Panel 1 of 4. The short version
Case study · Agentic AI and data
Morning Metrics
I set out to build a whole marketing team in one product.
Morning Metrics brings a business’s data from eight platforms into one place, then puts an AI agent to work on it: answering questions, finding what you missed, writing the reports and running jobs on a schedule.
- MCP server
- OAuth 2.1
- BigQuery
- ELT pipeline
- Multi-tenant app
The problem
Every morning I opened eight platforms to find out how the business was going, and their numbers never added up.
Google Ads, Meta, Microsoft Ads, GA4, Search Console, Merchant Centre, the email platform and the store’s admin. Add up the sales each one claimed and the total was more than the business actually made.
None of them was lying. Each counts a sale its own way, so people stopped trusting all of them and went with gut feel instead.
It is not only a marketer’s problem. Anyone who needs to know how the business is going hits the same three walls:
Nobody agrees on a number. “Revenue” means something different in every tab.
Every question is a project. Asking which search terms wasted money means an export, a pivot table and half a morning.
The answer comes after the decision. By the time the report is ready, the week it described is over.
Most businesses cannot afford an analyst, a reporting person and a strategist. I wanted to build all three.

What I built
Morning Metrics is a whole marketing team in one product, working from one set of numbers.
The analyst. An AI agent connected through MCP. Ask it anything about the business and it works the answer out from your own data.
The reporter. The 7am brief: yesterday in plain English, in your inbox before the first meeting.
The strategist. The app’s opportunity and action lists: wasted spend, missed sales and what to do next, each with a dollar value.
The dashboard is where they all sit: revenue, campaigns, products, customers and attribution, on a laptop or a phone.
Every number is worked out once, so the whole team agrees. The agent, the brief and the dashboard never give you two answers to one question.


How it works
Behind the scenes, four things happen every night.
Collect
Morning Metrics pulls the day’s data from every connected platform into one database, and keeps the full history in a data warehouse.
Collect → Define → Check → Serve → Collect again
- Collect. Morning Metrics pulls the day’s data from every connected platform into one database, and keeps the full history in a data warehouse.
- Define. It works out each number once, with every definition written down, so revenue means the same thing everywhere.
- Check. It looks for wasted spend, quiet email flows and missed sales, and puts a dollar value on each.
- Serve. The dashboard, the 7am brief and your AI agent all read the same numbers.
The AI agent
Connect Claude, or any AI agent that supports MCP, and the business gets an analyst who already knows every number.
It has 30 tools covering sales, ad campaigns, search, products, customers and social, and it can check the ad platforms, analytics and the store live. It does far more than answer questions:
Answers in plain English. “What did Meta really bring in last week?” It weighs what each ad platform claims against real store sales.
Finds what you missed. Products selling without ad support, searches your site does not answer, seasons starting in the next six weeks, products people buy together.
Explains its working. Ask how a number is worked out and it gives you the definition, the formula and the caveats.
Writes the report. A weekly summary, a channel review or a board update, pulled straight from the numbers.
Then you put it to work, like a member of the team.
Schedule a channel review every Monday, a stock check every night, or an alert when spend runs ahead of sales. Turn the way you like a job done into a skill, and anyone in the business can run it the same way. This is the whole marketing team I set out to build.
It can look, but never touch.
The agent is read-only: it can see your ad accounts and store but never change them. Access can be withdrawn at any time, and text from ads and shoppers is treated as data, never as instructions.
Most of the work was not the AI. It was making the numbers trustworthy enough for an agent to rely on.

Try it
Put your own marketing team to work.
See every number in one place, get yesterday in your inbox at 7am, and point your AI agent at the lot.

Want to see it on your own business’s numbers? Get in touch and I will plug them in for you.
morningmetrics.app
76 likes
Like this project





