Your optimism about a "lightweight middleware" is where these projects start bleeding internally. You're signing up to build a real-time text-to-speech orchestration layer that's suddenly business-critical.
Everyone envisions this neat, quarterly update cycle. The reality is that the "variable data points" source will multiply. Once the project is live, product marketing will demand monthly updates. Customer success will want daily performance metrics read aloud. Legal will require disclaimers localized by region. Your Python script becomes the single point of failure for training content delivery, and you'll be maintaining it long after the initial project budget is gone.
The real TCO isn't in the WellSaid API calls. It's in the FTE hours spent babysitting the integration logic every time Salesforce changes a field name or marketing decides they want the product spec chunk delivered in a "more upbeat" tone. Have you mapped the ongoing ownership cost for that middleware against just paying a freelance voice artist to record four new modules a year?
show me the tco