Applying Lessons from CX Text Analytics to Generative AI

Not Doomed to Repeat: learning from the past of Customer Experience text analytics will ensure we obtain the most value from Large Language Models

The author(s)
  • Fiona Moss Customer Experience
  • Rich Timpone Global Head of Ipsos Science Lab
Get in touch

Applying Lessons from CX Text Analytics to Generative AI | IpsosGenerative AI has rapidly democratised the power of text-based Artificial Intelligence. Essentially, anyone and everyone with access to the internet can now ask questions and get responses from these super-smart bots.

AI tools – particularly Large Language Models (LLMs) – can be leveraged for many practical text analytics’ use cases. While we are in a new landscape, learning from the past of text analytics will ensure we don’t repeat errors and can leverage these new tools to their full potential.

From a niche offering over a decade ago, text analytics has now grown to become standard in most large or ongoing Customer Experience (CX) programmes. It can be used to provide identification and quantification of key topics and sentiment across solicited (e.g. open-end questions) and unsolicited (e.g. social media) feedback.

While we are in a new landscape, learning from the past of text analytics will ensure we don’t repeat errors and can leverage these new tools to their full potential.

In this paper, our CX experts, drawing on text analytics’ learnings from the past 15 years, and using our AI framework of Truth, Beauty, and Justice, outline five key lessons that teams must keep in mind as they apply LLM-powered Generative AI tools:

  1. Demand transparency. Ensure that you are clear on the capabilities of the LLM, the nature of data used to train the model, and whether it is able to learn and adapt as it experiences new data. Transparency also means putting in place contracts, governance and infrastructure to protect the privacy and security of sensitive data and information.
  2. Don’t forget the data. If the data involved is not representative or relevant to your business question, or does not contain sufficient detail to answer that questions, then the results will not deliver against your objectives.
  3. Formal evaluation still matters. To get the most value from text analytics, the quality of specific use cases must be systematically evaluated. Holding LLMs to the same rigour as traditional text analytics ensures you will obtain the most value.
  4. Remember to manage expectations. Just as for text analytics in the past, we need to manage end-users’ expectations about commentary provided by LLMs. Outputs it provides should not go unchecked.
  5. Establish a reporting/usage mechanism that meets business needs. LLMs and Generative AI pick up where text analytics already is – with existing, configurable interfaces for live interactions. These interfaces, together with models that support the right functionalities, need to be put into the hands of the right users.

If we treat LLMs with the respect they deserve, learn from the past, and embrace the future, they will undoubtedly lead to better, more loyal, and more profitable customer relationships.

Download the white paper

For more information about Ipsos’ wider point of view, check out our latest Ipsos CX thinking.

Learn more about Voice of the Customer, Customer Experience Analytics and Customer Experience Advisory at Ipsos. To discuss how Ipsos can help you and your organisation, get in touch with your local Ipsos CX contact, or one of Ipsos’ global experts.

The author(s)
  • Fiona Moss Customer Experience
  • Rich Timpone Global Head of Ipsos Science Lab