One warehouse for ads, orders and CRM,
and an AI analyst who answers in numbers
Most businesses have the data and no picture: ads in one dashboard, orders in another, CRM in a third, each telling a different story. We put it into one database, make the numbers agree, and give you an analyst, human and AI, who answers questions with SQL instead of opinions.
What we connect
Meta Ads, Google Ads, TikTok Ads and TikTok Shop, LINE OA, Shopify, WooCommerce and custom stores, amoCRM, KeyCRM, HubSpot and Google Sheets, GA4 and PostHog, Telegram order channels, marketplaces, and your ERP for real cost of goods. Each source gets a raw layer and a daily sync with backfill, so history is never lost.
What you get
A single order book that agrees with your bank, not with a platform’s attribution. In one project we rebuilt the entire history of a sales channel from a Telegram chat: 571 orders, parsed 571 of 571, and found that the analytics platform had been seeing about 15% of real purchases.
Dashboards in money. Spend, revenue, ROAS, CAC, LTV, retention, repeat-purchase cycle, product and bundle economics. A market-basket analysis of 568 orders once showed that family-size carts were 36% of orders and 48% of revenue, which changed the whole bundle strategy.
An AI analyst in Telegram. “Top three cities by revenue last month”, “which campaign brought repeat buyers”, “margin by product line with real cost of goods”. Answers with numbers from the mart, with the SQL shown.
Monitoring. Competitor ads and prices, marketplace listings, distributor price undercuts (184 of 218 events caught with zero false alarms in one deployment), reviews, search trends. Alerts land in Telegram.
Unit economics and research. CAC, LTV, churn and payback models on your real data; market sizing and competitor teardowns run by research agents and checked by a human. We have built these for a trading app, a supplements brand and a game-server business plan with 113 sources.
How it runs
- Audit. What exists, what is tracked, what is wrong. One week.
- Model. Which questions the business must answer; the marts are designed from the questions, not from the sources.
- Connectors and sync. Raw layer, daily jobs, backfill, reconciliation checks.
- Dashboards and alerts. Built with the people who will use them.
- AI analyst and monitoring. Read-only role, guarded SQL, tested on recorded questions before going live.
Stack
PostgreSQL 16, Python, FastAPI, APScheduler, platform APIs, dbt-style SQL marts, Metabase or custom Next.js panels, PostHog, GA4, Claude, Telegram Bot API, Docker, daily backups.
Packages and pricing
Find out what your analytics actually sees and what it misses, with a fix list.
- Pixel, GA4, CAPI, UTM audit
- Orders reconciled against platform numbers
- Fix list for your developer or us
- Dashboard of the gaps
One database for ads, orders, CRM and site events, with dashboards that agree with each other.
- Connectors: Meta, Google, TikTok, Shopify, WooCommerce, CRM, GA4, PostHog
- Daily sync with backfill
- Marts: orders, spend, cohorts, products
- Dashboards: ROAS, CAC, LTV, retention
- Alerts to Telegram
- Documentation and handover
Ask questions in Telegram, get answers with numbers. Plus monitoring of competitors, prices and distributors.
- Everything in Warehouse
- AI analyst with guarded read-only SQL
- Competitor ads, prices and reviews collectors
- Price-undercut and anomaly alerts
- Market-basket and segment analysis
- Monthly review with a human analyst
Prices are starting points for typical scope. The exact number comes with a written plan after a 30-minute call; it does not change once agreed.
Structure, documentation and tests ship with every build, so any developer, or a person with an AI assistant, can continue without us. If you want us to keep making changes, support and development by our team start at $500 a month.
FAQ
Can the AI analyst hallucinate numbers?
It can only run read-only SQL on a prepared data mart, one query at a time, with forced limits and timeouts, and it shows the query it ran. We audited our own SQL guard and closed eight bypasses before trusting it with live data. The answer is as good as the mart, which is why we build the mart first.
What tools do you use?
PostgreSQL as the warehouse, Python connectors to the platform APIs, dbt-style marts, dashboards in your tool of choice (Metabase, Looker Studio, a custom Next.js panel), PostHog or GA4 for site events, Claude for the analyst.
We are small. Is this overkill?
Start with the tracking audit. For a shop with a few hundred orders a month the audit alone usually pays for itself by finding the orders the platform does not see.