One area that Large Language Models can advance research and understanding is in text analytics. There has been much work in the field prior to the advent of Generative AI solutions though, and the lessons learned over the years in those efforts can help us avoid pitfalls when applying new tools to get the most value from them.
At Ipsos, our Customer Experience (CX) team has been leveraging text analytics for over 15 years, and with Fiona Moss, the global head of CX analytics, we have summarized 5 areas from the past that are still relevant today. Just like leveraging the Truth, Beauty, and Justice framework that Charles Taber and I originally applied to evaluate an earlier era of AI, 25 years ago, these 5 items are still relevant with Generative AI today.
1-Demand transparency- transparency in the models and data as well as contracts, governance and infrastructure.
2-Don’t forget the data- data representativeness, relevance, and sufficient detail for your question is key to deliver against your objectives.
3-Formal evaluation still matters- the quality of specific use cases must be systematically evaluated and leverage the same level of rigor used with NLP, ML, and other analytics.
4-Remember to manage expectations- manage end-users’ expectations about commentary and check what is provided by LLMs.
5-Establish a reporting/usage mechanism that meets business needs- picking up where text analytics already is – with existing, configurable interfaces and models for live interactions that support the right functionalities for the right users.
To download our full paper "Not Doomed to Repeat: Lessons from CX Text Analytics Applied to Generative AI" that expands on these topics as you apply the latest AI solutions to your organization's challenges, you can follow the link- https://lnkd.in/eMqUE2Ft.
#ipsos #IPSOSHiAi #GenerativeAI #LLMs #customerexperience #textanalytics #TruthBeautyJustice
Startup Executive | Product and Innovation Expert
2moThanks for sharing. Great example of creating a collaborative AI/Human UX!