Editor’s note: The following is a guest article from Melissa Fletcher, senior principal researcher in Gartner’s customer service and support practice. Emily Potosky, Kim Hedlin, Olivia Foster and Aman Jagdev contributed to this article.
Generative AI-powered chatbots and interactive voice response systems are often promoted as revolutionary self-service solutions that could reduce the need for frontline customer service agents.
However, implementing customer-facing generative AI poses risks — from the potential degradation of customer experience and lack of human connection to AI inaccuracies and unexpected costs.
Nonetheless, such challenges are manageable, and investments in the technology remain significant and worthwhile. Effective generative AI solutions can create seamless customer experiences, and careful content management can reduce errors. With strategic planning and implementation, leaders can effectively leverage generative AI's potential.
Most customer service and support leaders are using generative AI to enhance service capabilities, primarily as a tool to augment human agents and streamline internal processes.
To date, its application in customer-facing scenarios has been less prevalent. But this may soon change.
Service leaders — who have long recognized the transformative potential in this rapidly developing technology — are beginning to seriously consider its direct application in self-service.
The potential for enhanced self-service and cost savings through generative AI-driven applications is accompanied by significant risks, however.
Poor implementation could have a negative impact on customer experience, at best driving them back toward assisted service and at worst driving them away entirely. The risk of disseminating inaccurate information to customers could result in dissatisfied customers and legal liabilities, while the potential for unforeseen or escalating costs could undermine anticipated savings.
Risk 1: customer rejection of generative AI in self-service
A lot of customers prefer interacting with human agents, but results of Gartner survey research indicate that this preference doesn’t matter when it comes to customer loyalty. What really matters is the ease of resolution.
Consider two individuals with similar experiences and preferences who resolved their issues and preferred assisted-service interactions. One customer interacted with a human agent during their journey, while the other did not. The probability of disloyalty between these two individuals is practically identical, with no statistically significant difference.
Our analysis shows that, regardless of contact with assisted service, higher-effort experiences correlate with a higher probability of disloyalty, whereas lower-effort experiences correlate with a lower probability.
The lesson for leaders: When customers experience low-effort resolution after engaging with generative AI, they are likely to engage with it further.
Risk 2: inaccurate information reaching customers
It is not uncommon for vendors to imply that their generative AI solution can accommodate even the least organized or optimized knowledge bases. Some leaders may even be led to believe that a knowledge base isn’t needed at all, and that pointing generative AI toward the enterprise website is sufficient.
But the reality is that generative AI is only as good as the data it relies on. While GenAI can sift through unstructured enterprise content, it cannot address insufficient knowledge management on its own and cannot reliably select only the most valuable information.
If it accesses incomplete, outdated or low-quality content, this may still be reflected in its responses, even with safeguards like retrieval-augmented generation in place.
To minimize the risk of inaccurate information reaching customers, organizations must enhance their knowledge base in two ways: ensuring that their conversational generative AI solutions have access to content that is comprehensive, accurate and up to date, and optimizing content specifically for AI consumption, rather than for human users.
Risk 3: generative AI run costs
The operational costs associated with generative AI can be challenging to predict; without appropriate controls, these solutions may end up costing more than the savings they generate.
To safeguard against unexpected or escalating expenses, leaders must gain a thorough understanding of vendor pricing structures and identify ways to influence these costs. Leaders can implement measures to control expenses by carefully designing the customer journey with AI usage costs in mind, ensuring that financial efficiency is maintained throughout the implementation and operation of generative AI solutions.
Many generative AI solutions are priced based on consumption, meaning that increased usage leads to higher costs. When considering the variability and volatility of GenAI expenses, leaders must take three major factors into account:
- Number of users: While leaders may estimate usage based on past customer interactions with chatbots or IVR systems, effective GenAI solutions that resolve issues efficiently may lead to increased engagement.
- Amount of input and output: Predicting usage requires considering the length and frequency of exchanges between users and the AI. As customers interact with generative AI in a more conversational manner, these interactions may be longer than those with previous chatbot or IVR iterations, potentially inflating costs.
- Cost structure itself: The cost per unit can vary, with usage often denominated in tokens. These token prices are set by vendors and can change unless negotiated otherwise, adding another layer of complexity to cost management.
The components of successful generative AI implementation
While the benefits of customer-facing generative AI have not fully met expectations, they remain significant and worthwhile.
Successful implementation of customer-facing, conversational AI solutions can allow leaders to serve more customers more effectively while also conserving resources.
To mitigate the unique risks associated with generative AI:
- Provide low-effort resolution with generative AI: Meet customers where they are when deciding where to place your generative AI solution in the customer journey. Additionally, avoid creating cycles that trap customers in self-service channels, and instead use generative AI to create seamless handoffs to live agents when they are needed to achieve resolution.
- Ensure your knowledge management is curated and optimized for AI: Focus on creating a meticulously curated, application-specific knowledge repository and adapt content formatting to suit AI retrieval methods rather than human consumers.
- Mitigate risks of ballooning costs linked to generative AI’s variable costs: If able, use a flexible tech stack to maintain negotiating leverage with vendors in a volatile pricing environment. Additionally, institute cost-mitigation strategies such as limiting which users have access to the generative AI solution or capping the length of interactions.
