Artigo versão em português

Birdie Assistant is an AI application that generates text responses solely based on consumer feedback. It comprehends natural language and creates text-based interactions in a conversational context by drawing knowledge exclusively from user input. This approach ensures that its responses remain firmly rooted in the users' perspectives and experiences, delivering contextually appropriate responses based on the information contained within its designated dataset.

Navigate to the Feedback tab, then hover over the "Ask me a question" tooltip in the screen's bottom-left corner to load the assistant.

The Birdie Assistant becomes operational upon data upload, and you can access all the familiar filter features available across the platform. These filters allow you to dissect your data and pose specific questions within contextualized datasets.

The Assistant tool is quite simple. On the top, you will find three icons with the following features:
- Reset Chat
- Copy Conversation
- Settings

💡Pro tip #1: By default, the assistant will load the entire data set without any filter applied. Relying on the data categorization, a previously defined topic structure, or saved views, you can extract the maximum potential of the Birdie Assistant using them as filters to narrow down specific contexts.
​
💡Pro tip #2: The assistant relies on a concept known as a "context window," which defines how much information it can use to generate responses. We carefully choose a subset of feedback that fits within this context window to uphold accuracy and efficiency based on the assistant's preferred settings. This approach guarantees that the assistant can deliver insightful responses grounded in pertinent information, all while respecting the context window's constraints. This method balances performance and responsiveness, ultimately improving the user experience. Applying the proper filters to narrow the context window is key to getting accurate responses.

The choice between the faster and slower models depends on your context window size and the associated advantages. The faster model excels in speed and accuracy, operating within a smaller context window, enhancing its processing efficiency. In contrast, the slower model can handle a larger volume of feedback, but this comes with a higher risk of false positives, given the increased information it considers for generating responses.
​

💡Pro tip: The assistant is configured to use the faster model and random sample for optimal performance by default. If the amount of feedback you analyze is larger than the context window can handle, try narrowing down your search by combining filters like Feedback Intention, Sentiment, Rating, Topics, or any relevant Custom Field.

You can utilize this option for comparative analyses, enabling the assistant to reference specific frameworks when creating comparisons. For instance, if you wish to compare points related to a Company ID, you can select Company ID as the reference point to guide the assistant in generating relevant insights.

Due to its limited context window, the assistant's strength lies in exploration and hypothesis generation rather than quantification, although it might be useful to determine trends. To delve deeper into specific issues and uncover actionable insights, use the chat feature for more details. Additionally, leverage the Topics classification and Saved Views, complemented by filters, to review and quantify the feedback in Analytics or Dashboards.

Building on the previously discussed best practices, you can now ask refinement questions to uncover actionable insights. Here is a list of tested prompts to assist you in enhancing your analysis:

1) Finding Opportunities

[Part 1] Act as a PM at [Company Name] and produce an analysis of the top 5 opportunities based on negative feedback. First, create a table with each opportunity, name each opportunity in the context of the feedback (use neutral nouns), and generate a definition, bring a feedback quote to support each.

Ask follow-up question immediately, no need to clear the chat

[Part 2] I want to know more about “XXXX”, "XXXX", and “XXXX”. Break these opportunities into the top 3 sub-opportunities brought by consumers for each (bring 9 in total), name each sub-opportunity in the context of the feedback, and generate a definition, bring a feedback quote to support each.

​2) Rank frequent topics that generate feedback in a predefined set

Create a table with the most frequent topics for discussion. For each topic, give me a definition in the context of the feedback, the frequency percentage, and a feedback quote that supports the definition.

​3) Root cause analysis
I want a root cause analysis for [TOPIC] issues/inquiries. First, start with generating a short executive summary in the context of the feedback in one short paragraph. Second, bring the list of the root causes. Bring only the top 3.

4) Produce a detailed analysis of a topic of interest
​Part 1: Produce an analysis of the most frequent triggers for [Insert Topic Issues]. First, create a table with each trigger, name each trigger in the context of the feedback, and generate a definition of the trigger in the context of the feedback, its frequency percentage, and a feedback quote to support each trigger.

Part 2: Act as a Product Manager at [Insert your Company Name] and thoroughly analyze these topics. Produce a new table with each trigger, an executive summary of the current overall status, a bullet point of the main issues, and point out potential solutions. Finish up with an action plan to address these issues.

5) Compare the evolution of topics over time

[Fields to use selection: Posted At]

Act as a Feedback Analyst for [Company Name] and analyze the evolution of problems related to [Topic] from [First period] to [Second period]. What problems increased their incidence in [Second period] compared to the previous period?

6) Configure a weekly digest

Create a weekly digest, mentioning what main themes emerge from the user, including the percentage of all feedback related to each theme, and add samples of what users talk about each one. Keep it under 2000 characters.

7) Kano Model Analysis

Perform a Kano Model Analysis. Give me a table where: rows are the specific features or issues in text format, limited to the top 20 features; columns must be: Must-Have, One-Dimensional, Attractive, Indifferent, and Reverse; and each cell should be marked with X if selected.

8) Top 5 Requested Features table

What are the top 5 requested features mentioned in the feedback list in the last month? Create a table with each request, its frequency, and a feedback quote where the request is mentioned.