Optimizing Product Discovery with AI - Accelerating SaaS Development for Complex Financial Modules

In the competitive landscape of high-pressure SaaS development, launching sophisticated financial platforms under tight deadlines demands a transformation in requirement engineering. During our recent SaaS project, the team was tasked with navigating a rigorous eight-month delivery cycle.

To enhance project velocity and maintain high quality, we implemented an agile AI discovery framework. This approach shifts the focus from simple task execution to high-impact user outcomes. Here is how our AI-powered product management workflow functions across four key phases.

Step 1: Automated Requirements Gathering & Stakeholder Alignment

The goal of this phase is to align the cross-functional team on business objectives and high-level project scope before diving into technical implementation.

• The Workflow: Instead of traditional manual note-taking, we utilize Google Meet transcripts to capture every nuance of our requirements discussions.
• AI Implementation: We use Gemini AI to process conversational data into structured meeting minutes. This automated documentation captures core decisions, technical constraints, and action items, which are then centralized in Notion for easy access.

Step 2: Drafting Functional Specifications with Generative AI

Moving directly from discussion to coding often results in logic errors. A crucial prerequisite: all discovery discussion minutes, specs, and requirement documents are centralized and managed on Notion. 

This ecosystem makes analyzing and cross-referencing information with Notion AI much more seamless.

This "User Story Draft"  ("US Draft")  stage ensures we define a solid business baseline and identify complex edge cases early in the SaaS lifecycle.

• The Workflow: Product teams trigger Notion AI using a standardized prompt*, and feed it three primary inputs: the meeting minutes link, relevant Figma design frames, and existing software specifications.
• Automated Structure: The AI compiles a functional draft covering:
  • Overview: A concise executive summary of the story theme.
  • Why: The measurable business goal and user value, structured as: As a [persona], I want to [action], so that [value/emotional relief].
  • How: A reflection of the current user pain points (limited to three points) paired with a high-level solution concept.
  • What: Functional requirements focusing strictly on capabilities, alongside design reference links.
  • Specs to Verify: A checkbox list detailing critical variables requiring alignment (e.g., data models, master data integration).

*Our team has pre-configured a library of AI rules/prompts on Notion that can be quickly triggered on demand (e.g., for creating PRDs, User Stories, Release Announcements, etc.).

Step 3: Developing Implementation-Ready Agile User Stories

This transition converts the strategic draft into precise development and quality assurance specifications.

• The Workflow: By invoking the "Create User Story" command within Notion, an automated agent checks the draft against established specs and glossaries to enforce technical alignment.
• Output Standards: The system generates structured functional and non-functional engineering criteria. Critically, acceptance criteria are automatically written in an industry-standard Given-When-Then format, providing QA teams with instant test cases.

Step 4:  Human-in-loop Review & Refinement    

Finally, all AI-generated requirements are validated by senior product managers and engineers. We treat automated outputs as high-quality baselines for human expertise.

• The Workflow: Through collaborative Notion comments and daily discovery meetings, the cross-functional team evaluates the story against a core checklist: Is it value-centric? Are error handling and edge cases explicitly defined? Is it conflict-free against historical specifications? Once validated, the ticket shifts directly to "Refinement Ready".

Keys to Long-Term Adoption

Managing an automated discovery flow requires strict process discipline. We maintain a dedicated page to define global teamspace rules, leverage advanced large language models within Notion for high-resolution contextual processing, and track our custom prompts across a strict lifecycle (WIP, Testing, Live, Deprecated) to avoid the dilution of document quality.