Command Cockpit

Overview of ad spend, installs, and channel performance (HKT business day / UTC+8)
→
placeholder
—
placeholder
—
placeholder
—
placeholder
—
placeholder
—
placeholder
—
placeholder
—
placeholder
—

Spend / Install / CPI Trend

loading
—

Data Quality

—

Channel Performance

—

AI Daily Brief

—

Data Anomalies

—

Today Action Queue

—
Data Sources: BigQuery AppsFlyer Push (Primary) + Platform API (Cost). Incomplete days do not participate in budget decisions.
BQ 24h scan: —

iOS / Android UA Daily Report

HKT natural day; AF installs are authoritative. Allocated country cost is explicitly labelled.

—
—
—
—
—
—

Country template matching Feishu

AF paid installs and ad conversions are separate; First-day photo (camera_photo_confirm_click) and First-day solve (solve_finalizing_show) are shown side-by-side. Hover over ℹ️ in the table header for underlying event definitions.

—
—

GPT Report Analysis & Recommendations

gpt-5.6-sol · high
—

Channel totals

—
—

Country / region detail

—
—
Metric basis and Obsidian preview
—
—
—
—
—
—
—
—

Channel Matrix

—
—
—
—
—

Google Ads Account Performance

—
—
—
—
—
—
—

Country × Channel Matrix

—

Campaign Performance

—

Creative Performance

—

ASA Keyword Performance

ASA keyword/search-term report waiting on ASA API credentials + fact_asa_keyword_daily. Scheduled for P1.

Google W2A Keyword Performance

—

Channel New Users

—
—
—
—
—

Module penetration

Main event UV / platform UV
—

Macro trend

PV · UV · Sessions
—

Adoption bands

Main event signal
—

Interaction preference

Feature usage share / platform UV
—

Module Key Funnels

User journey & step conversion rates
—

Daily module report

PV · UV · penetration
—

User Quality

Real AppsFlyer event aggregates: activation proxy, core actions, payer cohorts, and quality signals by channel.
real data
—
—
—
—
—
—

Core Actions by Channel

—

Cohort Revenue Proxy

—
—
Reviews sync automatically every four hours and reports share the inbox date range. Store APIs do not expose profile photos, so missing avatars use nickname initials. Google Play does not expose reviewer country; reports/appeals remain manual console workflows.
Written reviews
—
Average rating
—
5-star share
—
5-star reply rate
—
Low-rating share
—
Overall reply rate
—
Deep AI coverage
—
Avg response time
—

Store performance

Full 1–5 star distribution

Rating and review trend

5-star advocacy assets

5 ★

Low-rating risk topics

1–2 ★

App Store countries

Language performance

Version performance

Representative 5-star voices

AI growth brief

SEO · ASO · GEO · Product Ops

AI review operations weekly

Automatically analyzes the latest complete Monday–Sunday week against the preceding week.

Operations queues
Active filters

Review inbox

—
—
Select a review to inspect

Report workspace

manual only

Activation Quality

HKT install-cohort quality from AppsFlyer push data. First-day solve uses solve_finalizing_show and D1 retention requires app_launch, split by media source and geo.
BigQuery Live
—
—
—
—
—
—
—
—

By Channel

Installs / first-day solved users & rate / first-day pay users & subscribe rate / app_launch D1 retention (by first-touch media source).
—

By Channel × Geo

Media source drilled down to country — same metrics, searchable.
—
—

Weekly Growth Dashboard

Monday-anchored weekly reproduction of the Looker Studio Weekly-Active board. AppsFlyer Push behavior events cover iOS and allow-listed paid Android traffic; download and first-week-activation cohorts remain iOS-only because the Push install tables currently contain no Android rows.
BigQuery Live —
—
—
—
—
—
—
—
—
—
—

Weekly Trend

Active, downloads, retention and dormancy across the last 12 calendar weeks.
—
—

Monetization

Paywall funnel, purchase economics, subscription signals, and revenue by channel from AppsFlyer BigQuery events.
real data

Paywall Performance (Superwall)

Paywall opens, conversions, and A/B experiments from Superwall ClickHouse.
Superwall
Loading…

LTV & ROAS

Observed cohort revenue, current ROAS, payback status, and LTV:CAC by channel.
real data

Recovery Efficiency

Install-day cohort: ad cost vs. subscribe / paid / revenue, with CPI, CPS, CPP, RPD, expected 1-year revenue and 1-year ROAS by channel × campaign × country.
real data
—
—
—
—
—
—

Recovery Matrix

—
—

Web W2A

Proxy Web-to-App chain from Google Ads / Google W2A spend, clicks, app installs, and app-side purchase revenue.
proxy data
—
—
—
—
—
—
—
—

W2A Funnel Proxy

—

Google W2A Cohort Revenue

—
—

Ad Attribution

Resolve every registered user back to the AppsFlyer-attributed channel, campaign, ad ID and creative file. Forward lookup (ad → users) and reverse lookup (user → ad) over the last 30 days.
real data
—
—
—
—

