What kind of help is the customer looking for?
The customer did not come to meet your AI. They came for a coffee, a product that suits them, or a way to fix an order.
A familiar purchase may need almost no conversation. A problem may need someone who understands what happened. Same customer. Different moment. Different help.
An AI rollout can start with the wrong end of the question: What can we launch? How many requests can we automate? First ask what the person on the other side needs to get done.
AI might answer directly. It might help an employee give a better answer. Clearer options might remove the need for a conversation altogether. Adding a chatbot is a feature decision. Choosing how help reaches someone is a relationship decision.
Starbucks: an answer for the person serving the customer
In an account published in January 2026 and updated in June, Starbucks describes Green Dot Assist as a real-time resource for coffeehouse employees. It provides conversational access to recipes, routines and service standards. The company frames the tool as support for beverage craft and human connection. [1]
Notice where the answer goes. To the employee, who is still there with the customer. Making technology useful does not always mean putting it in front of the person being served.
This does not mean Starbucks keeps all AI backstage. The same account describes customer-facing digital discovery as well. The unit of comparison here is a particular encounter, not an entire company. Nor does this company account isolate the tool’s effect on satisfaction or sales.
L’Oréal: making expertise available as a service
L’Oréal Paris places Beauty Genius closer to the customer. Its official service page describes skin and hair analysis, recommendations, virtual try-on and connections to major US retailers, with access through its US website and WhatsApp. The page is undated; these are the capabilities described when rechecked on 17 October 2026, not a claim of identical availability worldwide. [2]
Here AI is in front. Which product suits me? How should I use it? The uncertainty before a choice becomes part of the service. Count only the advisers it might replace, and you miss the other choice: organizing expertise around a customer’s question.
That is my interpretation of the design choice, not a measured business outcome. The service is built around the brand’s own products. It should not be presented as independent advice across the whole market, or as a medical diagnosis. A fluent conversation must not obscure whose recommendations the customer is receiving.
Taobao: a human handoff is not a guarantee
A randomized field experiment in Taobao customer service compared employees supervising agentic AI for eligible chats with employees handling chats without agentic AI. Average chat duration fell, but customer ratings for AI-eligible chats also fell. The effectiveness of human intervention varied with the failure, the effort after escalation and its timing. Technical problems and emotional dissatisfaction did not respond in the same way. [3]
The paper was released in May 2026 and revised in June. It is a preprint, not a confirmed peer-reviewed journal publication. Its findings do not establish that AI support is worse everywhere. They do challenge the assumption that “a human is available if needed” fully describes the experience.
A person takes over. The customer starts again. Has the service really connected?
What happened before the transfer? Does the employee know—and have the means to act? These are questions I draw from the study, not a checklist its authors tested.
The same technology can create different relationships
A 2026 Journal of Service Research article examines service routines in four technical customer-support organizations, drawing on 41 interviews and other qualitative material. It considers both customer-facing encounters and the work behind them. The 2020–2024 observation period covers pre- and post-GenAI phases. The count refers to interviews, not necessarily distinct participants. This is not an experiment proving one arrangement superior across today’s AI services. [4]
Replacement or assistance? That distinction leaves too much out. Giving an employee knowledge is not the same as helping a customer choose. Neither is the same as acting on a request. One company can need all three. So can one customer journey.
A brand is choosing more than whether to use AI. It is choosing whom customers ask, what they entrust and how help reaches them.
Brand Design Management asks where the technology belongs
A friendly voice is not a helpful relationship. A fluent conversation can leave the problem untouched. Quiet technology can give an employee what they need to solve it.
In this note, Brand Design Management is the author’s lens for arranging expertise, products, employees, AI and customers around the experience a brand intends to offer. I am not claiming that these companies adopted a named BDM framework. The lens connects operating choices to the relationship customers actually encounter.
Not always a person in front. Not always AI out of sight. Choose the place that serves the relationship, rather than copying the latest interface. Part II puts that choice to work along one customer journey.
The question is not how much technology customers should see. It is what help should reach them.
Evidence and limits
[1] and [2] are company accounts of services and operating intentions, not independent outcome evaluations. [3] is a preprint whose journal peer-review status has not been established. [4] is qualitative research. These cases are not a controlled comparison of companies. The cross-case interpretation, BDM framing and review questions are the author’s proposals, not guarantees of satisfaction, loyalty or sales.