TESLA
AI-Driven Charging Journey
Experience Optimization Proposal
Charging Experience · Product Manager
Jinyao Ouyang
Contents
01
Market Status & Pain Points
Industry Data · User Research · Competitive Analysis
02
Core Concept: AI-Driven Charging Decision System
From Static UI to AI-Personalized Experience
03
User Journey & Examples
Full-Journey AI Optimization · Wireframes / Hi-Fi Mockups
04
AI Recommendation & Operations Loop · Metrics
Data Feedback · System Optimization · Continuous Learning · Metrics
05
Technical Feasibility & Implementation
Tech Stack · Difficulty Assessment
06
Roadmap & Business Value
4-Phase Plan · Success Metrics
01
China EV Charging Industry — Macro Data
12.82M
Total charging devices
As of end of 2024
110B
Annual charging volume (kWh)
38% YoY growth
+80%
National Day expressway growth
2024 vs 2023
98%
Expressway service area coverage
35,000 charging piles
Tesla China Charging Network
→ 2,300+ Supercharger stations
→ 12,000+ Supercharger stalls
→ 100% provincial capital coverage
→ 640+ Destination Charger sites
→ 450+ sites open to non-Tesla EVs
→ 15-min charging radius (BJ/SH/SZ)
Sources: NEA Jan 2025 / EVCIPA / Tesla official site
Core User Pain Points
79.2%
ICE vehicles blocking charging spots
Source: EVCIPA
32.4%
Charger info not updated in real time
~60%
Frequently encounter broken/faulty chargers
Source: CCA
71.2%
Concerned about unstable voltage/current
Source: EVCIPA
24.5%
Clunky end-to-end charging flow
AI helps users avoid infrastructure issues and directly solves decision-making issues
27.3%
Inaccurate facility availability status
Source: CCA
Source: CCA
Source: CCA
Infrastructure pain points
Information & decision pain points
The core problem with public charging is that users are forced to repeatedly make complex decisions, alone, under high uncertainty.
Tesla's Structural Advantages
Car, App, Charging Network & Payment Account natively integrated
Tesla isn't a single charging platform — it's a complete closed-loop ecosystem.
Higher-quality real-time data
Including remaining battery, navigation route, station status, and historical charging behavior.
Online experience can directly improve offline network efficiency
Beyond helping users find chargers — it also optimizes queue distribution, station utilization, and operational scheduling.
The charging network has shifted from coverage-driven to experience-driven — the window for AI optimization is now
02
Core Concept: From Information Lookup Tool to AI-Driven Decision System
Today
Information lookup, static UI
→
AI-Driven
From "user triggers decision" → "system triggers decision", adaptive dynamic UI
AI Personalization — Layered Logic
Dynamically infer what the user prioritizes from behavior and context signals
Input signals
- Commute frequency
- Historical station preferences
- Dwell time / click depth
- Battery / trip / time of day
System infers current priority
- Time-first
- Cost-first
- Context-fit-first
Corresponding recommendation & UI
- Nearby available stations
- Low-price time slots
- Mall / long-trip plans
Information density auto-adapts
Same scenario, different UI density auto-generated from user behavior
Simple Mode · Action-Oriented Users
Detailed Mode · Data-Oriented Users
03
User Journey: AI-Driven Charging End-to-End
1
Sense the Need
- Predict future trips from commute patterns
- Combine battery and network load to recommend at the optimal moment
→
2
Find & Decide
AI generates explainable multi-option recommendations to speed up the decision:
- Dynamic ranking
- Multi-option
- Availability prediction
- Cost estimation
- Composite recommendation
→
3
During Charging
- Real-time progress and predicted completion
- Live cost updates
- Timely move-your-car reminders
- Improves user comfort and resource efficiency
→
4
Pay & Leave
- AI bill summary (time vs cost vs alternatives)
- Auto-invoice & reimbursement, personalized perks
- AI summary of choice value (time / cost / fit)
→
5
Operations Loop
- 72-hour demand forecasting
- Dynamic pricing suggestions
- Behavior data feeds back to operations, continuously improving recommendations and network efficiency
Step 1 Sense the Need — In-Car Display
Scenario: Sat 12:30 · battery 32% · approx 128 km range
AI detects charging need
App: early push
Car display: real-time prompt
User enters quick decision:
View recommended stations
Decide whether to charge now
If user dismisses or doesn't respond,
recommendation persists in dashboard
supporting later re-entry into the flow
