Three questions stand out. Only the first is backed by results I've run so far. The other two are well supported by the columns but still need to be computed. ## 1. Which AI use cases deliver the most value, and do productivity, accuracy and satisfaction agree? This is the question behind the "which use cases to prioritise" decision, and the purpose breakdown already gives a first answer: - **Office Work** has the highest average productivity (5.95), but only 11 users, so it is a thin sample. - **Writing** (5.54, 49 users) and **Customer Support** (5.53, 22 users) follow. - **Coding** is the largest group (70 users) but has the lowest productivity (4.91) and the lowest satisfaction (5.81) of any named purpose. - **Accuracy** barely separates the purposes, ranging only from 3.30 (Coding) to 3.60 (Customer Support). Productivity and satisfaction carry most of the signal. - Research has the highest satisfaction (6.41) but mid-table productivity (5.41). So the purpose that feels best is not the one that measures most productive. - 13 users have a blank `AI_Purpose` and the lowest productivity (4.63). That is a data-quality issue worth resolving before ranking anything. The gaps between purposes are under one point on a 10-point scale, and the groups are small. I would treat these as directional, not conclusive. ## 2. Does heavier or costlier AI use actually pay off? This is the "is a monthly subscription justified" question. It would compare `AI_Usage_Hours_Per_Day` and `Monthly_AI_Cost` against `Productivity_Score`, `Time_Saved_Hours_Per_Week` and `Satisfaction_Score`. It could show whether returns flatten beyond some number of hours, and whether paid users really outperform free ones. I haven't run it. Several of these columns are stored as text (usage hours, cost, time saved), so they need casting first. Because the data is observational, any correlation shows association, not that AI caused the gains. ## 3. Who benefits most, and which tool-and-audience pairings are worth recommending? This is the "students vs professionals, and at which education levels" question. It would cut productivity, satisfaction and `Would_Recommend` by `User_Type`, `Education_Level`, `AI_Tool` and `AI_Purpose`. The pre-aggregated gold tables for these combinations already exist. The interesting part is whether the best tool differs by purpose, for example one tool leading for Coding but not for Writing. With 300 rows spread across about 16 tools and 17 education labels, many cells will be very small. Education labels also look inconsistent, so they may need cleaning first. If you tell me which one to pursue, I can run it next.