AI for data and analytics
Connecting AI to your data, and checking the numbers it gives back.
AI business intelligence
What the AI in Power BI, Tableau, Looker, ThoughtSpot, Omni, Sigma, Hex, Metabase and Count does, which to choose, and how to check its answers.
AI dashboards
What AI dashboard generators in Claude, Power BI, Looker and Omni really do, how to build one where every number traces to a query, and a two-minute check.
AI data analyst
What an AI data analyst does well, where realistic benchmarks show it failing, whether it can replace an analyst, and how to review its work like a new hire's.
AI for data analytics
Set up Claude for a data team: the context it needs, official connectors for Snowflake, BigQuery and dbt, and five data analytics workflows with prompts.
AI report generator
How AI report generators in Claude, ChatGPT and Power BI work, a project setup where every number is footnoted to a query, and the sentences to check first.
Best AI tools for data teams
The 9 best AI tools to run alongside Claude on a data team, ranked, with a winner, what each gets wrong, and the setup for your team. Checked October 2026.
BigQuery MCP
Google's BigQuery MCP server: its six tools, how to connect it to Claude, how to block writes with IAM, the 3,000-row cap, and how to check the answers.
Budget variance analysis
Budget variance = actual − budget. Sign conventions for revenue and costs, a worked price and volume split, flex vs static budgets, and budget vs actual traps.
Click-through rate (CTR)
CTR = clicks ÷ impressions. Google Ads, Search Console, Meta and email count both differently, which is why your CTRs don't match. Examples and benchmarks.
Conversational analytics
What conversational analytics is, the two ways chat-with-your-data tools build answers, where they guess at definitions, and a two-minute check for each answer.
Conversion rate
Conversion rate = conversions ÷ visits. Why Google Ads, GA4 and your store disagree, lead conversion rate by source, and a worked example with checks.
Cost per acquisition (CPA)
CPA = ad spend ÷ conversions. Which conversions count, and which platform claims them, is why Google Ads, Meta and your CRM disagree. Worked example included.
Customer acquisition cost
CAC = sales and marketing cost ÷ new customers. Which costs and customers you count can halve or double it. Blended and paid CAC, payback and a worked example.
Databricks MCP
Databricks' managed MCP servers (Genie, DBSQL and more): how to connect them to Claude, the Unity Catalog grants to set, the risks, and how to check answers.
dbt MCP
dbt's official MCP server, remote and local: its tools, how to connect it to Claude, which defaults to change before you do, and how to check the answers.
Forecast accuracy
Forecast accuracy = 1 − |actual − forecast| ÷ actual. MAPE, WMAPE and bias on a worked sales forecast, why MAPE breaks near zero, and why to snapshot forecasts.
Google Analytics MCP
Google's GA4 MCP server: set it up in Claude Desktop or Code, grant read-only access, know its quotas and thresholds, and check the numbers Claude reports.
Postgres MCP
Postgres has no official MCP server and the reference one has a SQL-injection hole. What to use instead, how to connect it to Claude, and how to lock it down.
Return on ad spend (ROAS)
ROAS = conversion value ÷ ad spend. Four choices inside it are why your ad platform and finance team report different numbers. Break-even table included.
RFM segmentation
RFM analysis scores customers on recency, frequency and monetary value. Quintile scoring worked through, the named segments, NTILE SQL and the traps.
Search term analysis
Read the Google Ads search terms report: find wasted spend, add the right negatives, find new keywords, and read impression share lost to budget and rank.
Semantic layer for AI
Why AI answers about your data need a semantic layer, what goes in one, how dbt, Snowflake, Looker and Power BI compare, and how to connect yours to Claude.
Snowflake MCP
How Snowflake's managed MCP server works, how to connect it to Claude step by step, which roles and grants to set, and how to check what it returns.
Support ticket analysis
Turn Zendesk or Intercom ticket data into numbers product and leadership trust: contact reasons, median reply times, escalations, reopens and a heatmap.
Text-to-SQL
How text-to-SQL works, what Spider 2.0 and BIRD really show in 2026, why semantic layers matter, and prompts and a two-minute check for AI-written SQL.