Snowflake Data Analyst

Raw data hides problems that distort results. You cast types, parse dates, handle nulls and blanks, standardize codes and text, remove duplicates, and package the result as views that analysts can reuse.

• Casting types and parsing dates
• Nulls, blanks and default values
• Standardizing text and codes
• Removing duplicates
• Reshaping with PIVOT and UNPIVOT
• Analysis-ready views

You build A cleaned, documented view over a messy raw table.

Complex questions need layered queries. You structure work with common table expressions, use window functions for ranking and running totals, filter on windows with QUALIFY, and combine results with set operators.

• CTEs and subqueries
• Ranking, running totals, lag and lead
• QUALIFY for filtering window results
• Set operators and conditional aggregation
• Recursive queries for hierarchies
• Date and time analysis

You build A set of queries answering six layered business questions with windows and CTEs.

JSON arrives from applications and devices, not from tidy tables. You store it as VARIANT, address nested values with path notation, flatten arrays into rows, and keep such queries fast enough to use.

• VARIANT, OBJECT and ARRAY types
• Path notation and casting values
• FLATTEN and lateral joins
• Parsing nested JSON records
• Making semi-structured queries faster

You build A query set that turns nested event JSON into a flat, typed analysis table.

A query that runs can still be wrong. You read error messages, find joins that multiply rows, spot null-handling traps, use the query profile to find slow steps, and check results against totals you trust.

• Reading and fixing common errors
• Fan-out joins and wrong row counts
• Null traps in filters and joins
• Using the query profile to find slow steps
• Checking results against a known total

You build A debugging log of four faulty queries with the cause and fix of each.

Most analysis needs are already covered by built-in functions. You work with string, date and time zone, statistical and conditional functions, use approximate functions for very large counts, and learn to find the right function quickly.

• String and regular expression functions
• Date, time and time zone functions
• Aggregate and statistical functions
• Approximate counting for large data
• Conditional and conversion functions
• Finding the right function in the documentation

You build A function cookbook of twenty worked examples drawn from one data set.

When built-in functions fall short, you write your own. You create user-defined functions and table functions in SQL, JavaScript and Python, decide when a function is worth writing, and test and share them safely.

• When a function is worth writing
• SQL user-defined functions
• JavaScript and Python functions
• Table functions that return rows
• Sharing and securing functions
• Testing a function and checking its speed

You build Three user-defined functions that clean or score data, each with tests.

Every analysis begins with an honest summary. You profile data first, then describe it with aggregates, distributions, trends, segments and cohorts, and check that each summary really reflects the data beneath it.

• Profiling data before analysis
• Aggregates and distributions
• Trends over time
• Segments and cohorts
• Percentiles and outliers
• Checks that a summary is right

You build A descriptive report of one data set with each figure reconciled to its source.

Knowing a number moved is not knowing why. You break a change into its components, compare segments and periods, weigh correlation against cause, and write hypotheses with the evidence for and against.

• Drill-down and decomposition
• Comparing segments and periods
• Correlation and its limits
• Finding the driver of a change
• Recording hypotheses and evidence

You build A diagnostic note explaining a revenue change with evidence for each cause.

Some questions look forward. You forecast time series, detect anomalies and classify records using Snowflake’s ML functions, read prediction intervals, and decide when a result is not trustworthy enough to act on.

• Forecasting a time series
• Anomaly detection
• Classification with ML functions
• Training data and evaluation
• Reading forecast intervals
• When a prediction should not be trusted

You build A forecast and an anomaly report for one metric, with their limits stated.

A correct analysis that no one understands changes nothing. You turn a business question into requirements, build charts and dashboards in Snowsight, pick views that suit the audience, and share results on a schedule.

• Turning a question into requirements
• Charts and dashboards in Snowsight
• Choosing views for the audience
• Sharing results and scheduling refresh
• Telling the story with numbers
• Handing data to BI tools

You build A Snowsight dashboard answering a stated business question for a named audience.

You take one business question from raw files to a presented answer: load and clean the data, query it with advanced SQL, analyze what happened and why, add a forecast, and present it to its audience.

• Framing the question and the data needed
• Loading, cleaning and modeling
• Analyzing what happened and why
• Adding a forecast and its caveats
• Presenting to the business

You build A complete analysis from raw data to dashboard, with notes on method and limits.