports, interview transcripts etc
• CX Studies
• UX / Design Studies
• Support tickets
• Customer chat logs
• Internal knowledge base or FAQs.
• Product description or documentation
• Academic articles
• Market reports
• Etc i.e. Any type of text data not available publicly on the internet that can better inform your specific project.

Where possible, provide data in string format i.e. data labels rather than numeric codes.

Yes! You can upload up to 300 files (max file size 2MB each) to one Virtual Audiences project, and they do not need to be the same file type. All proprietary data is summarized when it is uploaded – we recommend uploading less than 10 files to get the best output from your proprietary data. 

There are several different file types that the Yabble proprietary data feature accepts, including .PDF, .CSV, .TXT, and .DOCX (max file size 2MB each). You can upload .XLSX files, though the AI will only process the first sheet in the file, and will not process oddly structured Excel files. 

Virtual Audiences projects can take as little as 15 minutes to complete from topic input to insights output – but this will vary based on the quantity and size of the proprietary data you upload to the project i.e. the more data you provide, the longer the process will take for the AI to read and process the data.

When creating the Knowledge Lake for your Virtual Audience the AI will read across all data provided.

The extent to which it ‘relies’ on your proprietary data to create its personas, synthetic respondents and answers will vary depending on your data volume, content, completeness and the relevance of your data to the questions you have have asked. As well as the recency of your data.

As all data is multi dimensional, some data is more valuable to the process then other pieces, and this is decided by the AI on a case by case basis depending on the research or business question you have asked. Also the ability for the data to be cross referenced / validated between sources.

Yabble will always create the Knowledge Lake using a combination of sources, but the utilization rate of each source will vary dynamically depending on the question being asked.

Yabble’s AI also has a recency bias inherent – so, as an example, if the proprietary data you upload is from 3 years ago, the AI may find more recent information from our other sources.

If you have any specific questions about Virtual Audiences and how this works with your own data to provide insights, contact the team at support@yabble.com.

Yabble’s AI will read across all data provided. It is assumed that you will provide data that is related to the topic, however, if there is some unrelated data, the AI will be able to ‘clean and ignore' this data to an extent.

As the product is in beta, edge cases where data is entirely unrelated may provide poor results.

If you have any specific questions about Virtual Audiences and how this works with your own data to provide insights, contact the team at support@yabble.com.

No, not at this stage.

No, not at this stage.

If your data doesn’t include any demographic data, the Yabble AI will recommend and generate target demographics, based on data it has gathered from other sources. This can always be regenerated later by you before you complete the project, if you would like to.

Note: The proprietary data will still be used to form the persona’s personality characteristics and behavior even if demographics are absent.

There will be demographic-specific insights and general insights in the proprietary data. Insights specifically relevant to the non-target audiences might be disregarded when generating insights. But, all the relevant information will still be used.

Yes! You can now save groups of personas to your persona library and use them across your projects. Read more about this feature here.

You can continue to chat with the personas within your topics as long as you have access to the Yabble platform. You cannot add new data to an existing project. You can save your personas to your persona library and start a new Virtual Audiences topic with those personas to ask your new survey questions and add additional proprietary data.

At this stage, the AI cannot process images embedded in your proprietary data. It can pull some context from text in tables, but it’s not seamless. At this stage, text data is best for all data qualitative or quantitative i.e. labelled data rather than numeric codes.

No, not at this stage. You can however download the data from the platform, then upload it in the VA process.

No, that would need to be uploaded separately via the normal ingestion process in Yabble.

Currently, the proprietary data is used as a layer in Knowledge Lake and blended with other data Yabble collects. At this stage you will not be able to ask the AI-generated personas about specific data you have uploaded.

Currently Yabble can read closed ended data that is in string format i.e. data labels rather than numeric. It will look at the question for context i.e. The question must be clear for the model to understand the context of ‘Very satisfied’ for example.

If you have any specific questions about Virtual Audiences and how this works with your own data to provide insights, contact the team at support@yabble.com.