Step 1 Sense the Need — Mobile App
Detect charging need early, reach the user at the optimal moment
Low-interruption reach, influence decisions early
Multi-objective recommendations balancing efficiency and cost
Convert recommendations into actionable decisions
Step 2 Find & Decide
AI generates explainable options, letting users choose between efficiency and control
Single-option recommendation (low decision cost)
AI surfaces the best option
Best for quick execution
One-tap go
Multi-option comparison (high user control)
Provides 2–3 candidate options
Compare price / time / distance
User chooses
Simple Mode
Detailed Mode
Step 3 During Charging
Real-time status + key decision prompts + charging strategy hints
Key status · instantly readable
Time remaining and current state shown
No extra interpretation needed
Enhanced decision info · supports dynamic adjustment
Price trend and charging dynamics shown
Helps decide whether to keep charging
Builds user understanding and trust
Simple Mode
Detailed Mode
Step 4 Pay & Leave
AI explains the decision result, and feedback continuously improves recommendations
Explain the decision result,
reduce user uncertainty
Close the transaction loop, secure the experience
Collect decision and experience gaps to improve the recommendation system
04
AI Recommendation & Operations Optimization Loop
1
User Decision
User selects a station
- Recommendation / comparison
- App / car display
2
Behavior Feedback
Real-world behavior
- Did they go
- Wait or abandon
- Actual charging duration
3
Data Collection
Data flows back to the system
- Behavior data
- Station load data
- User preference updates
4
Operations Optimization
Ops & system tuning
- Demand forecasting
- Dynamic pricing
- Station scheduling / maintenance
5
Recommendation Upgrade
Model update
- Ranking strategy
- User profile refresh
- Next-round recommendations sharper
UI / Recommendation system
Behavior tracking
(event logging)
Data pipeline
(real-time stream)
Demand forecast / dynamic pricing
Ranking model
(recommendation algorithm)
AI Solution Metrics Analysis
North Star: reduce user decision cost (time + uncertainty)
1
Sense the Need
Improve reach efficiency
Push CTR
Baseline Industry 2-4%
Target >15%
Recommendation acceptance
Baseline Industry 20-35%
Target >40%
Dismiss rate
Baseline —
Target <15%
2
Find & Decide
Core stage: AI replaces manual comparison
Decision time
Baseline ~5 min
Target <3 min (-40%)
AI option selection rate
Baseline MVP-period data
Target >60%
Station match accuracy
Baseline —
Target >90%
3
During Charging
Dynamic optimization during charging
Move-your-car response rate
Baseline —
Target >70%
Overstay rate
Baseline Current level
Target -30%
Charging strategy execution rate
Baseline —
Target Continuously tracked
4
Pay & Leave
Capture experience value
NPS score
Baseline 30-40
Target 45-55 (+15)
Survey completion
Baseline Industry 10-15%
Target >25%
Operations loop
Data feedback · system tuning
Peak/off-peak charging ratio
Baseline 3 : 1
Target 2 : 1
Demand forecast MAPE
Target < 15%
Recommendation CTR monthly growth
Target +2-3%/mo
Monthly active charging retention
Baseline ~60%
Target ~70% (+10%)
All baselines are industry data or reasonable assumptions; calibrate via A/B testing post-launch. Metrics validated stage-by-stage per roadmap; MVP focuses on Step 1-2 metrics.
05
Technical Feasibility Analysis
Module | Difficulty | Notes |
User Profile Model | ⭐⭐ Low | Behavioral clustering and preference modeling on historical data |
Real-Time Station Data | ⭐ Very Low | Real-time stall API already exists |
Proactive Trigger Engine | ⭐⭐ Low | Rules-based push timing (battery / trip / off-peak / favorite stations) |
Dynamic UI Rendering | ⭐⭐⭐ Medium | Server-Driven UI; backend returns JSON config |
Recommendation & Ranking | ⭐⭐⭐ Medium | Multi-objective ranking on time / cost / distance / preference, supporting multi-option generation and dynamic adjustment. MVP can use a rules engine, unifying proactive recommendation and user-initiated query |
Natural Language Interface | ⭐⭐⭐⭐ High | LLM-based; MVP can use rules + templates |
Charging Demand Forecasting | ⭐⭐⭐ Medium | Predicts station load from historical data, supports ops-side resource optimization (time-series methods) |
Conclusion: This proposal is buildable on existing data and system capabilities. The MVP can be validated quickly with a rules engine, then progressively augmented with ML models for recommendation and forecasting.
