AI Customer Service Is Rewriting Peak Retail Economics

Australian online retailer Kogan said on September 22 that its AI service agents now resolve 67% of the conversations they handle, while 81% are completed without escalation to a person. Chief technology officer Goran Stefkovski told MARKETING-INTERACTIVE that customer satisfaction is seven percentage points higher among customers using the bot.

The company also reported more than 44,000 agent actions during a peak week and said 78% of customers prefer the automated channel. Those results are company-reported rather than independently audited, but they move the customer-service debate beyond pilot counts and simple deflection.

Kogan’s case suggests that the commercial value of service AI sits in demand elasticity: handling retail peaks without rebuilding seasonal headcount each time demand jumps. For retail CMOs, the decision is no longer whether to automate routine questions. It is which outcomes, safeguards and roles should govern the system.

What Kogan Automated

Kogan moved from a one-day hackathon to a live customer agent in six weeks, then reduced deployment time for additional agents to about two weeks. The system now covers order tracking, recommendations, returns, warranties and image verification, and the same architecture is being extended to New Zealand retailer Mighty Ape.

“Ninety percent of questions don't need a human to actually respond,” Stefkovski said. The operational logic is strongest during Black Friday, Christmas and other spikes, when human teams require recruitment and training while digital capacity can expand quickly.

Salesforce’s July case study adds useful detail. Kogan began with order-status enquiries, which represented roughly 60% of inbound contacts, then added more complex returns workflows. It reported absorbing 20% more sales volume while reducing customer-servicing costs by 10%.

Retail CMOs Are Taking Control Of The Commerce Loop

Target has put marketing, guest experience, Roundel and Target+ under one executive. The structure gives its CMO direct influence over customer trust and commerce revenue, raising a sharper governance question for retailers building media and marketplace businesses.

Mission MediaMission Media

Why Resolution Beats Deflection

Those economics can tempt executives to treat the smallest contact-centre payroll as the main measure of success. Kogan’s better idea is “true resolution,” which checks whether a question was fully answered without a follow-up contact.

That distinction matters because deflection can hide failure. A customer who gives up, repeats the question later or switches channels may look like a contained interaction even when the underlying problem remains. Retail leaders need a scorecard that combines verified resolution, repeat-contact rates, satisfaction, conversion, returns outcomes and cost to serve.

The operating model also matters. Kogan’s customer-care analysts, rather than a central engineering team alone, help develop and improve the agents. “What matters most to me is that the customer care team can own this channel,” Stefkovski said in Salesforce’s case study.

Where Human Access Still Matters

Kogan’s performance should not be read as permission to make human service difficult to reach. Gartner reported in August that 87% of 3,566 surveyed customers considered access to a human essential when companies use generative AI for service, even though half said AI made interactions easier.

The tension is not automation versus people. It is routine capacity versus accountable escalation. Order status and standard returns can be automated aggressively, while disputed payments, vulnerable customers, unusual product failures and policy exceptions need clear transfer rules and decision authority.

Kogan routes unresolved conversations to a representative and provides a summary on handoff. That continuity is commercially important: forcing customers to restart the conversation turns an efficiency gain into visible service friction.

What Retail CMOs Should Change

For CMOs and customer-experience leaders, the immediate task is to define automation boundaries before peak demand sets them by default. Each use case needs an owner, a resolution standard, an escalation trigger and a review process for errors.

The workforce question is equally concrete. Kogan is moving some employees from repetitive responses into knowledge management, agent design and exception analysis. That may reduce seasonal hiring, but it also makes frontline expertise part of the product-development process.

Kogan’s results suggest customer-service AI can improve both capacity and experience when brands measure completed outcomes and keep human support available. The executive test is not how many contacts an agent contains. It is whether customers get a correct answer, a clean handoff when needed and a better reason to return.