What key metrics have become most valuable for measuring contact center success since implementing AI tools? Has this changed your definition of what "good performance" looks like? How do you evaluate the balance between automation and human interaction in your contact center? What framework do you use to determine where each should be applied?

AI has made us rethink how we judge performance in contact centers. We no longer just look at average handle time or post-call surveys. Now we measure things like how often the AI identifies the customer's intent correctly in the first message and how many interactions are fully resolved without human involvement. If a virtual agent handles a return request or billing question from start to finish without confusion, that's real value. We also look at transfer quality, how well the AI sets up the human agent. If the customer has to repeat information, that's a fail, even if the final outcome is positive. A less obvious but powerful metric is how often AI intercepts problems before they become tickets. For example, if the AI detects that a customer has had two delivery delays in a row and triggers an apology message with a discount before they even reach out, that's a success. We track these preemptive interactions because they reduce volume while improving satisfaction. That's something traditional contact center metrics never captured. When deciding what should be automated and what should stay with people, we focus on the level of cognitive load and emotional risk. If a task is emotionally neutral and follows a clear logic tree, like changing shipping addresses or updating payment info, automation works well. But when emotions are high or there's ambiguity, we hand it off to people. For example, we never automate account closures or anything involving loss, complaints, or gray areas. We do not follow a fixed formula. We A/B test flows monthly and let the data tell us where friction shows up. If drop-off spikes after a script change, we revisit the AI logic. Our team treats the AI like a junior team member that's always being trained. That's how we scale both volume and quality. Best, Arthur

Since adopting AI, we value resolution quality and customer effort more than handle time. Metrics like first-contact resolution, AI containment, and intent accuracy now shape our view of "good performance." It's less about speed, more about fit and follow-through. To balance automation and humans, we map each contact type by complexity and emotional load. AI handles predictable, low-stakes queries; people take what needs judgement or empathy.

On one occasion, I had five airport pickups booked—and thanks to an AI tool, three out of four were rescued from potential catastrophe in less than 60 seconds. At Mexico-City-Private-Driver.com, we do not have a conventional call centre. However, we do have a real-time dispatch and support line injecting many call centre features. When we added an AI layer for WhatsApp routing and follow-up booking confirmation, the very first metric we saw change was "Response Delay"—which was down from 2m 14s to less than 30 seconds for 82% of inquiries. That one change saved two misfortunate arrivals from occurring and a high-value customer who almost cancelled due to a delayed pickup clarification. So, my measure of "good performance" indeed changed—from measuring how fast we replied through to how reliably we made intent match with outcome. We started measuring successful resolution in the first response, which increased by 40% after adding the AI. I am much less concerned with volume of chats, and more concerned with human error reduction. When trying to define the balance between automation and human touch, I use a "heat versus heart" metaphor. High-volume, low-emotion interactions like pickup location change requests, questions about luggage, siting ETA, etc. go to AI. Any interaction with a human (or "heart") involved, such as VIPs asking for custom-designed tours, complaints, unexpected changes, or alterations mid-transfer - get escalated to me or one of my team. The moment I see the word "cancel" or "refund," a human steps in. The real win? Guests have no idea it was AI - they felt the peace of mind we want them to feel. The metric that matters to me is frictionless confidence.

Hello. I am the Chief Digital Officer and I am responsible for the development and implementation of the company's digital development strategies, including the use of AI for automation. Our team implemented a system where generative AI answered users' questions, taking over some of the work of customer support staff, thereby reducing operating costs, speeding up the response process, and, consequently, the level of customer satisfaction. This change somewhat altered our approach to evaluating contact center performance. Previously, we were interested in such indicators as the average time to process a request, the average time to the first response, the level of service, customer satisfaction, and the quality of the operator's work. After the introduction of AI, many requests were delegated to generative models, and such an indicator as "human-AI" appeared. After the introduction of AI systems, the metrics remained the same, but contact center operators primarily began to respond to non-trivial requests and make suggestions to improve the work of generative AI and errors in the system. In our company, AI is the first point of contact for users with questions. It mainly answers frequently asked questions, technical questions that have a clear answer and a certain algorithm of actions. For example, if a user has found a bug in the system and the AI bot does not know how to answer the question, we recommend that they refer it to our support specialists. To estimate the number of user requests to AI, we ask for feedback on the "Was this information useful to you?" question and then compare the figure with the number of requests to support staff.

I am a Business Analyst at 8rental. Our company provides car rental services in different European countries. I analyze and optimize logistics processes. My responsibilities include analyzing the efficiency of contact centers and finding ways to improve them. Our team has implemented AI tools in the work of contact centers. And this implementation has led to changes in the approach to the work of operators. Currently, AI processes simple popular requests such as "Rent a car", 'Deliver to X address', etc. And more complex requests, such as receiving claims are handled exclusively by humans. Our AI tool saves the processed requests in the database, so we can see a complete picture of how many requests were processed automatically and how many were processed with the help of people. After implementing AI in the contact center, the success indicators added the "Autonomy level" item, which shows what percentage of requests a virtual assistant/chatbot can process without human intervention. A high or low percentage indicates the level of development of the tool and the possibilities for its improvement. The "Request processing time" item is also important: the faster the AI handles the request, the better. And "Customer satisfaction level" indicates how accurately the AI provides answers. This change hasn't altered drastically our approach to evaluating contact center performance. Previously, we paid attention to such indicators as the average time of request processing, the first response average time, the service level, customer satisfaction, the operator's work quality, etc. After the implementation of AI tools, the metrics remained the same, but such an indicator as "human-AI collaboration" emerged. This indicator shows how well AI tools respond to customer requests and the percentage of fully processed requests.

We're still tracking traditional metrics such as Average Handle Time (AHT), but we interpret them differently. Now that we've been using AI to answer basic questions such as how to register for an event or how to navigate on our platform, the AHT for human agents, in fact, went up 15%, but the reason is the right one. Agents are now able to spend more time having the high-value, complex conversations with trade show organizers around reputation management strategies and data-driven insights from their reviews. The game-changer has been measuring 'Resolution Depth', our proprietary metric which measures whether an interaction merely answered a question or created new business opportunity. We have had to make this same shift, and since then have realized a huge increase in upsell conversions from support interactions. We deploy a light yet powerful "Three-Layer Filter" to blend the automation with the human touch: (1) AI takes over tier-1 transactional questions (access an account, basic reporting), (2) tandem AI-human squads guide tier-2 operational inquiries (discrepancy reconciliation, review moderation), and (3) humans alone facilitate tier-3 strategic dialogues (damage control, exhibitor retention tactics). We are cutting our operational costs while driving up customer satisfaction scores by 22 points in the process.