Frontend → API Gateway → Business Services → AI/ML Layer → Data Layer → Infrastructure
Frontend Touchpoints
MVP
Tesla App (iOS/Android)
React Native · APNs/FCM · WebSocket
Push · Find chargers · Monitor · Pay + NPS
MVP
In-Car Display
Qt/C++ · CAN Bus · 10-sec notification bar
In-drive charging hints · live battery/nav
Months 6-12
Server-Driven UI
JSON Schema UI config
Dynamic render · simple/detailed switching · A/B without releases
API Gateway
MVP
API Gateway + Auth + Rate Limiting
Kong/Nginx · OAuth 2.0 · RESTful + gRPC · CDN
Tesla Account auth · rate limit · internal/external routing
Business Services
MVP
Proactive Trigger Engine
Drools rules · Kafka events
Battery/trip/price → push timing → backoff
MVP
Charging Station Service
Go/Java microservices · reservation system
Stall status · availability prediction · booking
MVP
Payment & Billing
WeChat/Alipay SDK · invoice API
Auto-settlement · invoice · reimbursement
Months 3-6
Dynamic Pricing
Price-band engine · electricity-rate API
Peak-shaving · ±20% · compliance checks
AI / ML
MVP
Recommendation Ranking
Weighted rules → LightGBM
Multi-objective → recommended/fastest/cheapest
Months 3-6
72h Demand Forecast
Prophet → LSTM Ensemble
Station load forecast → ops + pricing
MVP
User Profile
Collaborative filtering · K-Means
Dynamic context tags · not static profiles
Months 3-6
AI Summary + NPS Analysis
Rule templates → LLM
Value quantification · alternative comparison · feedback loop
Data Layer
MVP
Real-Time Data
Redis · Kafka · WebSocket
Stall status · power/cost · queue
MVP
Business Database
PostgreSQL · MongoDB
User · order · invoice · NPS
Months 3-6
Data Warehouse + Tracking
ClickHouse · Mixpanel
Behavioral analytics · A/B testing · training sets
MVP
External Data Ingestion
Weather API · electricity API · maps
Weather · pricing · traffic · holidays
Infrastructure
MVP
Cloud Infra + Monitoring + Edge Compute
AWS/Alibaba Cloud · K8s · Docker · CI/CD · Prometheus · ELK · vehicle-side edge
Container orchestration · log monitoring · privacy-first edge processing
Step 1 Sense the Need�
Trigger engine → ranking → push/in-car
Step 2 Find & Decide�
