“Good service never makes everyone crowd into one channel.”
Of Beijing Mobile's average 20 million monthly customer-service inquiries, 60% are standardized requests such as checking balances, subscribing to plans, or querying data. Yet under the traditional model, human agents must handle both these "simple tasks" and complex issues like "seniors who don't know how to disable auto-renewal" or "signal troubleshooting" — and the usual result is "users frustrated by slow service, companies frustrated by high cost."
In truth, the pain point of telecom customer service has never been "not enough staff." The logic behind Beijing Mobile's AI agent "Xiaobei" is essentially a return to the first principle of "tiered demand response": let AI handle the standardized work that must be "fast, accurate, and cheap," and let people focus on the human-centered work that is "hard, warm, and specialized" — ultimately achieving the best of both cost and experience.
01 The Essence: Customer Service Is Essentially "Tiered Demand Matching"
The core cost of telecom customer service is labor cost, accounting for over 60%. Yet the biggest waste in the traditional model is using costly human agents for low-value, repetitive work. Beijing Mobile's 2024 customer-service research showed that 30% of human agents' working hours were spent on "balance checks" and "data-plan subscriptions," while the elderly users and complainants who most needed conversation waited over 10 minutes in queue. Behind this lie three inescapable contradictions:
1. The First Contradiction: Misallocated Resources — High-Cost Labor on Low-Value Work
The core value of human agents is "solving complex problems and conveying warmth." Yet on traditional hotlines they are forced to be "repeaters": when a user asks "what's my balance," they read out a number; when asked "how much data is left," they read it again. This thoughtless work wastes labor and frustrates users with the wait.
2. The Second Contradiction: Fuzzy Identification — Human Agents Kept When AI Should Take Over, and Vice Versa
When a user says "this plan is such a rip-off," are they "complaining" or "wanting to switch plans"? Traditional voice menus can only guide with "press 1 for balance, press 2 for plans," and a misidentification triggers complaints. When Beijing Mobile trialed a general-purpose AI model in 2023, its plan-interpretation error rate reached 12%, and many users griped that "the machine can't understand human speech."
3. The Third Contradiction: Redundant Cost — Night-Shift Staff "Watching Empty Lines"
Customer service must run 7x24, but overnight (22:00-6:00) inquiry volume is only 20% of daytime. Still, companies must schedule on-duty staff, and overtime pay makes night labor 1.5x the daytime cost. One on-duty agent said: "After midnight we often get just one call an hour, and most are data checks — pure time-wasting."
02 AI Agent Xiaobei: Confronting Real Business Problems, Not a One-Size-Fits-All
Xiaobei is not an "all-powerful smart agent," but a "precision patch tool" for the traditional model — fixing exactly where it falls short, never overstepping into "what machines do poorly."
1. First, Fix "Fuzzy Identification": Three-Dimensional Tags to Decide "Who Should Handle It"
To tier demands you must first "identify" them. Xiaobei uses a three-dimensional mechanism — "keywords + emotion + profile" — to sort demands with perfect clarity:
Keyword recognition captures standardized needs: with over 12,000 built-in telecom keywords, when a user says "check my balance" or "get a 5G plan," it matches an AI response directly, at 92% accuracy.
Emotion recognition captures humanized needs: through voice analysis it catches traits like "fast speech, repeated questions, emotional words" — for example, when a user says "this is so annoying, I'm over my data again," the system automatically transfers to a human.
Profile recognition captures high-priority needs: by connecting to the user database, when a senior (over 65) or VIP customer calls 10086, the system skips the voice menu and goes straight to a human.
2. Next, Fix "Misallocated Resources": AI Takes 80% of Standardized Work, Done in 10 Seconds
The standardized demands Xiaobei handles are centered on "replacing repetitive labor with technology" — not chasing "flashy tech," only "efficiency and cost savings":
Query-and-act in one: no human needed. When a user says "activate a 10GB data pack," Xiaobei identifies it and connects straight to the back-end billing system, completing it in 10 seconds with an error rate of just 0.3% — lower than the 2% for humans.
Autonomous night service, no human on duty: overnight "data checks" and "bill checks" are all handled by Xiaobei, with only complex issues routed to the "daytime ticket pool."
Proactive reminders to cut call volume: for issues like "plan expiry" or "data overage," Xiaobei pushes a "one-tap activation link" via SMS.
3. Finally, Fix "Redundant Cost": Staff Focus on the 20% High-Value Work, Doing It Well Rather Than Doing More
The staff freed by AI were not laid off, but redirected to "what machines cannot do":
Complex problems as "diagnosticians": signal anomalies, international-roaming failures, billing disputes — all experience-dependent issues — are handled entirely by humans.
Emotional needs as "advisors": for the "chat a bit more" needs of elderly or complaining users, human agents can communicate patiently.
The AI transformation of telecom customer service need not be a "great leap forward." Like Xiaobei, first understand the essence that "demands must be tiered and resources matched," then use AI for precise patching — not chasing "omnipotence," only "the right remedy." That is the real logic of landing value beyond technical showmanship.
03 Lessons from the Case: Don't Blindly Chase Technology — First Dig into the Business Essence
When many companies adopt AI customer service, they obsess over "are the model parameters big enough" or "are the features complete enough." But Xiaobei's case proves that landing telecom customer-service AI hinges on three essential industry rules:
1. Precise Identification Matters More Than "Intelligence"
A general large model can chat about weather and tell jokes, but cannot interpret the details of the "M-Zone plan." Although Xiaobei's vertical knowledge base focuses only on telecom business, it identifies needs precisely — showing that the first step for telecom customer-service AI is "knowing the business," not "knowing small talk." When Beijing Mobile used a general model in 2023, the error rate was 12%; after switching to a vertical knowledge base it rose to 92% — the best proof.
2. Resource Matching Matters More Than "Replacement Rate"
Some companies chase "AI replacing 80% of humans" without figuring out "what humans do after replacement." Xiaobei's AI replaces 80% of standardized demand so the remaining 20% of staff create higher value — that is the essence of "cost reduction and efficiency," not "layoffs and headcount cuts."
3. Human Boundaries Matter More Than "Full Automation"
Telecom service cannot lack "human warmth": the elderly need "hands-on" guidance, complainants need "empathy" — things machines cannot do. By keeping "one-tap to human" and a "silver-hair channel," Xiaobei guards the "human boundary" and avoids cold, impersonal technology.
The ultimate battleground of business is always efficiency, but the essence of efficiency is not speed but precision. Just as hospital triage is not to move patients faster, but to match the right patient with the right doctor at the right time, the lesson of Mobile's AI Xiaobei may lie right here: good service never makes everyone crowd into one channel.
