AI for analysts
Analysts are one of the highest-ROI use cases for AI right now. Most of analysis is moving messy questions into well-structured queries, then translating numbers into a story stakeholders can act on. AI accelerates both ends — the SQL and the storytelling — while leaving the actual analytical judgment to you.
What AI handles well
SQL queries you can trust
The problem. You need a query that answers a specific question. Writing it correctly takes 20 minutes. AI gives you something that runs but might be subtly wrong.
What AI does. Use the SQL-trust prompt. AI explains its query step by step, flags edge cases (NULLs, duplicates, joins gone wrong), and suggests sanity checks before you commit to the result.
Data interpretation that has a take
The problem. You ran the query. You have the numbers. Your stakeholder asks "what does this mean?" You shrug and say "looks like X went up."
What AI does. Use the interpret-patterns prompt. AI surfaces 2-4 plausible explanations, ranks them by likelihood given context, and suggests the follow-up data that would discriminate between hypotheses.
Sanity-check your own analysis
The problem. You built a model and want to ship it. But you're too close — you might be missing obvious questions a skeptical reader would ask.
What AI does. Use the what's-missing-from-analysis prompt. AI flags confounders, selection bias, effect-size-vs-significance, and what a smart reader would push back on.
Cohort analysis structure
The problem. PM asks for a cohort analysis on retention. You can build the query but the cohort definition could go five ways. Which slice tells the real story?
What AI does. Use the cohort-analysis prompt. It surfaces the right cohort dimension, the metric definition (with edge cases), the chart shape, and what the result would MEAN for the business.
Hypothesis testing for product questions
The problem. PM asks "is feature X driving retention?" You could run a regression, do a cohort comparison, or build an A/B test. The right approach depends on the question.
What AI does. Use the hypothesis-test prompt. Translates the business question into the right test, surfaces confounds, and recommends a cheaper smell-test before the rigorous version.
Chart recommendations from data shape
The problem. You have data. The default chart in your tool is bar. You're not sure if that's right for this story.
What AI does. Use the chart-recommendation prompt. Given data shape + audience + the story you want to tell, AI suggests the right chart and what to highlight or strip.
Stakeholder briefs from technical analysis
The problem. You did the analysis. Stakeholders want a 1-page summary. Your draft is too technical. Your edits dumb it down past usefulness.
What AI does. Use AI to translate technical findings into stakeholder briefs. Lead with the takeaway, support with key evidence, end with the action implied. Skip the methodology unless asked.
Survey design that doesn't produce vanity stats
The problem. Marketing wants a survey about brand perception. Default questions ("how likely to recommend?") produce numbers that everyone agrees mean nothing.
What AI does. Use the survey-design prompt. Limits questions to ones that'll actually inform a decision. Catches leading questions before they bias the results.
Your AI stack
Start with the foundation. Add specialized tools as the work calls for them.
Foundation LLM
Specialized add-ons
Prompts ready to use
Get started in 30 minutes
Pick an LLM and load your data dictionary
15 minClaude Project with: your warehouse schema, common joins, business definitions, weird data gotchas. Any future query prompt is pre-loaded with context.
Run the SQL-trust prompt on your next ad-hoc query
10 minDon't take the AI-generated query at face value. The prompt asks AI to explain its work and flag edge cases. Use it the next time someone asks for a quick number.
Run the what's-missing prompt on a recent analysis you shipped
5 minPick something you delivered to a stakeholder. See what AI flags as missing. You'll either learn something or confirm the analysis was solid. Both are useful.
Common mistakes
Trusting AI-generated SQL without testing edge cases. AI writes queries that run. Whether they're CORRECT requires you to think about NULLs, duplicates, and the specific joins. Always sanity-check.
Letting AI interpret data without business context. AI surfaces patterns. Whether they MATTER depends on context AI doesn't have. Ask AI for hypotheses, not conclusions.
Recommending charts AI suggests without considering audience. AI optimizes for clarity. Stakeholders sometimes need a chart that's clearer to THEM, even if it's technically less optimal.
Pasting raw customer/transaction data into public AI. Use enterprise tiers, anonymize identifiers, or skip.
Using AI to confirm analyses you've already decided on. AI is a yes-machine if you frame the prompt that way. Use it to stress-test, not to validate.