Steps to build your own AI analytics dashboard

- Identify the repetitive report you want to automate: what data do you currently pull manually, how often, and what decisions does it drive? - Confirm API access: check that your marketing platform offers an API (most paid tiers do) - Choose an AI assistant: Cinca used Google Gemini; ChatGPT and Claude work the same way - Pick a hosting platform: Google Cloud is a reasonable default; your AI will suggest options based on your setup - Set the right expectations: the first version will not be perfect—plan to iterate

- Open a new conversation with your AI tool: treat it like talking to a technically fluent colleague, not a search engine - Describe your problem conversationally: explain what you currently do manually, why it's slow, and what you wish you could see—for example: "I pull subscriber reports by hand every week and calculate churn in spreadsheets. Is there a way to automate this with data from [platform] API?" - Ask the AI to confirm feasibility: it should tell you whether API access plus Python code can solve the problem - Ask for a step-by-step outline before any code: understand the shape of the solution first - Ask the AI to write the initial code: request Python scripts that pull data from your platform's API and process it

- Ask the AI where to host your code: request a comparison of options (Google Cloud, AWS, Render, etc.) and let it recommend based on your existing tools - Follow the AI's setup instructions step by step: ask it to walk you through the hosting setup, development environment, and deployment—not just the code - Set up authentication: Cinca used Google Workspace login so his entire team could access the dashboard with their work email and no separate login required - Deploy a first working version: get something running before refining it—a rough dashboard is better than a perfect plan - Connect a subdomain: give the tool a clean URL your team will actually remember and use

- Use exact field names from your platform's API: never say "the date"—say "the subscription date," "the list join date," or "the unsubscribe date" depending on what you mean - Test against known data first: run the dashboard against a period where you already know the correct numbers, then compare - Repeat context the AI might have lost: as Cinca put it, "It's better to repeat yourself and say too much rather than not enough"—restating field names and logic in every prompt costs nothing - Handle edge cases deliberately: what happens if someone enters a start date after the end date? What if a field is empty? Ask the AI to test those scenarios - Validate calculated metrics against manual calculations: especially any multi-step metrics like true cost per lead or adjusted churn rate

- Report unexpected results in plain language: tell the AI what you expected, what you got, and why it seems wrong—for example, "I ran this for March and got 47 subscribers, but I know we had around 200. Something's off." - Set up error logging when the AI recommends it: copy error messages exactly and paste them back into the conversation - Think of AI as a brain without hands: Cinca described it as having someone with a strong analytical mind who "doesn't have hands and eyes to look through and click"—your job is to feed it the information it needs to diagnose the problem - Start each new session by having the AI review existing code: context is lost when you close a conversation, so paste the current code and ask for a summary before making any changes - Test edge cases before you consider it done: date ranges, empty fields, overlapping cohorts—anything that might happen 1% of the time will eventually happen