Using Customised GPTs to Supercharge your Team

We used to spend a significant amount of time writing and refining requirements, ensuring that all necessary edge cases were covered in our acceptance criteria. Despite this effort, we often missed obvious ones and would still receive many questions from developers and QA during development—or worse, encounter observations, bugs, and blockers right before or after release!

Keeping track of all aspects of a complex or large-scale system is extremely difficult, especially in a startup environment where documentation may be incomplete. This is where I introduced a customised GPT.

Initially, I was simply looking for a tool to capture and provide complex product knowledge on demand. However, as I began experimenting with the Custom GPT, I realised it could do far more. The more background information or “knowledge” you provide, the more powerful and versatile it becomes—often in surprising ways.

What the GPT can be used for

A GPT has a wide range op applications, but I found it particularly helpful in preparing for development and improving the accessibility and clarity of product information.

Here are some of the things I found it most useful for:

• Generating User Stories with complete BDD acceptance criteria from a natural language description of a change or new feature.
• Retrieving system rules, user types, and access control details on demand (also automatically included in BDD mentioned above).
• User guides and detailed descriptions of user journeys.
• Internal and marketing documentation, and knowledge sharing.
• Structure and content for presentations.
• Brainstorming new features.

The GPT Knowledge Base

When creating a GPT, you can provide it with context when answering questions. This context may include documents, spreadsheets or API Specs. By building a comprehensive knowledge base in the GPT, it increases the usefulness and accuracy of the tool.

Here is what I included in the Knowledge base, each bullet point as a separate document:

• A description of each feature, any interactions between them, and any known limitation or future plan for it.
• Rules governing access control, user types, and any limitation on how each of them can use the system features.
• A structured representation of the data model or API spec.
• Screenshots of main user journeys.
• A template of the structure in which you want outputs such as user stories with acceptance criteria and requirements documents.

These items help teams keep current development aligned with future plans while ensuring adherence to product design rules and limitations. It further produces requirements in a standardised structure, helping with communication across stakeholders, especially the development team.

With an advanced and reasonably up to date knowledge base, it would even be able to identify existing bugs, improvements or “low hanging fruit” to tackle whilst working on a related piece of functionality.

Tips for creating a GPT

• Use one document for both a product or feature spec and the GPT knowledge base. Replace the version in the knowledge base after each spec update, or after each release.
• Assign a dedicated person to maintain the knowledge base to prevent errors and inconsistencies.
• Start with the product team and gradually expand to other relevant teams.
• Make sure to exclude information of your GPT from teaching the Chat GPT model, and if necessary, get approval from your company before using the tool.
• Work with engineering on integrating this tool with theirs, or identifying another tool which may be more useful across product and tech teams.

Finally, a word of caution

Ultimately, this is another tool that requires maintenance. Keeping an up-to-date and comprehensive solution can be time-consuming, so I recommend focusing on the most critical challenges your team faces first and expanding as you identify new uses for it that might be worth the effort.

Our GPT transformed how we prepare for development and respond to questions and requests. However, it is crucial to remind users that the GPT may provide inaccurate information, and proper reviews, human creativity, and expertise are still essential for both inputs and outputs. It is also important to reassure team members who may feel threatened. The GPT is designed to assist—not replace—human ingenuity.

For more information and ideas on how ChatGPT could help your team, visit their website here.