The most consistent productivity result in 425 pages of data should change how enterprises think about AI solutions.
6 minute read
This is Part Two of Making Every Conversation Count, an ongoing video series featuring 8x8 CEO Sam Wilson and Liz Miller, VP and Principal Analyst at Constellation Research. Miller covers customer engagement, CX strategy, and AI-driven business outcomes.
Stanford's 2026 AI Index contains 425 pages of data, charts, and footnotes. After working through it, one number sits at the top of my notes… 14-15%. According to the study from Erik Brynjolfsson and his colleagues, customer support agents using a conversational AI assistant saw a 14% increase in overall productivity. They resolved 14-15% more issues per hour. Less-experienced agents saw even bigger gains of 30-35 points.
But the same technology that helped a junior support agent resolve three times as many issues in an hour slowed down an experienced developer by 19%. The report calls these discrepancies “the jagged frontier.” Models that won gold at the International Mathematical Olympiad read analog clocks correctly only half the time. Hallucination rates across 26 leading models range from 22% to 94%.
The jagged frontier makes a great case for not just flinging AI around your organization every which way. It’s all about fit.
Customer support is the cleanest, most consistent, and most evenly distributed AI-boosted productivity result in the entire report. For over a year now, we’ve been arguing about whether AI is for cutting costs or improving the customer and employee experience. Perhaps we can put that argument to rest.
Instead, we should be focusing on whether our environments are set up for AI to work, and, more importantly, whether we’re applying it to the correct use cases.
In a recent conversation with Liz Miller, VP and Principal Analyst at Constellation Research, we discussed how just like in the AI index, we’ve seen that the most successful use cases for AI have the AI and agents working together.
Moreover, the more customized the use case, the better that AI actually works. A good result isn’t that we deployed AI, it’s that when a customer calls us with a specific problem, we come in with a solution that’s specific enough to actually succeed.
If we only see AI as a way to save money, deflect tickets, eliminate queues, and reduce headcount, we miss its real opportunity, just as early web companies missed the full potential of the internet by treating it as a digital brochure.
The Stanford data takes that argument from a hunch to a numbers-backed theory. AI is creating measurable productivity where customers benefit directly from the experience. That 14% is not coming from a chatbot replacing an agent. It is coming from a human agent being made better by AI directly inside a conversation.
Cost savings and improving customer experience were never opposites, despite us treating them like they were.
Step back, and you can see why the support number is so much tidier than the other statistics in the report.
Customer support has structured workflows, repeating patterns, and real-time feedback. Its goal unambiguously boils down to: Did the customer's problem get solved? And, critically, its medium is voice or chat. Support conversations produce mostly language-based data that an AI can immediately learn from. They happen at human pace, and nuance, tone, hesitation, frustration, and urgency are exactly the signals AI has gotten really good at picking up.
Voice is not a relic of some bygone support era. It is the highest-bandwidth, highest-fidelity channel humans use when something matters. When AI listens to that channel with high-quality transcription and the right context, it gets dramatically better. We chat in bullets, but we speak in paragraphs.
The Stanford report makes a point about this that deserves underlining. Capability is jagged. The same models that won gold at the International Mathematical Olympiad read analog clocks correctly only 50.6% of the time. Hallucination rates jump when models are pushed outside their comfort zone. Safety performance drops under adversarial conditions.
The way you keep AI working is to put it inside structured, well-instrumented, high-context environments. A customer support conversation, with full history, clean audio, and immediate feedback, is the exact right environment for AI to work in.
Most enterprise workflows aren’t; that’s why productivity numbers vary so widely across the report. The same (broadly speaking) technology can lift a junior support agent's output by a third, and yet slow an experienced developer by 20%.
Stanford asked organizations how far along they are with AI agent deployment. Despite a year of breathless conversations around "agentic AI," scaled usage was in the single digits across nearly every business function. In service operations, two-thirds of respondents reported no agent use at all. Even in the technology sector, a leader in adoption, only a fraction of organizations had moved past pilots.
So while the AI industry is selling the future, its customers are still working out the present. The companies winning right now are not the ones deploying autonomous agents that may or may not work in 18 months. They are putting AI inside the conversations they have every day, directly helping the humans who run those conversations.
If you are running a business right now, here’s what I think you should take from all of this. Stop thinking of your contact center as a budget item that AI will eventually replace. The data says that it is perfectly backward. Your contact center is the richest training environment AI has access to in your entire organization.
It is the first place where you will see consistent returns. It’s also the place where the productivity gain and the customer experience gain are one and the same.
That changes how you should be investing; put your money in plain old infrastructure rather than flashy, aspirational AI use cases. That includes:
Infrastructure improvements rarely make it into a flashy keynote. But they’re exactly what the 14% gain depends on.
Stanford's number is going to get a lot of headlines for the wrong reasons. People will read "14% productivity gain in customer support" and conclude that AI is coming for support jobs. But they’re looking at it all wrong: the number is just proof that AI works best when applied to the right use case.
Most organizations are waiting for a future that is already available to them in a more modest form. AI succeeds when it’s embedded in conversations, supporting humans in environments where context flows freely.
Voice still matters. Conversation still matters. Customers still pick up the phone, open the chat, and reach out when it counts. And after two years of searching for places where AI actually delivers measurable value, we just got handed a pretty clear answer.
It works best where your customers already are.
Stay current on what matters in CX and IT. Subscribe to our LinkedIn newsletter for regular analysis on the decisions shaping customer experience, technology, and AI. Clarity you can act on.
