Your ultimate AI research assistant – meet Gen.

Gen is an AI Research Assistant that allows you to have a conversation with your own data in a similar way to how you use ChatGPT’s technology – all within Yabble’s secure, walled garden data environment. You can prompt Gen to query your dataset to provide an insight. Gen can even suggest actions if you ask it to.

Gen uses a natural language model which means you can ask your data all types of questions and Gen will be able to answer your question instantly and conversationally. While ChatGPT is asking questions of the internet up to 2021, Gen allows you to ask questions on your own proprietary dataset – whatever topic that might be. Once your dataset is uploaded into the Yabble platform you can prompt Gen to query your dataset and provide high level insights or dive deeper to explore your data holistically – all with data security peace of mind.

Simply type a question related to your own dataset, and Gen will do all the work for you.

You can specify which particular part of your project your question relates to, or for more general questions, ask them across ‘All Questions’. 

As Gen is a natural language model, it will understand what you are trying to achieve even if you do make the odd spelling mistake. However, Gen will not like characters such as ‘[' or '>' as these are not generally used in conversational text.

Where you have a large dataset, you are likely to get a deeper insights by selecting a specific Question. Selecting ‘All Questions' will factor in more than may be required.

Further, being specific always helps. Rather than asking ‘What are the main complaints people had?’, ask something like ‘What are the main comments people made with respect to pricing?’. By saying the word ‘pricing’ in the question, Gen is able to apply a hybrid search that picks up the word pricing, and everything related, allowing for a better response. 

As a user, I want to know about what’s really driving peoples opinion about the prices at Walmart form their responses to a survey question.

OK: Why?

Good: Why did people say that?

Better: Why did people say that about pricing?

Best: Why did people have those concerns about pricing at Walmart?

Gen has its own context builder and should be able to handle most questions it receives. However, it is helpful for a user to be as specific as possible if they want to dial in to the insights they want from their data.

There are multiple ways to approach getting actionable insight from Gen. One of the most effective ways is to structure your questions in a way that leads Gen toward the outcome you want. This is done by building context. Say you want to know what you could do about some customer feedback, you might input as follows:

• Input 1: Tell me about the feedback related to online shopping.
• Input 2: Why did customers say that?
• Input 3: What could we do about it to help our customers?

This method is effective in that Gen will not only have the context of the survey/interview question, and the responses behind it but now also some conversation history it can use as context.

Alternatively, you could also ask:

• Input 1: With regards to negative customer feedback on online shopping, I want to understand why customers gave those responses, and what we could do about it to improve the customer experience?

Both are valid methods and will arrive at a similar output. However, the first method allows Gen more capacity to respond and breaks up the question set. The user may even learn something about their data that they weren’t expecting along the way.

Yes, Gen can chat with any type of qualitative data you have in the Yabble platform.

While Gen can talk to data that has been theme counted, at present it does not yet factor in the results of the theme counting. 

Gen is a language model, and for the time being doesn’t like counting. Gen can however give you an answer like ‘most’ or ‘the main topics people discussed'.

Asking Gen an identical / similar question in the same chat will give different responses. Often the subsequent responses can add more detail, as Gen will not want to repeat itself. 

The more detail you give, the more information Gen has to work with. However, you can simply ask a question like ‘Why?’ and Gen will go and find context from your previous conversation. 

Gen will find responses related to your question in order to give an insightful answer. It may not need your entire dataset before it can give an answer like ‘most people preferred…’

Further, survey data can contain irrelevant answers. Gen will filter these answers out, or ignore them in its response generation

For now, Gen only likes data that is in English. Please watch this space while Gen becomes multi-lingual!

Gen is designed to be a Market Research AI Agent

This means that as a request comes in, Gen will review the request and then:

• Allocate the task to one of a series of MRX agents Yabble has built to complete the request.
• This allows Gen to do lots of different tasks e.g. name the top 3 mentions, summarize, work with short form as well as long for data etc
• However, if a request comes in that Gen deems to not be Market Research related or related to the project, but straight fact, it will bi-pass the agents and the project data and ask the LLM directly - that is why it can do things like tell you who Joan of Arc is.
• The problem is that LLMs by their generative nature have an element of randomness, and sometimes they won’t answer the question. This unfortunately isn’t something we have a lot of control over.
• The probability of this happening is also likely to increase with the number of questions we have asked Gen in a single chat, as the history and context becomes more complex.  

As more of our customers use Gen and we collate feedback we see different use cases they would like Gen to help them with. Every use case (eg finding direct quotes) is a different ‘tool' we need to build - which Gen's agent framework can then call on to answer. For example, pulling 'Direct quotes’ is not a tool we currently have but obviously one we would like to look at in a future state.

In short Gen and Count are totally different processes and were designed for different purposes.

An Explore shows ‘counted’ responses for a related Question and is intended more as a statistical analysis of the data, while Gen has all information in a project at its disposable to review in order to answer the users question (and this could also include ‘all responses, for all questions’).

When Gen talks about, for example the ‘top 3 things about this topic’, it is a much more holistic answer based on all the information it has available.

Gen will aim to find responses/comments related to your question in order to give an insightful answer. It may not need your entire dataset before it can give an answer like ‘most people preferred xyz’. Comments (e.g. in survey data) can also contain irrelevant answers, so Gen may filter these answers out, or ignore them in its response generation. This is the main reason why Gen won’t have ‘referenced' 10,000 comments in its answer like you may see on the Explore.