AI Build Week — Day 2: From Payments to Predictions

Today we’re connecting Retool to Stripe’s OpenAPI to build a real-time “Churn Radar” dashboard — a fast, practical way to spot at-risk customers before they churn :chart_with_upwards_trend:


:clock10: Live at 10 AM PT — streaming right here
:speech_balloon: Say hey in the comments if you’re tuning in! We’ll be chatting and answering questions live.

---

:brain: What We’re Building

@angeliklaboy (Developer Advocate at Retool) will show how to —

  • Use Stripe billing signals (failed invoices, expiring cards, cancellations) to build a subscriber table enriched with AI-generated risk scores
  • Transform raw payment data into a polished dashboard that flags churn risk
  • Link directly to Stripe’s Customer Portal for quick retention actions
  • Iterate and refine the build using Assist + AppGen

:file_folder: Download today’s resources
:wrench: Register for the rest of the week →


:speech_balloon: How to Participate

  • Ask questions or share your takeaways in this thread
  • Post screenshots or snippets of your own build
  • Tag your project with app-gen to earn badges + bragging rights :sports_medal:
  • Poll + discussion below :backhand_index_pointing_down:

:light_bulb: Have a prompting tip to share?
Drop it in the Prompt Pro Challenge — whether it’s a clever shortcut or a funny fail, it helps the community get smarter about building with AI.

← Back to the full AI Build Week schedule

1 Like

Thanks to everyone who joined us live (or async!) for Day 2 of ai-build-week-2. We built a real-time Churn Radar dashboard using Stripe + Assist and tackled some thoughtful technical questions along the way. Here's a recap of what came up:


:red_question_mark: Do we have visibility into the source code of the apps that Assist creates?

You can always inspect the JavaScript and SQL queries Retool writes. Retool apps themselves are built on an abstraction layer, so while you don’t see raw source code like in a typical React app, you do get full visibility into the logic, queries, and components within Retool.


:red_question_mark: Can I migrate my custom Next.js codebase to Retool?

Definitely! You’ll need to manually recreate your UI using Retool’s building blocks, but for anything not natively supported, you can create your own custom components and plug them right in.


:red_question_mark: How do I know Assist won’t run incorrect or destructive queries on my database (or Stripe)?

Assist never runs anything automatically — by default, every query must be manually approved by a human before execution. If you’re confident, you can set certain queries (or even full resources) to be auto-approved in your settings.


:red_question_mark: Does Assist work with Workflows or Agents?

Not yet — currently Assist is scoped to app building. That said, we're actively working on expanding Assist to more areas of Retool, including Workflows and Agents in the future.


:red_question_mark: How is LLM behavior monitored and logged in terms of user security?

Retool offers comprehensive audit logging to track AI-related actions. These logs include inputs, outputs, timestamps, and metadata around queries generated or modified with Assist.


:red_question_mark: Do I need both OpenAI and Anthropic API keys on self-hosted to use Assist?

Only if you’re self-hosted. If you're on Cloud, Retool manages this for you. On self-hosted, you can add either an OpenAI or Anthropic key — both are supported, and having just one is fine.


:red_question_mark: What plans include Assist?

Assist is available for all Cloud plans. For self-hosted, you'll need to be on Edge version 3.283+ or Stable 3.284+, and you’ll need to enable Assist in Settings → AI with the necessary connectors toggled on.


:red_question_mark: Can Assist be used to improve existing apps?

Yes! Assist can explain how an existing app works, analyze query-component relationships, and suggest improvements. It doesn’t make any changes without your approval — everything is surfaced to the user for review.


:red_question_mark: How was the risk score determined in the demo? Was that AI-generated?

In the Churn Radar demo, the risk score wasn't determined by the AI itself — the underlying logic (like a scoring formula) was set up in the Retool app and Assist helped visualize it. In other words, the AI didn't invent the risk logic; it worked with data the app already had.


:red_question_mark: Where can I access the prompts used in this session?

You can find the sample prompts in the Day 2 Resources folder. Bonus: If you have your own tips (or funny fails), add them to the Prompt Pro Challenge thread!


If we missed your question, or you're watching async — drop it in the comments! :raising_hands: