Utilizing modern tooling to support the Product Discovery process
Artificial intelligence (AI) has found its way into every aspect of our lives in a very short time. About a year ago, we started having a conversation about how to leverage AI to help in our day-to-day efforts working within the fields of Product and user experience to support product conceptualization. In this article, we explore how AI can streamline the creation of product Discovery documents.
Writing and maintaining product documentation can be a tedious and labor-intensive process. Maintaining consistency across documents and ensuring all necessary information is included can be challenging. It requires attention to detail, a deep understanding of the project, and the ability to communicate complex ideas clearly and concisely.
Our initial idea was to utilize AI to take part of the documentation load away from our Discovery engagements. To tackle this issue, we turned to AI to automate part of the documentation process. ChatGPT, an AI language model developed by OpenAI, can generate content based on predefined templates and guidelines by leveraging natural language processing (NLP) and machine learning algorithms. This automation allows us to significantly reduce the time and effort required to prepare documents, allowing teams to focus more on problem-solving, requirements creation, ideation, and iteration processes.
Over time, we learned that when using AI like ChatGPT to assist in your documentation process, ensure alignment with your intentions by asking questions you already have answers to, such as “What makes X topic relevant for my X audience?” or “Tell me best practices for X subject.” When you get junk results, course correct and refine responses to address misalignments or disagreements. This initial step is crucial in establishing a foundation of understanding between you and the AI, setting the stage for a more productive collaboration.
To start training the AI, add context, including client, user, or product information, and correct any summaries or interpretations provided by ChatGPT. Providing this context helps ChatGPT understand the specific requirements of your project. It ensures that its responses are tailored more directly to your needs, like prepping a user for discussing a certain topic and putting them in the right frame of mind before outright asking questions. This process of contextualizing the conversation improves the accuracy of the AI’s responses and enhances the overall quality of the generated content. After providing context, you can provide ChatGPT with a template outlining the document structure and key components.
Now we can ask ChatGPT to perform the task based on the template and context. For more, see the “Creating and Using ChatGPT Templates” section later in this article. By guiding ChatGPT through this structured approach, this method reduces the need for corrections and ensures that ChatGPT’s responses align with your intentions and objectives.
Our research experience directly inspired our approach, where priming users and mentally preparing them to talk about a certain feature was crucial. Have you ever tried to rapid-fire answer random questions that are not related? You are basically making your brain work extra hard! In research, it is important to prime users and mentally prepare them to discuss a specific feature. This puts them in the right mindset before asking specific questions and supports them more easily in context-switching, making it easier for them to articulate their feedback.
This is the same purpose for simplifying prep questions with ChatGPT. Aligning ChatGPT with your intentions can save you time and effort in the long run. We have streamlined this process by saving these prompts and reusing them like a script.
Templates for Discovery documents are easy to create. We ended up taking existing documentation and reducing that content into templates. Here is an example of our problem overview template:
Now, we can replace the variables and generate the document. We keep our variable lists stored with our client documentation.
Then, we ask ChatGPT to generate the problem overview. As with everything using AI, we read through the content to identify and correct errors and inaccuracies before we paste it into our official client documentation.
Tip: One tip that we have found most effective in this process is to tell ChatGPT to “Please don’t add any new ideas or concepts to this content.”
Summarize Meetings, Interviews, & Observations
Transcribe Video Recordings
We have found a lot of value in using AI meeting assistants. However, we have found it more accurate and effective to feed ChatGPT our meeting notes, 1:1 interview feedback, and user observation transcripts. Currently, most video meeting applications have their own AI transcript generators. If your product doesn’t offer this feature, ask the people on the call for permission to record the user or stakeholder interview, and don’t forget to press the record button.
If you are hosting a face-to-face workshop or interview, you will need to think ahead and have a device to record the meeting, ensuring that audio is picked up accurately. The clearer the audio, the better the transcript will be.
If you have a video recording of a meeting, interview, or observation, you can upload the file to an affordable, for-pay AI transcription service like Temi. You can utilize free methods, like using Google Docs in combination with Google Voice Typing and playing your videos into text. There are downsides to most free methods we have tried, like Google Voice Typing, which happens in real-time. We have found it more valuable to use a service like Temi, which can transcribe multiple videos simultaneously and doesn’t take long to complete. A service like Temi is the way to support both to save time and streamline processes.
Review Transcripts
As with everything with AI, you will want to review the transcripts for the accuracy of the content. Also, keep an eye on who the generator thinks is speaking and update the assigned speaker. We have seen much more value in using a service like Temi, being more accurate in identifying the speaker and making it easy to reassign who is speaking. When trying to cut costs and using Google Docs to read your audio into a document, all this becomes a manual task.
Summarize Transcripts
We then paste our transcripts into ChatGPT with the prompt: “Can you summarize the transcript of this video without adding any new content or concepts?”
Generate Content & Edit
We read through the summary and correct any inaccuracies as we paste it into our official documentation.
Bonus Method
After summarizing your interviews or observations for a single group or event, copy and paste them into a single prompt. We then start by prompting ChatGPT with questions to influence its response. For this example, we would ask, “What makes an analysis of a transcript effective and relevant for clients?” We can then follow that with a prompt: “Please analyze these transcript summaries to identify similarities and differences without introducing any new concepts.” Using this method will produce comparisons between transcriptions to highlight shared successes and points of friction and call out edge cases between conversations.
Warning
Ensure Compliance with NDA
Before using AI to automate the writing of product documents, it is crucial to understand and follow any non-disclosure agreements (NDAs) or confidentiality agreements with your client or business. This is crucial to ensure no proprietary or confidential information is inadvertently disclosed.
Review Content for Accuracy
As we mentioned several times in the article above, while AI can greatly streamline the documentation process, it is essential to double-check the generated documents for accuracy and completeness. While powerful, AI is fallible and can regularly produce errors or omissions, especially if you provide vague prompts.
Establish a Double-Double-Check Review Process
After using AI to draft content, we have started working as a team across the department and sending our content to team members for review. This extra set of eyes can help catch any errors or inconsistencies you may have overlooked. Once a colleague has reviewed the document, you should thoroughly review it to ensure that it meets the necessary standards of accuracy and completeness.
Conclusion
In our experience, AI has proven to help speed up content generation and review. When used correctly, AI tooling can help to speed up transcript reviews and summaries. Utilizing content templating within ChatGPT helps to speed up initial documentation drafting. However, it is important to be cautious when utilizing new tooling like ChatGPT for client, concept, and company security. You must establish and follow review processes to catch and call out any potential errors or inaccuracies in generated content, protect sensitive information, and adhere to legal agreements.
To learn more about how Callibrity can help your organization, email us at contactus@callibrity.com to schedule a time to connect.