Information Quality: The Lifeblood of AI-Powered Insights

Introducing Aware’s Information Quality model, enabling accurate, actionable Generative AI experiences across your business 

The last couple of weeks in AI have been huge. Google rolled out a large expansion of their Bard chatbot. Meta announced several new AI-powered experiences. Adobe introduced Firefly, designed to facilitate AI-powered image creation.  

And now, we release our latest purpose-built behavioral machine learning (ML) model called Information Quality (InfoQ). Our Information Quality Models deliver enterprises more actionable and accurate AI-Powered insights by determining the information quality, or meaningfulness, of workplace conversations anywhere from Slack, Teams, and Workplace to Qualtrics, ServiceNow, and Outlook. 

This fast-growing data set from survey, collaboration and social media platforms contains valuable insights from every part of the business. About 30% of these digital workplace conversations include high-quality data with the context and substance to power generative AI capabilities and drive strategic decisions across your business. 

Filtering out the noise

However, collaboration data is also messy and contains low quality conversation that lacks sufficient context and substance. According to Aware’s research, 25% of collaboration data is now considered low quality, consisting of: 

• Spam messages 
• Gibberish 
• Messages from bots 
• One-word responses 
• Stand-alone gifs or images

Relying on this low-quality data to drive business-critical decisions or power generative AI capabilities results in false positives, overly broad themes and—most importantly—less relevant and actionable insights. By isolating and removing that noise, info quality improves and so does the quality of the insights, and Generative AI summaries.  

The InfoQ model solves this problem by evaluating the quality of each conversation, assigning an information quality score, and then excluding or filtering low-quality data from future results. By isolating and removing that noise, overall information quality improves, and with it, so does the value of the insights, trending topics, themes, and Generative AI summaries that drive business decisions. 

Purpose-built models designed to power the enterprise workflow

This ML model is, more accurately, a Recurrent Neural Network (RNN) model and is part of the Aware library of proprietary models. Aware’s models include RNN, convolutional neural network and mixture density-type models, and are purpose-built to capture the nuance inherent in workplace conversations. While most AI in use at the enterprise level is trained on publicly available data, Aware's models are trained on real collaboration messages. These messages are expertly hand-labeled by professional annotations and have gone through exhaustive quality assurance checks to ensure reliability in real-world situations. The result? Our models can be deployed confidently out-of-the-box and boast near-human accuracy. 

Thanks to the combined power of Aware’s data platform, machine learning operations platform and our in-house team of data scientists, Aware is at the forefront of ML model development. With access to a pipeline of trustworthy, timely data combined with targeted training methods, Aware can quickly develop and optimize cost-effective and highly accurate models in weeks. In the past quarter alone, the Aware data scientists have refreshed our Code Detection, Sentiment, and Toxic Speech models, with the latter now boosting accuracy of 94%, higher than the human benchmark. The Aware team was also able to introduce the industry’s most accurate Spanish sentiment model, designed to bring greater insight to multilingual organizations. 

Closing the gap from analysis to action

This capability powers insights across the platform, starting with Aware’s new Generative AI Summaries. InfoQ helps to determine which messages contain enough information to contribute meaningfully and eliminates information that would dilute quality of the summary. InfoQ powers additional use cases within Aware including: 

• Filtering contextual Search & Discover results 
• Reducing false positives within Signal events 
• Excluding uninformative trending topics and themes 
• Surfacing more impactful message verbatims in Spotlight 

By accessing only truly high-quality data within Aware, organizations are better equipped to make faster decisions with greater certainty. 

Why does accuracy matter? Because of what lurks in your data. Get the report to find out more.