“Zendesk’s Head of AI on the promise and pitfalls of generative AI”- Cristina Fonseca, the Head of AI at Zendesk
Here’s the interview snippet from the interaction with Cristina Fonseca, the Head of AI at Zendesk.
Saumya: What excites you about technology like ChatGPT?
Cristina: ChatGPT is based on an LLM which is a machine learning model that can recognize, predict, and generate an output based upon training from very large text-based data sets. LLMs can even be used to generate images, or even music, based upon a word input.
ChatGPT has brought LLMs mainstream and made them readily available for everyone.
Their ability to analyze large amounts of data and become smarter over time can accelerate the development of AI capabilities as well as AI adoption. The more accessible, reliable and helpful AI can be, the faster businesses will see its value and choose to implement.
LLMs are not new, but what’s really innovative around ChatGPT is the conversational, chatbot-like user interface that makes AI easy to understand and use. I believe AI should be accessible to companies of all sizes which is aligned with the Zendesk AI mission, and part of what we are working on for customers currently.
Saumya: Can you tell me more about the potential for ChatGPT and Zendesk?
Cristina: Expectations are at an all time high, and customer experience, more than any other area, benefits from data and a deep understanding of what the customer wants. We know AI will be adopted broadly and play a major role in all CX interactions.
The biggest opportunity lies in using AI to eliminate much of the manual workload that can be low value and incredibly time consuming. Imagine an agent going from having to read through pages of text to get a summary of a customer’s previous issues, to getting an accurate, customized summary allowing them to solve customer issues much more quickly. That is the type of scenario LLMs enable and a Zendesk and ChatGPT partnership will provide to businesses.
Zendesk has been creating an AI Suite built specifically for CX that leverages the high amounts of CX-specific data we have. For example, our intent detection and sentiment analysis models are proprietary and custom to the CX industry. We are also partnering with LLM providers like ChatGPT to extend our core offering, creating highly personalized and valuable customer conversations. Businesses will be able to start using our solutions in minutes, without costly training or expensive integrations. This increases time to value and ensures a positive return on investment in AI.
Zendesk is the only company taking this unique approach and believes integrations like these paired with home grown AI solutions will revolutionize CX and provide companies with faster, relevant data and insights to create custom solutions.
Saumya: How are Zendesk customers already benefiting from AI and LLMs?
Cristina: Zendesk has been working for years to apply machine learning to provide excellent customer experiences at a staggering scale to better understand patterns in how customers request support. We have one of the biggest CX datasets and are leveraging it to train our own AI capabilities including intent detection and sentiment analysis, that we use to better understand customers.
In addition to our AI-powered self-service tools, many of our customers are already driving business value using features like intent detection and sentiment analysis. For example, when a customer submits an inquiry about a broken product, Zendesk can detect the customer’s specific intent to make a return which then gets instantly routed to the correct customer service representative.
All of this is done by Zendesk, behind the scenes, without the customer having to touch, train, or even really understand the model exists. We’ve learned a lot from it – and are applying it to other portions of the customer experience that might benefit from having bespoke CX models – as opposed to wide ranging LLMs.
Saumya: What advice do you have for companies looking to get AI and LLMs for customer service right?
Cristina: Businesses want and deserve to see immediate results when it comes to manual, repetitive work performed today, but it can’t be done with generic models only. AI can help CX teams be more consistent, better understand customers and derive insights from data. If we appropriately apply technology with the right level of CX strategy, we can find unique opportunities to boost agent productivity and to completely automate customer queries without degrading the customer experience. Additionally, AI insights can help businesses identify knowledge gaps and pinpoint problems before they become large volume issues.
While the benefits are clear, there are some things to keep in mind as companies look to implement LLMs:
It’s important that AI can guide human agents on what is true versus false and understand nuances of individual businesses. We design our models to be grounded in facts and to learn from specific company data for improved performance.
AI should be available with one click but still give the control to customize and adjust to specific business needs.
Know that AI makes a lot of mistakes. ChatGPT has to be fed the right answer, content, or knowledge if it’s going to be expected to get the customer to the appropriate resolution. Especially as you are first training a model, keep humans in the loop so that you are providing the best CX.
Don’t let AI run wild. Understand the confidence of an answer before you provide it so you can supervise and improve solutions with a human feedback loop.
Don’t be afraid to narrow the scope of answers and escalate as needed. Use a data-driven approach to restrict AI to only engage in topics relevant your business selects to reduce the margin of error.
Don’t forget the humans. As beneficial as AI is, the human touch shouldn’t be replaced. Escalating to agents can help mitigate quickly if something goes off course and provide customers with the best solutions.
Saumya: It seems every company is releasing a new ChatGPT feature or offering. How should a company decide what is right for them? What steps can be taken so that the customer experience won’t be impacted when implementing?
Cristina: ChatGPT is powerful, but we’re seeing companies so caught up in the hype that they’re often not being thoughtful in its application. Simply implementing a ChatGPT API and promoting to customers can come at a huge cost. Moving quickly is important, but without the right steps to safeguard and test, it can cause a lot of pain to their customers. We believe one of the biggest challenges in adopting AI solutions is to build trust in the technology, so making sure AI is properly enabled is key.
It’s critical for brands to provide accurate responses and know when to escalate to an agent for more nuanced 1:1 conversations. ChatGPT is great at finding general responses that are highly conversational, but it doesn’t have the context to answer questions about your business – this is a problem, especially for customers looking for correct information. Companies must have a mixture of deep CX specific technology and human agents who remain involved to supervise the AI and mitigate risks. Providing a good customer experience and maintaining customers’ trust is a top priority – businesses cannot lose sight of this while adopting new technology.
The burden should be on providers, not customers. At Zendesk, we want to get it right, keeping a human in the loop until the AI gets accurate enough to not ruin the customer experience we partner with our customers to achieve.
Saumya: How will Zendesk utilize ChatGPT moving forward?
Cristina: Zendesk offers AI solutions that are pre-trained and can be customized for unique customer use cases. Our solutions uniquely combine the power of our entire support suite, LLMs trained against conversational service tickets and intents, and soon the power of OpenAI and ChatGPT. Combinations like these allow companies to focus less on the technical aspects of the rapidly changing AI spaces, and instead focus deeply on the customer experience.
We commit to being a partner in helping our customers deploy AI systems that are built on trust, security and reliability, so we are carefully considering partnerships based on our core values.