AI Agents for Retail: From Kiosk to Checkout, a Face-and-Voice Store Guide

Direct answer: A conversational AI in-store assistant, also deployed as an AI retail agent online, is a face-and-voice system deployed on a kiosk in-store or on a product page online. Working as an AI shopping assistant, it greets shoppers in their language, guides them to the right product, handles size and availability questions, explains returns policy, and in premium contexts, delivers the service experience of a personal shopping assistant, at every price point, 24 hours, without staffing constraints. Ojin makes the two face models this runs on, Portrait where the same agent has to hold up across an estate, and Presence where it is carrying a luxury brand at close range.

The staffing model for retail has not changed much in 30 years. A floor associate greets you. You get good service if you get an associate who knows the product and has time to talk. You get indifferent service if they are busy or do not match your language or do not know the category. The service quality is associate-dependent, shift-dependent, and language-dependent.

A Human AI agent on the retail floor or the product page changes all three dependencies. It knows every product. It is never busy. It speaks the shopper's language. The service quality floor rises to a standard that previously required the best associate on the floor. Portrait, Ojin's fast face model, is aimed at exactly this shape of problem, fast enough and light enough to run to hold a conversation on every kiosk in an estate rather than one flagship installation.

Live deployments

Puma's Las Vegas flagship is among the most visible retail AI agent deployments. The face-and-voice agent provides product guidance, brand experience information, and personalised recommendations matching the high-energy, experience-forward positioning of the venue. The agent is not a FAQ kiosk, it is a brand presence.

K11 and Songdo, mixed-use commercial developments in South Korea and Greater China, deployed Human AI kiosk agents alongside the touchscreen self-service kiosks that preceded them. Vendor reporting from kiosk supplier Wavetec on these deployments cites meaningfully more completed interactions with the Human AI agents, a directional figure (commonly cited around 40%) rather than an independently audited benchmark; a broader search turned up no independent third party that has verified the figure, so it is presented here as vendor-sourced only. The difference: shoppers who would not commit to a touchscreen form, too many steps, wrong language, unclear UI, were willing to have a conversation.

None of the deployments above run on Ojin, and they are cited here as category evidence rather than as customer proof. Magic Moment Resort's 88-language deployment demonstrates the language coverage ceiling of a Human AI system: no retailer in any tourist-heavy location can staff 88 languages. The AI system covers every shopper, regardless of origin.

Where retail AI agents create the most value

The honest answer is that not every retail interaction needs a face. A shopper checking stock availability for a specific SKU is well-served by a search box. What face-and-voice changes is the category of interaction where language, trust, or guidance is the service.

High-consideration purchases. When a shopper is choosing between a £2,000 sofa and a £1,400 sofa, the decision is not made on spec-sheet comparison alone. The service interaction, an agent that asks the right questions, understands the space and the lifestyle, and makes a confident recommendation, changes the conversion rate. McKinsey's research on personalisation finds that guided, assisted experiences most often drive a 10 to 15% revenue lift, with company-specific gains ranging from 5 to 25% depending on sector and execution, a range consistent with what a face-and-voice agent would need to deliver to justify replacing the consultative service model without the staffing cost.

Cross-lingual service. The agent an Ojin deployment puts on the floor is created from one still photograph and speaks whatever languages it is configured for, which is the part no roster change can match. In any urban retail environment with significant tourist or immigrant shopper traffic, the gap between the languages a shopper speaks and the languages staff are able to deploy is real. A Human AI agent bridges it without a roster change.

Checkout and abandonment. Online, the failure mode is measurable. Baymard Institute's aggregate of 50 studies puts the average cart abandonment rate at just over 70%, with unexpected costs the most-cited reason and mobile worse than desktop. A large share of that is an unanswered question at the wrong moment, which is precisely what a live agent on the product or basket page exists to catch. The expectation behind it is not retail-specific: Zendesk's CX Trends research tracks the same impatience with waiting for an answer across every consumer sector.

Returns and policy handling. Returns conversations are a disproportionate source of negative brand experience. A frustrated shopper at a returns desk, waiting for an associate who is with another customer, is a preventable situation. A face-and-voice agent that handles returns policy questions and initiates the process, conversationally, in the shopper's language, takes the friction out before it becomes an experience-ending moment.

