Pattern Recognition

Hidden Signals

Patterns Are Everywhere

Every dataset tells a story, but most of the story is invisible to the naked eye. You might notice that sales dip in February, but did you notice that customers who buy Product A in their first order are 3x more likely to buy Product C within 60 days?

AI can hold an entire dataset in view simultaneously and spot relationships across thousands of data points. This is where AI analysis goes from convenient to genuinely powerful.

Taxonomy

The Pattern Type Catalog

Patterns in data fall into distinct categories. Knowing the types helps you ask AI the right questions and interpret results with confidence:

Seasonal patterns: Recurring cycles tied to calendar periods. Retail spikes in December. Gym memberships surge in January. Ice cream sales peak in July. Ask AI: "Is there a seasonal pattern in this data? Show me the same metric for the same month across multiple years."

Cyclical patterns: Recurring fluctuations not tied to a fixed calendar. Business cycles, economic expansions and contractions, product adoption curves. These are harder to spot because the period length varies.

Step changes: Sudden, permanent shifts in a metric. A new pricing model that moved average order value from $30 to $45 overnight. A policy change that cut support tickets by 40%. Ask AI: "Are there any abrupt level changes in this time series? When did they occur, and what could have caused them?"

Gradual drift: Slow changes that are invisible day-to-day but significant over months or years. Customer satisfaction slowly declining. Average response time creeping up. These are the most dangerous because nobody notices until it is too late.

Clustering: Groups of similar data points that form naturally. Customers who behave similarly, products with similar sales patterns, regions with similar demographics. Ask AI: "Are there natural clusters or groups in this data based on these variables?"

Absence patterns: Sometimes the most important pattern is what is missing. No sales on certain days. No support tickets from a region that should be generating them. Zero values where there should be data. Ask AI: "Are there any gaps, zeros, or missing data points that seem unusual given the surrounding context?"

Three Types

Trends, Outliers, and Correlations

Trends are directional patterns over time. Revenue is growing 5% month-over-month. Customer support tickets increase every Monday. Your email open rate has been declining since September.

Outliers are data points that don't fit the pattern. One customer spent 20x the average. One day had zero traffic when every other day had thousands. Outliers are either errors or the most interesting data points you have.

Correlations are relationships between variables. When ad spend goes up, conversions go up. When temperature drops, hot chocolate sales rise. Correlation doesn't mean causation — but it always means investigation.

Going Deeper

Ask "Why," Not Just "What"

Critical Warning

The Correlation Trap

Ice cream sales and drowning deaths both increase in summer. That doesn't mean ice cream causes drowning. Both are caused by heat. This is the correlation-causation trap, and AI can accidentally reinforce it if you're not careful.

Always ask AI: "Could there be a confounding variable here?" and "What would I need to prove this is causal, not just correlated?" This habit separates people who find patterns from people who find truth.