Best Channel

—

Best Ad

—

Best Creative

—

Channel Leaderboard

—

Ad Leaderboard

—

Creative Leaderboard

—

Campaign Leaderboard

—
—

Creative Performance

—

Ad ID Performance

—

Channel → Business User IDs

Exact user-source mapping from fact_user_attribution. Open a channel to list the business user IDs acquired by that channel.
—

Reverse Lookup — User → Ad

Attributed Users

—

Ad-User Grade Cohort

Which educational segment do ad-acquired users belong to? Maps iOS ad installs from AppsFlyer BigQuery Push to the onboarding profile (grade level), and exposes per-user email, business ID and AppsFlyer ID with full filtering.
BigQuery Live
—
—
—
—

Grade Distribution (ad users)

—

Grade Cohort by Channel

Installs / matched / fill rate / grade counts (by first-touch channel)
—

Grade Distribution Breakdown

Aggregated by the selected dimension — grade counts and share (uses the top date range; first-touch attribution)
—

User Detail (Grade / Email / Business ID / AFID)

Per-user detail of ad-acquired users; filter by channel, grade, email and keyword
—
Aggregate
—
Worst of 8 checks
Green / 8
—
Yellow / 8
—
Red / 8
—

Check Matrix

—

Active Anomalies

—

BigQuery Cost — last 24h

—
—
Local registry only Actions here update the Sovi.AI incident registry only; they do not pause ads, submit appeals, or change live ASA / Google / TikTok resources. View button impact guide

Policy Incident List

—

Incident Maintenance

Response Advice

—

Advertising Automation Control

Read live Google Ads values and validate approved changes through the official validate-only API. No live delivery changes are applied.

GOOGLE VALIDATE ONLY
Production safety mode

The server forces validateOnly=true on every Google Ads mutate request. Permissions, ownership, snapshot freshness, account allowlists, and budget limits are checked without applying changes.

Generate Google Ads dry-run

Latest change preview

No controlled actions yet.

Action queue

—

Language Settings

Personal API Keys

Use API keys to programmatically query marketing, attribution, and growth datasets via REST API.
—

Open Data API Quick Start & Documentation

Supports Bearer token or X-API-Key HTTP headers for authentication.
View OpenAPI Spec
📊 GET /api/v1/data/raw-metrics

Query raw daily metrics with filtering and direct CSV export.

?start_date=2026-08-01&end_date=2026-08-31&format=csv
📥 GET /api/v1/data/export

Universal dataset export for channels, countries, campaigns, and attribution.

?dataset=channels&format=csv
🚀 GET /api/cockpit & /api/daily-report

Core cockpit and daily report APIs accessible via personal API Key.

Authorization: Bearer pci_ak_...
Python Code Example:
import requests
import pandas as pd

API_KEY = "pci_ak_your_personal_key_here"
headers = {"Authorization": f"Bearer {API_KEY}"}

# 1. 读取渠道投放指标 JSON
url = "https://mysovi.ai/apps/ua-command-center/api/ua/channels?start_date=2026-08-01&end_date=2026-08-31"
res = requests.get(url, headers=headers).json()
df = pd.DataFrame(res["data"]["channels"])
print(df[["channel", "spend", "af_installs_bq", "cpi"]])

# 2. 直接下载 CSV 格式原始数据
csv_url = "https://mysovi.ai/apps/ua-command-center/api/v1/data/raw-metrics?format=csv&start_date=2026-08-01"
csv_data = requests.get(csv_url, headers=headers).text
with open("sovi_metrics.csv", "w", encoding="utf-8") as f:
    f.write(csv_data)

Data Sources

—

Connection Configuration

Fill in the required configuration for each channel. Secret fields will not be echoed; leave blank to keep current value.
It is still recommended to use Cloudflare Worker secrets in production. Saved here to D1 as a fallback for Cloudflare page configuration, used to fix missing_credentials and let sync tasks read the config.
—

AI Model

—

TikTok Business OAuth

Creative Assets in R2

Pull ad-platform creative media into Cloudflare R2 and expose stable public URLs via sovi-ua-command-center-assets.mysovi.ai.
—

Recent Audit Log

—

Add User

Note: Currently all added users are granted standard access. Only @gmail.com or Google Workspace emails are supported for Google SSO login.

Users

—

System Overview

Sovi.AI UA Command Center is an automated BI system that integrates data from mainstream ad channels and attribution platforms to support daily business operations.