Station service → ranking → dynamic pricing → UI
Step 3 During Charging�
Real-time data → WebSocket → power/cost/reminders
Step 4 Pay & Leave�
Payment → AI summary → NPS → data loop
MVP (Months 0-3)
Recommendation Engine (Months 3-6)
Adaptive UI (Months 6-12)
System Architecture
06
Roadmap — Phased Plan
Months 0-3
MVP
Smart recommendation card
Rules engine
A/B test validation
Card acceptance rate>40%
(industry recommendation acceptance: 20-35%)
Months 3-6
Recommendation Engine
Multi-option recommendations
Real-time prediction
Option selection rate>60%�(baseline = MVP-period data)
Months 6-12
Adaptive UI
Server-Driven UI
Information density auto-adapts
NPS +15 (baseline: 30-40)
Retention +10% (baseline: ~60%)
Months 12+
AI Ecosystem
FSD integration
Solar-storage-charging integration
Charging experience
fully intelligent
Core Design Principles
AI Assists, Never Replaces
AI recommends; the user always has the choice
Progressive Personalization
Sensible defaults for new users, sharper with use
Explainability
Every recommendation comes with a reason
Privacy-First
Edge processing preferred over cloud
Offline Graceful Degradation
Falls back to local basic recommendations when offline
Use AI to extract more from the existing charging network
Charging is no longer the user's task — it's a service the system performs for the user
TESLA
AI 驱动充电全旅程
体验优化方案
Charging Experience · 产品经理
欧阳瑾瑶
目录
01
市场现状与痛点
行业数据 · 用户调研 · 竞品分析
02
核心理念：AI 驱动的充电决策系统
从静态界面到 AI 个性化体验
03
User Journey与示例
充电全旅程 AI 优化 · 线框图/高保真示意
04
AI推荐运营优化闭环与指标分析
数据反馈 · 系统优化 · 持续学习 · 指标分析
05
技术可行性与落地方案
技术栈 · 实现难度评估
06
Roadmap 与业务价值
4 期规划 · 衡量指标
01
中国充电行业宏观数据
1,281.8万
充电设施总量
截至2024年底
1,100亿
全年充电量(kWh)
同比增速38%
+80%
国庆高速充电增长
2024 vs 2023
98%
高速服务区覆盖
3.5万台充电桩
特斯拉中国充电网络
→ 2,300+ 座超充站
→ 12,000+ 根超充桩
→ 100% 省会覆盖
→ 640+ 目的地充电站
→ 450+ 座已向非Tesla开放
→ 15分钟充电生活圈（北上深）
数据来源：国家能源局 2025年1月 / 中国充电联盟 / 特斯拉官网
用户充电核心痛点
79.2%
燃油车占位是首要问题
来源：EVCIPA
32.4%
充电桩信息更新不及时
~60%
经常遇到充电桩损坏或故障
来源：中消协
71.2%
关注设备电压电流不稳定
来源：EVCIPA
24.5%
充电全流程操作不顺畅
基建问题AI帮用户规避，决策问题AI直接解决
27.3%
设施使用状态不准确
来源：中消协
来源：中消协
来源：中消协
基础设施痛点
信息与决策痛点
公共充电在线体验的核心问题是“用户需要在高度不确定的信息中，反复独立完成复杂决策”。
特斯拉优势
车、App、充电网络、支付账户天然打通
特斯拉不是单一充电平台，而是完整闭环生态。
拥有更高质量的实时数据
