GenAI Gaining Traction in Customer Feedback
Generative artificial intelligence will eventually change the way organizations collect, analyze, and act on customer insights, Forrester Research says in a new report.
But for now, only one-third of customer experience professionals are leveraging genAI for voice-of-the-customer and CX measurement programs.
As they change those numbers, CX leaders need to avoid shiny object syndrome and carefully balance risks and rewards when considering genAI use cases for customer feedback management, the research firm concludes.
Forrester advises CX and customer insights professionals to take advantage of the AI push to intensify their investments in predictive and other “traditional” AI capabilities.
Colleen Fazio, Forrester senior analyst and lead author of the report, says genAI will continue to be a technology for internal use for employee efficiency, summarization, and analysis.
She cautions that many consumers are still very distrustful of AI, even for simple responses to simple questions. Forrester’s research found that 78 percent of U.S. online adults who have heard of genAI agree that companies should disclose when they are using genAI during customer interactions, and only 30 percent say they would trust information provided by genAI. However some companies have started to use the technology for some limited customer-facing uses, such as closed-loop emails, according to Fazio. “They’re using AI drafts for customer feedback, with a human in the loop.”
Wider customer-facing use is still some time away, especially for regulated industries, Fazio adds. “You have to think about the value trade-off. You’re balancing using AI, with a human in the loop who’s potentially error-prone, with the risk/reward of using the technology. If you have a ton of scale around your close-the-loop program, you have automation, but you may not be eliminating a problem.”
Another drawback, according to Fazio, is that too much AI eliminates some of the human contact, which can be very valuable from a CX perspective.
She recommends that any use of AI with customers be tested on a limited basis before a complete roll-out.
The report also cautions that genAI is not a shortcut around other analytics. Non-generative-AI analytics techniques such as prescriptive and predictive analysis are more reliable for understanding revenue per customer, how to improve customer satisfaction, etc. GenAI is nondeterministic, which means a user would probably get a different answer if asking the same question twice.
It’s not just genAI that is underused; companies don’t use other analytic techniques as much as they should, Fazio adds. “With the silos that persist at these companies and the fragmented data architecture, it’s challenging to bring these datasets together.”
Forrester recommends establishing guardrails like prompt engineering and explainability to mitigate the risks of unpredictability.
The technology is still underused for analytics.
According to the research firm, slightly more than one-quarter (28 percent) of organizations currently leverage both generative and predictive AI techniques.
For improved analytics, Forrester recommends the following:
Hybrid AI for text mining.This is still the best approach for the enterprise level, according to Forrester, pointing to the use of linguistic rules and domain ontologies that provide unmatched accuracy, while summarization and Q&A interfaces on data are machine-learning-based AI’s sweet spot. “Vendors with hybrid AI solutions offer highly accurate, industry-specific models that many organizations will find too costly and time-consuming to replicate and build at scale with genAI,” the report says.
Predictive and prescriptive AI for recommendations.Currently, genAI can’t predict which customers are likely to churn, given that large language models are still not good at advanced math. So Forrester recommends traditional predictive and prescriptive models instead. The research firm expects advanced genAI users to adopt a hybrid approach with predictive AI measuring customer behavior and making recommendations and then validating those results with genAI topic extraction from customer feedback.
Structured databases for complex analytics. Relational, graph, and similar databases are a better fit and more cost-efficient for structured data storage, compression, security, and analysis than genAI, Forrester explains. “When analyzing time series data, such as trends in topics or sentiment over time, and conducting longitudinal analytics to identify factors contributing to a decline in sentiment within a specific timeframe, a relational database or an online analytical processing database remains indispensable.