8 大操作说明书聚合传送门 (8 Core Guides Bento Grid)

EX
经营驾驶舱 (Cockpit)
面向管理层与增长负责人的 30 秒高管决策大盘,高度聚合核心 KPI,指导高频预算执行。
UA
UA 买量表现 (UA Performance)
细分渠道、国家、计划与素材维度,提供媒体自称与 MMP 归因的偏差比率,精确定位算力。
UQ
用户质量与激活 (User Quality)
衡量留存率、激活天数 and 核心动作转化。监控高粘性交互,防范机器或低质虚假买量。
MR
变现表现 (Monetization)
聚合 Superwall 漏斗、购买经济学和套餐占比。扣除渠道分成与税率,反映真实实收。
LV
投资回报周期 (LTV & ROAS)
观察 Cohort 财务回收周期与 LTV 拟合曲线。严格按时区归集,精确算清回本天数。
W2
网页跨端归因 (Web W2A)
专门对齐 H5 落地页广告花费、点击,并跨端关联 App 内的首购。对齐归因防预算漏损。
DQ
数据质量监控 (Data Quality)
一站式运行 8 项关键数据校验指标的健康盾。秒级感知数据断流、订单重复和金额偏差。
PI
政策事件与资源预警 (Policy Incidents)
说明账号、素材、关键词等资源如何通过接口只读监控,并列清 Generate Advice、Move 等按钮是否会影响线上投放。

工具与对账百科 (KPI Reference & FAQ)

指标口径字典 (KPI Reference Manual) ▶
指标名称 数据来源 计算方法 指标意义 常规范围
Total Spend 各广告平台 API 实时回调 SUM(spend_usd) 反应实际获客花费 $1k - $20k / 天
CPI 广告平台 Cost + MMP Installs Spend / AF BQ Installs 衡量真实平均获客价格 $2 - $6 / 安
Paywall CVR Superwall / AppsFlyer SDK first_purchase_users / pay_page_show_users 评估产品内付费墙转化效率 5% - 15%
ROAS (D0/D7/D30) D1 orders net_price + daily cost Cumulative Cohort Revenue / Cost 核心回本指标 D0: 1-3%, D30: 25-45%
🚦 数据质量状态说明 (Data Quality States) ▶

系统会在“经营驾驶舱”和“数据质量”模块中展示不同的状态标识,含义如下:

正常 (Green)
指标处于健康阈值内,无需干预。例如:归因覆盖率 > 95%。
警告 (Yellow)
指标偏离最佳范围但未造成严重断层。建议持续观察或手动定位原因。
异常 (Red)
指标严重异常,通常意味着 API Token 过期、数据流水线断流或归因发生断裂。需立即排查。
未知 (Unknown)
系统尚在等待底层数据聚合,或该指标的前置依赖条件(如对齐延时窗口)尚未就绪。
数据同步与时区标准 (Data Sync & Timezones) ▶

数据同步与聚合标准:

平台默认每天夜间/凌晨自动定时拉取各渠道的最新数据(业务日历统一对齐为 香港时间 HKT / UTC+8)。

回溯机制 (Backfill): 每晚定时同步任务不仅拉取前一天,还会对过去 7 天的历史成本与归因订单流水进行全面校准重刷,抹平媒体归因延迟引起的账期偏差。

资源状态监控与按钮影响 (Resource Watch & Button Impact) ▶

可以通过接口做统一资源状态监控,建议第一阶段全部采用只读模式。系统可从 Moloco、TikTok、Meta、Google Ads、Apple Search Ads、AppsFlyer、GA4 / BigQuery 等数据源同步账号、Campaign、Ad Group、Ad、Creative、素材文件、ASA 关键词、W2A 关键词 / UTM / 落地页等对象,统一归一为在投、停投、禁用、拒登、受限、审核中、未知等状态。

AI 预警可以基于状态变化和业务异常触发:账号被封、活跃资源突然 0 消耗、高消耗计划被拒登、关键词点击或安装断崖、素材同步失败、W2A 关键词漏斗断裂、API 数据延迟、政策高危词命中等。命中后先生成 Policy Incident 或预警,不直接改平台。

页面动作 调用接口 是否影响线上平台 说明
KPI / 筛选 / 搜索 / RefreshGET 本地查询否只过滤或刷新 Sovi.AI 数据。
Generate Advice无写入否只在当前页面套用建议文案,不保存,不提交申诉。
Save IncidentPOST / PATCH 本地事件否只写 D1 事件台账和审计,不调用广告平台。
Move · Appeal Ready / ResolvePATCH 本地事件否只推进事件状态、跟进时间和本地投放状态字段。

查看完整 Policy Incidents 操作说明 →

常见问题咨询 (FAQ Helpdesk) ▶

1. 为什么我的账号无法登录?

系统目前仅支持通过 Google SSO 授权登录。如果您未在授权组织白名单内,请联系系统管理员 sbin 申请加入白名单。

2. 媒体自称转化数 and AppsFlyer BQ 安装数为什么总有细微差别?

媒体(如 TikTok、Facebook)使用的是各自的专属自归因模型(Self-Attribution),会倾向于夸大自己的战绩;而 AppsFlyer 采用的则是去重归因模型。指标看板提供的“归因偏差率”旨在帮助我们定量评估并发现超出正常范围(通常为 ≤10%)的数据异常风险。

—