包括车辆剩余电量、导航路线、站点状态、历史充电行为等。
线上体验可以直接影响线下网络效率
不只是帮助用户找桩，还能优化排队分布、站点利用率和运营调度。
充电网络已从覆盖驱动进入体验驱动阶段，AI优化的时机窗口已到
02
核心理念：从信息查询工具，到 AI 驱动的充电决策系统
当前
信息查询，静态 UI
→
AI 驱动
从“用户触发决策” → “系统主动触发决策”, 动态自适应 UI
AI 个性化推荐的分层逻辑
基于行为与场景信号，动态判断用户当前更关注什么
输入信号
- 通勤频率
- 历史选站偏好
- 停留时长/点击深度
- 电量/行程/时间段
系统判断当前优先级
- 时间优先
- 成本优先
- 场景匹配优先
对应推荐与信息呈现
- 推荐附近可用站
- 推荐低价时段
- 推荐商圈/长途方案
信息密度自适应
同一场景，基于用户行为自动生成不同密度的界面
简洁模式 · 行动型用户
详细模式 · 数据型用户
03
User Journey：AI 充电全旅程
1
感知需求
- 基于用户通勤习惯预测未来出行需求
- 结合当前电量和充电网络负载，在最优时机主动推荐充电
→
2
找桩决策
AI 生成可解释的多方案，帮助用户快速完成选择：
- 动态排序
- 多方案
- 空桩预测
- 费用预估
- 综合推荐
→
3
充电中
- 实时充电进度与预计完成时间
- 费用动态变化
- 及时提醒移车
- 提升用户安心感与资源利用效率
→
4
支付离场
- AI账单总结（时间 vs 费用 vs 替代方案）
- 自动发票 & 报销支持，个性化权益推荐
- AI总结本次选择价值（时间 / 费用 / 匹配度）
→
5
运营端
- 72h需求预测
- 动态定价建议
- 用户行为反哺运营策略，持续优化推荐与网络效率
Step 1感知需求 车机端
场景：周六12:30 · 电量32% · 约128km续航
AI识别充电需求
App：提前推送
车机：实时提醒
用户进入快速决策：
查看推荐站点
判断是否立即充电
用户未响应或主动关闭
推荐沉淀至 dashboard
支持后续再次进入决策
Step 1感知需求 APP端
提前识别充电需求，在最佳时机主动触达用户
低打扰触达，提前影响用户决策
多目标优化推荐，平衡效率与成本
将推荐结果转化为可执行决策
Step 2找桩决策
AI生成可解释方案，帮助用户在效率与控制之间做选择
单方案推荐（低决策成本）
AI给出最优方案
适合快速执行
一键前往
多方案对比（高决策控制）
提供2–3个候选方案
支持价格 / 时间 /距离对比
用户自主选择
简洁模式
详细模式
Step 3充电中
实时状态反馈 + 关键决策提醒 + 充电策略建议
关键状态 · 即时可读
显示剩余时间与当前状态
无需额外判断
决策信息增强 · 支持动态调整
展示价格趋势与充电变化
帮助判断是否继续充电
提升对系统的理解与信任
简洁模式
详细模式
Step 4支付离场
AI解释决策结果，并通过反馈持续优化推荐策略
解释决策结果，
降低用户不确定性
完成交易闭环，保障转化体验
采集决策与体验偏差，优化推荐系统
04
AI推荐与运营优化闭环
1
用户决策
用户选择充电站
- 推荐/对比决策
- APP/车机
2
执行行为反馈
真实行为反馈
- 是否前往
- 是否等待/放弃
- 实际充电时长
3
数据采集
数据回流系统
- 行为数据
- 站点负载数据
- 用户偏好更新
4
运营优化
运营与系统优化
- 需求预测
- 动态价位
- 站点调度/维护
5
推荐升级
推荐模型更新
- 排序策略优化
- 用户画像更新
- 下一轮推荐更精准
UI呈现 / 推荐系统
行为埋点
（事件追踪）
数据管道
（实时数据流）
需求预测 / 动态价位
排序模型
（推荐算法）
AI方案指标分析
核心目标：降低用户充电决策成本（时间 + 不确定性）
1
感知需求
提升触达效率
推送点击率 CTR
Baseline 行业 2-4%
目标 >15%
推荐采纳率
Baseline 行业 20-35%
目标 >40%
dismiss率
Baseline —
目标 <15%
2
找桩决策
核心环节：AI决策替代人工比较
找桩决策时间
Baseline ~5 min
目标 <3 min (-40%)
AI方案选择率
Baseline MVP期数据
目标 >60%
桩匹配准确率
Baseline —
目标 >90%
3
充电中
动态优化充电过程
移车提醒响应率
Baseline —
目标 >70%
超时占位率
Baseline 当前水平
目标 下降30%
充电策略执行率
Baseline —
目标 持续追踪
4
支付离场
沉淀体验价值
NPS评分
Baseline 30-40分
目标 45-55分 (+15)
问卷完成率
Baseline 行业 10-15%
目标 >25%
运营闭环
数据反哺 · 系统优化
峰谷充电量比
Baseline 3 : 1
目标 2 : 1
需求预测 MAPE
目标 < 15%
推荐CTR月增长
目标 +2-3%/月
月活充电留存
Baseline ~60%
