Give the AI to your agents first

Service leaders planning for 2027 have to decide where AI goes first: in front of customers, or beside the people who serve them. The evidence favours the second. Gartner expects half of the organisations that planned to cut their customer service workforce significantly because of AI to abandon those plans by 2027, and in its poll of 163 service leaders, 95% said they would keep human agents.¹

The strongest evidence for that choice comes from a study of more than 5,000 support agents at a software company, by Erik Brynjolfsson, Danielle Li and Lindsey Raymond. Agents given an AI assistant that suggested responses in real time resolved 14% more issues per hour on average. Novice and low-skilled agents gained 34%, while the most experienced saw little change. Customer sentiment improved, and more agents stayed in their jobs.²

The authors’ explanation is the part to plan around. The assistant appears to pass on the practices of the best agents to everyone else, which is why the newest staff gained most.² The payoff comes from new agents reaching competence sooner, and that is worth most in centres with high turnover or many languages to cover.

Klarna started at the other end

In February 2024 Klarna said its AI assistant had handled 2.3 million conversations in its first month, two-thirds of its customer service chats and the equivalent work of 700 full-time agents.³ Fifteen months later its chief executive, Sebastian Siemiatkowski, told Bloomberg the company was recruiting people again so that customers would always have a human to talk to, and described investing in the quality of human support as "the way of the future for us".⁴

Customers ask for the same thing. Our review of the 2026 evidence found most customers expect a company that uses generative AI for service to offer a route to a person.

What the assistant teaches you about automation

There is a second reason to start with agents. Each suggestion an agent accepts, edits or rejects is evidence about which contacts AI can handle on its own. After a few months the transcripts show which intents are routine enough to hand to a bot and which still need judgement, information that a team automating first has to guess at.

What to measure in 2027

Measure the assistant the way the study did, with issues resolved per hour, customer sentiment and agent retention, split by tenure, because the gains are concentrated among newer staff. Then track how long a new hire takes to reach the team’s median resolution rate. If that time falls, the assistant is paying its way before a single contact is automated.

At CAIS 2027 we want to hear from service teams in Asia-Pacific that have measured an agent assistant across more than one language, including where it did not help.