Project 01Marketing analytics
Marketing Dashboard
Every marketing channel, an in-house site audit and a monthly competitor report in one live dashboard, built by a marketer in Claude Code to replace a paid Semrush plan and a weekly report done by hand.
Live at Archetype Group, updated three times a day
- $2,000+saved a year by replacing Semrush
- 7channels in one live view
- 90automatic updates a month
01
At a glance
The whole project in three steps.
- The problem
Marketing data sat in seven platforms and a set of spreadsheets, reported by hand every week, with a paid Semrush plan for site health.
- What I did
Built one live dashboard in Claude Code: every channel, an in-house site audit, a weekly AI briefing and a monthly competitor report.
- The result
More than $2,000 a year saved on Semrush, 90 automatic updates a month, and two spreadsheet trackers retired.
Skills it shows
- Marketing analytics
- Marketing operations
- SEO and GEO
- AI in production
- Product ownership
- Product marketing
Tools
- Claude AI
- Python
- GitHub Actions
- Supabase
- Vercel
- Google Analytics 4
- Google Search Console
- Meta
- ElevenLabs
02
The challenge
Why the project had to happen.
Archetype Group’s marketing data lived in seven places: Facebook, Instagram, LinkedIn, Google Search Console, Google Analytics, PageSpeed and our email platform. Every report meant signing in to each one, exporting, and copying the numbers into a spreadsheet by hand.
Read the full context
The rest of the operation was just as scattered. Website health came from a paid Semrush subscription. Business requests sat in a separate Excel workbook for every year since 2023. Projects waiting to go onto the website and into our presentations were tracked in another sheet, and nobody had a regular view of what our competitors were doing.
The subscription was the visible cost. The bigger one was time: hours spent collecting numbers instead of acting on them, and reports that described last week by the time they were finished. Leadership needed one place to see what was working, and the team needed its time back.
03
How I did it
7 steps. Open any one for the detail.
August
Start with the data, not the screens
Stored every channel’s history before the platforms deleted it: 836 Facebook posts back to 2020.
How
Before drawing a single chart I checked how far back each platform lets you go. Instagram keeps 30 days, so anything older is lost unless you store it yourself. I set up a collector that saves every channel into one database and pulled the full history while it was still there: 836 Facebook posts back to 2020 and 210 Instagram posts back to 2018.
August
Design for the person reading it
Every view leads with one figure and its trend, in plain words.
How
People open a dashboard to answer one question: is it working? So every view leads with one figure and its trend, in plain words, with each platform in its own colour. Every number on the Home page comes from the same calculation as its own tab, so two screens never disagree.
September
Make it run without me
Moved collection to the cloud: three automatic runs a day, behind a company login.
How
The first version ran on my laptop. When the laptop slept through a weekend, two days of data were lost for good. I moved the pipeline to the cloud: it now collects, rebuilds and publishes three times a day, at 6:00, 12:00 and 18:00, behind a company login, and the Home page flags any source that falls behind.
September
Replace the paid tool
Built an in-house site audit: 114 checks covering every Semrush issue type.
How
The team relied on Semrush for its site audit. I built our own: a crawler with 114 checks covering every issue type in Semrush’s Site Audit, at the same thresholds, run every two weeks. Once a month an AI review reads the results and turns them into tasks for the team.
September to October
Turn data into decisions
A weekly AI briefing and a monthly competitor report for leadership.
How
A chart shows what happened; a team needs to know what to do next. Every Wednesday, Claude writes a briefing from that week’s data: what stood out, why, and the next steps. Every figure it quotes is checked against the data before it is shown. A monthly competitor report covers 13 international firms and local rivals for 22 country offices, and reaches leadership as a PDF by email.
October
Bring the spreadsheets in
Business requests and project approvals moved from Excel into the dashboard.
How
Once the team used the dashboard daily, the side spreadsheets moved in. Business requests, kept in a separate Excel workbook for each year, became one searchable list with contracts signed and their value. The project approval sheet became a six-step tracker. PR mentions and the project portfolio got tabs of their own.
November
Launch it like a product
Wrote and directed a launch film so the team would actually open it.
How
An internal tool only saves time if people open it. I researched how software companies launch products, then wrote, storyboarded and directed a launch film for the team, built frame by frame in code from the dashboard’s own design. It is at the top of this page.
04
See it
05
What it achieved
Replaced the Semrush subscription with an in-house site audit, saving more than $2,000 a year.
Ended the weekly routine of signing in, exporting and copying numbers across seven platforms. The dashboard now updates itself three times a day, about 90 times a month.
Gave leadership a weekly AI briefing and a monthly competitor report without adding a reporting job to anyone’s week.
Retired the business-request workbooks and the project approval sheet. Both now live in the dashboard the whole team can see, behind a login, with a review step before any change goes live.
Built in eleven weeks with more than 350 changes shipped, by a marketer working in Claude Code, with no development team.
06
What I took from it
Collect before you design. Instagram keeps 30 days of history, so data you do not store early is gone for good.
A tool only saves time if it runs without its maker. Moving collection off my laptop is what made the dashboard dependable.
An internal tool needs a launch like any product: if the team never opens it, it saves no one any time.