- Practically AI
- Posts
- 🧠AI Prompting Masterclass but for Data Analysis
🧠AI Prompting Masterclass but for Data Analysis
Using MOBY + GOOSE Frameworks developed by Logan Brown
We all know that data is king. But accessing and interpreting that data can be a major challenge. In a recent interview, Logan Brown, a Senior Product Manager at Triple Whale, shared a masterclass on how to leverage AI for data analytics [00:08, 00:17]. The secret isn't just using AI; it's knowing how to ask the right questions. Logan introduces two simple frameworks to help you move from ambiguous questions to getting actionable insights.
The 2 Prompting Frameworks for Data Analysis
To get better outputs from AI, you need better inputs. Logan shares two frameworks he uses: Moby and Goose.
The MOBY Framework [01:11] This is your foundation for clear, effective data prompts.
M - Measurable: Be specific. Instead of asking for your "best ads," ask for ads with the "highest ROAS (return on ad spend)" or "lowest CPA (cost per acquisition)" [01:39]. Define what "best" means to you.
O - Obtainable: Understand your AI's limitations. If you only provide two weeks of data, you can't ask for a reliable two-month forecast [02:43, 02:56]. Know what the AI can realistically do with the data you've given it.
B - Bounded: Set clear boundaries. Specify the exact time range for the data and, just as importantly, how you want it grouped (e.g., by day, by week, by month) [03:15, 03:20].
Y - Yielding: Have a goal. Know what you want to get out of the data before you even start writing your prompt [03:45].
The GOOSE Framework [01:17] This framework builds on Moby but adds one crucial element: Organization.
While the Moby and Goose frameworks share many similar concepts (like being specific and setting boundaries), Goose adds a focus on a "structured and logical flow" for your prompt.
Logan's key technique for this is writing what he calls "pseudo-SQL".
Instead of writing a long, narrative paragraph, you structure your prompt in a logical,-query-like format. This "no-code SQL" approach makes your intent incredibly clear to the AI.
Here’s an example
Logan provides a perfect example of the Goose framework's "Organization" principle [05:01]. Instead of a messy, single-sentence prompt, he structures it like a query:
He also shows an example where he uploads a dataset to an AI, first asking it to confirm it understands the columns and data types, and then prompting it to build a specific line graph comparing revenue for specific products over time [06:48, 07:53].
Here are some rapid-fire pro-tips from Logan to get the most out of your AI data analysis:
Confirm Understanding: Before you ask a single question, prompt the AI to "review the files and confirm you understand them" [06:56]. This ensures it has correctly identified your columns, time periods, and data types.
Be Specific with Output: Don't just ask for "insights." Tell the AI how you want the answer. For example: "Provide me with a bulleted list of insights" [15:01] or "Show me a line graph" [08:05].
Guard Against Hallucinations: AI can be convincingly wrong. To combat this, instruct it to "use data point references in your insight responses" [15:28]. This forces the AI to back up its claims with actual data.
Fact-Check the AI: A clever way to check the AI's work is to take its analysis, start a new chat, upload your original data, and ask the AI to fact-check the analysis it just gave you [17:34, 17:40].
Chunk Your Analysis: Don't overload a single prompt. If you want trend analysis, anomaly detection, and a regression analysis, break those into separate, focused prompts [28:36].
Use a Persona: Start your prompt with a role: "Act as a world-class data analyst..." [27:38]. This helps set the context for the AI and often improves the quality of the response.
Advance your team’s AI Fluency
Corporate AI Training → tailored for teams Gemini Mastery Course → join the waitlist |
Was this post useful? |
Until next time,
Kushank @digitalSamaritan
Reply