Different roles. One customer task.
The AI has finished its task. The employee is starting theirs. The customer is still trying to finish the same thing.
Part II separated information, advice and action. Now look between them. Does the story follow the customer when responsibility changes hands? Or does the customer have to bridge the gaps?
When a person takes over, does the story come with them?
The Taobao study found that human intervention worked differently depending on the failure and its timing. It does not supply a universal minute-by-minute escalation rule. It does question the assumption that emotional dissatisfaction can always be handled like a technical exception, even after frustration has built up. [3]
- Can the employee see what was tried and where it failed before asking the customer to repeat it?
- Can customers find an alternative to the AI route when their situation calls for one?
- Can the employee explain feasible remedies and constraints, rather than simply acknowledge the problem?
- Is transferred customer information limited to what is necessary and protected by appropriate access controls?
The button exists. What happens after someone presses it? Dividing work inside the company should not make the customer do it twice.
Put customer effort beside handling time
The conversation ended. Did the problem? Someone may have given up, switched channels or started the explanation again. In the Taobao experiment, handling time and customer ratings did not move in the same direction. [3]
| What happened | What to observe | What not to assume |
|---|---|---|
| The conversation ended quickly | Distinguish resolution, abandonment and repeat contact | Shorter must mean more satisfying |
| A person took over | Repeated explanations, context transfer and ability to act | A human handoff must have repaired the relationship |
| A recommendation was delivered | Whether its scope and reasoning were understood | More recommendation clicks must mean more trust |
These are proposed observations, not validated measures of this design exercise. Even improved before-and-after results would not isolate AI placement from staffing, policy, demand or product changes. Rather than force customer effort into one score, begin by finding where someone had to start again.
Give the intended relationship an operating form
You cannot promise expertise and leave the frontline with a script. You cannot promise human connection and withhold the information and authority that make it useful.
This is where I bring a Brand Design Management lens: the relationship between a promise and the conditions for delivering it. Who knows what? Who can change what? What must reach the next person? An interface alone cannot answer those questions.
A business can divide its work. That does not divide the customer’s problem.
Scope of the proposed exercise
This note draws on the company accounts [1][2], the Taobao preprint [3] and qualitative service research [4] used in Part I. The appliance journey, tables and review questions are the author’s illustrations, not documented operating procedures of those companies. They do not establish technical or legal suitability, guarantee outcomes, prescribe an org chart or set a universal automation ratio.