Online product discovery. The digital equivalent of the in-store associate is an AI agent on the product page. A shopper landing on a category with 200 SKUs and no clear entry point needs guidance. A face-and-voice agent that asks "what are you looking for today?" and narrows the selection to three relevant options converts better than a filter sidebar alone.

What the floor staff actually think

The objection that comes up first in a retail rollout is rarely from head office. It is from the floor. An associate who has spent five years learning a product range reasonably asks what a talking screen is for, and the honest answer has to be specific rather than reassuring. The agent takes the questions that do not need judgement, stock, sizing, returns policy, warranty terms, opening hours, where a category sits in the store, and it takes them at the moments when the floor is thinnest. It does not close the £2,000 sale. It does not read a hesitant customer and decide to leave them alone for two minutes.

Retailers that deploy this well tell their staff exactly that, and they measure it: if associate time is not visibly moving from repeat questions to selling conversations, the deployment is not working and no conversion figure will disguise it for long. Retailers that deploy it badly present the agent as a general-purpose replacement, discover it cannot do the judgement half of the job, and lose the goodwill of the people they need to make it work.

The premium brand register question

There is a legitimate concern in luxury retail: does a face-and-voice AI agent lower the brand register? The answer depends on how it is designed. A generic chatbot widget in a luxury boutique is jarring, it signals that the brand chose efficiency over experience. A purpose-designed AI persona, named, visually consistent with the brand aesthetic, voice-matched to the brand register, is experienced differently.

This is where the choice of face model stops being a technical detail. Ojin makes two: Portrait for speed and scale, and Presence when the face is the brand and has to survive close inspection. A luxury retailer buying the cheaper option and then complaining that the register feels wrong has answered its own question. The question is not "AI or no AI." It is "what does this AI represent about us?" A well-designed Human AI Agent at a luxury property answers that question the right way: it represents the brand's commitment to service quality and availability, extended to every shopper at every moment, in their language. Forbes' coverage of AI adoption across luxury fashion houses makes the same point from the brand side: the houses investing seriously are treating AI as a considered extension of service, not a bolt-on efficiency tool.

A retailer does not need a systems integrator to find out whether this works on its own product pages. The Ojin platform is a self-serve AI platform: build the agent, point it at the product data, put it on one category page and watch what happens to the questions the contact centre normally receives.

Frequently asked questions

Will shoppers use a kiosk if they can just ask a real associate?

Yes, if the experience is better. Vendor reporting on K11 and Songdo's deployments suggests a conversational face-and-voice agent outperforms self-service UI for shoppers who want guidance, though that figure is directional rather than independently audited. Associates remain present for complex, emotionally loaded, or escalating interactions. The kiosk serves the shopper who is browsing and not yet ready to engage a person.

How is product information kept accurate?

Product data is managed through the knowledge base layer (Human Agents), not hardcoded into the agent. Inventory, pricing, and spec updates feed through the knowledge base as they change. The agent does not carry stale information, it pulls from the same source-of-truth as the website.

Can the agent be branded with the retailer's own name and visual identity?

Yes. The agent character, face, name, voice, and persona, is custom per deployment. There is no visible Ojin branding in the shopper-facing interface. The agent is an extension of the retailer's brand, not Ojin's.

For the category, see what a Human AI Agent is and how it actually works, what a Human AI Agent is and when it beats recorded video, and what a conversational AI agent is and how it differs from a chatbot. For a shop-floor agent that has to stay on-brand in front of customers, see brand safety when deploying a Human AI Agent, and for the face itself, how realistic a Human AI Agent can look in 2026. Luxury retail, in-store kiosks, product recommendation agents, retail ROI, and the verified luxury case studies each get their own article in this vertical. Across the rest of this batch, the closest neighbours are telecoms, the upgrade conversation an operator runs at the same counter, and hospitality, the same face at a hotel front desk.

Multi-channel retail deployment with Ojin: ojin.ai/enterprise. Try a live Ojin agent: docs.ojin.ai.