What does AI change in a support organization?
It compresses the ramp for new agents more than it speeds up your best ones. In the largest field study so far, issues resolved per hour rose 14% on average and 35% for the least experienced agents, with minimal effect on the most experienced, and attrition among agents with the tool ran 8.6% lower. Plan for that shape rather than for headcount reduction.
- Increase revenue
- Improve operational efficiency
- Reduce cost
Customer Operations
The evidence, before the pitch
Researchers tracked about 5,000 support agents at a Fortune 500 software company through a staggered rollout of a generative AI assistant. Issues resolved per hour rose 13.8% on average, 35% for the least experienced and lower-skill agents, and barely at all for the most experienced. Customers expressed more positive sentiment and asked for a supervisor less often. Attrition among agents who had the tool was 8.6% lower, and the difference came from newer people staying.
The mechanism the authors describe matters more than the headline: the assistant spread the working habits of the best agents to the newest ones. It did not replace experienced people. It shortened the time it took to become one.
What we build
Agent assist grounded in your own material, not a generic model guessing. Draft responses cite the article or the prior resolution they came from, so an agent can check rather than trust. Wrap-up summarization and disposition coding remove the after-call work that agents skip when the queue is deep, which is why reporting is unreliable in most contact centres.
Quality review changes shape entirely. A human QA team reads perhaps 2% of contacts. A model reads all of them, flags the ones worth a human’s attention, and the QA team spends its time on coaching instead of sampling.
Customer-facing assistants are in scope here, for the contact types that are genuinely deflectable, with escalation paths designed before launch rather than bolted on after the first complaint.
What we measure
Resolved per hour, first-contact resolution, and handle time by contact type, baselined before rollout. Track attrition separately and by tenure, because the retention effect is real and it is the one most likely to pay for the engagement on its own.
Typical projects
- Agent assist: suggested responses grounded in your own knowledge base and prior resolutions
- Conversation summarization and disposition coding at wrap-up
- Quality review across a far larger sample than a human QA team can read
- Customer-facing assistants for the deflectable question types, with escalation designed in
- Voice-of-customer analysis across tickets, calls, and reviews
The same work, by industry
How it starts
A three-week assessment scoped to this area: where the hours actually go, what is worth building, and what to fix first. You keep the findings whether or not there is a next part.