We know consumers want generative AI, however they also have some valid concerns. According to The 2023 Zendesk CX Trends Report, more than a third are concerned about people and companies using AI in malicious or negative ways, and nearly half said they worry that bots and automated systems may show bias against certain groups of people. Business leaders share a similar concern: 73% of them said that ensuring AI and bots are not biased against specific types of customers is an important priority.

As our head of AI, Cristina Fonseca, shared recently when it comes to the promises and pitfalls of generative AI, “We believe one of the biggest challenges in adopting AI solutions is to build trust in the technology.” Our commitment to providing businesses with the most secure, trusted products and solutions possible is embedded in our DNA. As part of this, Zendesk established and leverages a set of design principles that not only set the standard for how we design, develop and build everything we do, but set a clear foundation for our use of generative AI for customer experiences (CX).

1. Privacy, security and compliance by design
It is imperative we keep our customers’ data secure, and this responsibility is of the utmost importance to us. We do not compromise on technology that does not meet our security and compliance standards, and we put in place checks and balances to evaluate our usage of data collected by our technology to ensure it is used responsibly. We believe AI should be used, where appropriate and necessary, to enhance the existing protections we put in place for our customers and their end customers. Examples of this when it comes to generative AI include: ensuring the training data that is used is anonymized; restricting the use of live chat data; respecting data locality; providing opt-outs for customers; and reducing the risk of bias by having a diverse set of developers working on the project and engaging users and other subject matter experts to identify potential areas for bias.

2. Transparency builds trust
We believe in being upfront about how our technology works, what data it collects, and how that data is used. We also provide industry leading opt-out processes for both end customers as well as our direct customers. When it comes to AI in customer service, our communications are direct, clear, and honest with our customers about how it’s used in our products and solutions. Not all AI predictions are the same in regards to quality. Sometimes models have a high confidence at doing a particular task, while in other scenarios the confidence is low and AI is simply trying its best (probably doing a new or not very frequent job). By making the quality of AI predictions available to our customers they can take that into account and act accordingly, while building trust in the technology.

3. Accuracy thresholds set the standard
With every early access program and general availability of each product, our teams must meet certain accuracy thresholds before we release them to our customers. When it comes to AI, the accuracy that must be met – be it high, medium or low – largely depends on the task or model being tested. At Zendesk, we only make AI available to customers if the model passes a quality threshold that is of the highest industry standard. Every AI prediction we make has an associated confidence score that is passed to the administrator or agent, so any workflow built on top of AI can take that into account. For example, high confidence predictions can lead to fully automated tasks, whereas those with less confident scores mean we recommend that a human is kept in the loop for a better experience for both the agent and customer. At its core, our technology is designed to provide accurate and reliable responses to customers’ needs, and we ensure that is our primary focus always.

4. Work with the right partners
Partner selection is a crucial aspect of our approach to product development. We work with different technology partners in order to use the best tools to continually create value for our customers. We recognize that generative AI is a tool with no one size fits all solution, and therefore an ecosystem of partners that helps us build the best products and solutions is what is best for our customers. That also allows us to work closely with them to overcome some of the current challenges and limitations.

5. Technology is created by people for people
Our technology is designed to enhance the customer experience, not replace human interaction. When it comes to applying generative AI to the customer experience, human oversight is particularly imperative. While our technology can help automate certain tasks, we know that a human connection is still crucial in providing exceptional customer service. As we stated recently, 81% of consumers told us having access to a human agent is critical to maintaining their trust with a business when they have trouble with AI-enabled customer service. Furthermore, we always ensure that there is human oversight so that the technology is being used appropriately and customers are receiving the level of service they expect.