A guide to an AI-built analytics dashboard

The build (minimum viable version)

Step 1: Describe the problem and get AI to write your data script—2–3 hours

Open your AI assistant and describe the manual process: ​“I currently export data from [platform name] to track [your metric]. I want a Python script that pulls this data through the API instead and calculates [metric].” If your platform has API documentation, paste the relevant URL into the chat.

Be specific about field names. Cinca ran into problems because his platform had three separate date fields—subscription date, list join date, and unsubscribe date—and vague prompts pulled the wrong one. Name the exact fields you want, not the concept. ​“I want the field called subscribe_​date, not just ​‘the date they signed up.’ ”

Quick version: Pick one metric and build only that. Cinca started with subscriber churn—not his whole reporting suite. A focused first build is easier to debug and gives you a working result faster.

You’re doing it right if: The AI produces a Python script with your platform’s API URL visible in the code and the specific field names you mentioned. If it writes generic pseudocode with placeholders, push back: ​“Write the actual API call for [platform name], using the endpoint at [URL from docs].”

Step 2: Deploy it to Google Cloud—3 hours

Paste your Python script into the chat and ask: ​“Give me step-by-step instructions to deploy this script on Google Cloud so the output is accessible through a web browser.” Follow the instructions literally, step by step, and paste any error messages back into the chat.

The AI will guide you through creating a Cloud Run service or App Engine instance, uploading your code, and getting a working URL. You do not need to understand what’s happening at every step—treat it like following a recipe.

Quick version: Skip Google Workspace authentication and custom subdomain for now. Share the raw Cloud URL with teammates while you confirm the data is accurate. Authentication and a clean URL come later.

You’re doing it right if: You can paste the Cloud URL into a browser and see output—even if it’s raw JSON or plain text. That means your script is running in the cloud. A formatted table comes in the next step.

Step 3: Add a web interface—2 hours

Ask the AI: ​“Create a simple HTML frontend for this script that displays the data in a table. Include date pickers so I can filter by date range.”

Paste the HTML it generates back into the AI and ask it to wire the frontend to your Cloud-hosted backend so the table updates when you change dates.

Quick version: An HTML table with date filters is the whole goal here. Do not ask for charts, CSV export, or visual styling yet. Cinca’s working dashboard is interactive tables—that is enough to replace manual reporting and prove the system works.

You’re doing it right if: You can change the date range in the browser and the table updates with different numbers. If the dates don’t filter correctly, paste the exact behavior you’re seeing (“I set the start date to Jan 1 and the table still shows all data”) into the AI and ask it to fix the date-handling logic.

Step 4: Validate against your own data—1 hour

Pull the same metric manually—the way you’ve always done it—for a date range you know well. Compare those numbers to what the dashboard shows. If they match, you’re done. If they don’t, tell the AI: ​“I expected [X], but the dashboard shows [Y] for [date range]. Can you check the code for what might cause that?”

Quick version: Test one date range you’ve already calculated by hand. Cinca’s debugging method was simple: describe the wrong result and ask the AI to trace through the code. You don’t need to understand the code to do this—describe the symptom and let the AI diagnose it.

You’re doing it right if: The dashboard number matches your manual number for the test period. A small rounding difference is acceptable. A 10%+ gap means a field or calculation logic is wrong—keep iterating.

You’re done when…

You can open the dashboard URL in a browser, select a date range, and see your key metric calculated automatically—with numbers that match your manual calculations. The system should run without you touching it again.