Konstantin Semenenko
June 25, 2026
4
minutes read
The best-measured result comes from a controlled rollout across more than 5,000 support agents: AI raised resolutions per hour by about 14% on average, and 34% for the newest agents. The gains are largest where the work is repetitive and the staff is least experienced.
Customer support is the function where AI's savings are best documented, because they've been measured in a real contact center, not a survey. Here's what AI actually saves in support, with hard numbers, and where the savings land hardest.
The best-measured result comes from a controlled rollout across more than 5,000 support agents: AI raised resolutions per hour by about 14% on average, and 34% for the newest agents. The gains are largest exactly where the work is repetitive and the staff is least experienced.
We build support automation that sits behind a human, so this is from deploying it, not from a brochure.
The flagship number isn't a survey, it's a randomized field experiment (Brynjolfsson, Li & Raymond, NBER) across a contact center of over 5,000 agents.
The headline: AI didn't just speed up the average agent, it lifted the weakest agents the most.
Because AI encodes what your best people already know. A new hire with an AI assistant gets the phrasing, the policy, and the fix that a five-year veteran would reach for, which is why the NBER study found the largest gains among the least-experienced agents. It compresses the learning curve, so the savings show up fastest in teams with high turnover or heavy seasonal hiring.
Fewer agent-hours per ticket and faster onboarding. McKinsey estimates generative AI could cut human-serviced contacts by up to 50% in industries like banking and telecom, and lift the productivity of customer operations by 30 to 45% of the function's cost. For a team of any real size, a double-digit lift in resolutions per hour is a measurable headcount saving.
When the AI answers confidently from the wrong source. Generic bots that aren't grounded in your own documents produce answers that look right and aren't, and the cost moves from answering to correcting. Workday found nearly 40% of AI time savings get eaten by rework when output isn't checked. In support, that shows up as escalations and angry customers, which erases the saving.
Put the AI behind a human, not in front of the customer, and ground it in your own knowledge base. The agent stays in control, the AI drafts and retrieves, and a senior review loop keeps the answers accurate. That's the difference between support automation that saves money and a bot that generates rework. If you want a team that builds it that way, that's AI Dev Team, and if you're deciding who should build it, start with who should build your MVP.