目标 ~70% (+10%)
所有 Baseline 为行业数据或合理假设，上线后需通过 A/B 测试校准。指标按 Roadmap 分期逐步验证，MVP 阶段聚焦 Step 1-2 指标。
05
技术可行性分析
模块 | 难度 | 说明 |
用户画像模型 | ⭐⭐ 低 | 基于历史行为数据进行用户分群与偏好建模 |
实时站点数据 | ⭐ 很低 | 已有实时桩位API，现有能力 |
主动触发引擎（Proactive Trigger） | ⭐⭐ 低 | 基于规则判断推送时机（电量 / 行程 / 低峰时段 / 常用站点） |
动态UI渲染 | ⭐⭐⭐ 中 | Server-Driven UI，后端返回JSON配置 |
方案推荐与排序 | ⭐⭐⭐ 中 | 综合时间 / 成本 / 距离 / 偏好进行排序，支持多方案生成与动态调整，MVP可用规则引擎，支持主动推荐与用户主动查询的统一逻辑 |
自然语言交互 | ⭐⭐⭐⭐ 高 | 接入LLM，MVP可用规则+模板 |
充电需求预测 | ⭐⭐⭐ 中 | 基于历史数据预测站点负载，支持运营侧资源优化（时间序列方法） |
结论：该方案基于现有数据与系统能力即可实现，MVP阶段通过规则引擎快速验证，后续可逐步引入模型优化推荐与预测能力。
前端触点 → API网关 → 业务服务 → AI/ML层 → 数据层 → 基础设施
前端触点
MVP
Tesla App (iOS/Android)
React Native · APNs/FCM · WebSocket
远程推送 · 找桩 · 充电监控 · 支付+NPS
MVP
车机端 In-Car Display
Qt/C++ · CAN Bus · 10秒通知条
驾驶中充电建议 · 实时电量/导航
6-12月
Server-Driven UI
JSON Schema UI配置
动态渲染 · 简洁/详细切换 · A/B无需发版
API 网关
MVP
API Gateway + 鉴权 + 限流
Kong/Nginx · OAuth 2.0 · RESTful + gRPC · CDN
Tesla Account鉴权 · 限流 · 内外部API路由
业务服务
MVP
主动触发引擎
Drools规则 · Kafka事件
电量/行程/电价→推送时机→退避
MVP
充电站服务
Go/Java微服务 · 预约系统
桩位状态 · 空桩预测 · 预约
MVP
支付与账单
微信/支付宝SDK · 发票API
自动结算 · 发票 · 报销
3-6月
动态定价
价格带引擎 · 电价API
削峰填谷 · ±20% · 合规校验
AI / ML
MVP
方案推荐排序
加权规则 → LightGBM
多目标排序→推荐/最快/最省
3-6月
需求预测 72h
Prophet → LSTM Ensemble
站点负载预测→运营+定价
MVP
用户画像
协同过滤 · K-Means聚类
动态场景标签·非静态画像
3-6月
AI总结+NPS分析
规则模板 → LLM
价值量化·替代方案对比·反馈回流
数据层
MVP
实时数据
Redis · Kafka · WebSocket
桩位状态·功率/费用·排队
MVP
业务数据库
PostgreSQL · MongoDB
用户·订单·发票·NPS
3-6月
数据仓库+埋点
ClickHouse · Mixpanel
行为分析·A/B测试·训练集
MVP
外部数据接入
天气API · 电价API · 地图
天气·电价·路况·节假日
基础设施
MVP
云基础设施 + 监控 + 边缘计算
AWS/阿里云 · K8s · Docker · CI/CD · Prometheus · ELK · 车端边缘计算
容器编排 · 日志监控 · 隐私优先车端处理
Step 1 感知需求�
触发引擎→推荐排序→Push/车机
Step 2 找桩决策�
充电站服务→推荐排序→动态定价→UI
Step 3 充电中�
实时数据→WebSocket→功率/费用/提醒
Step 4 支付离场�
支付服务→AI总结→NPS→数据回流
MVP（0-3月）
方案引擎（3-6月）
动态UI（6-12月）
系统技术架构
06
Roadmap 分期规划
0-3月
MVP
智能推荐卡片
规则引擎
A/B测试验证
推荐卡片采纳率>40%
（推荐采纳率通常20-35%）
3-6月
方案引擎
多方案推荐
实时预测
方案选择率>60%�（baseline为MVP期数据）
6-12月
动态UI
Server-Driven UI
信息密度自适应
NPS +15分（baseline:30-40)
留存率+10%（baseline:~60%)
12月+
AI生态
FSD联动
光储充一体
充电体验
全面智能化
核心设计原则
AI 辅助而非替代
AI提供推荐，用户始终有选择权
渐进式个性化
新用户合理默认值，随使用时间越来越准
可解释性
每个推荐都附带理由
隐私优先
车端处理优先于云端
离线降级
无网络时回退到本地基础推荐
通过AI 让现有桩网络发挥更大效率
充电不再是用户的任务，而是系统为用户完成的服务