Diagnostic Analytics

Saturday sold 95 cold coffees when every other day sold between 36 and 46. Nisha's first reaction: "Why?" Her barista says it was hot. Her manager says it was the new poster. Both sound reasonable, and at most one of them is the main reason. Diagnostic analytics is how you find out which story the data supports.

The rule

To find why, break the number down and compare it to a normal case.

A diagnosis has three moves:

1. Confirm the change is real. Check for a duplicated import, a changed definition, or a till counted twice before hunting for causes.
2. Drill down. Split the surprising number by the dimensions you have: hour, item, branch, customer type, payment method. Find where the extra actually sits.
3. Compare and test causes. Set the odd case beside a normal one, and check each suspected cause against the data. A real cause should explain this gap and be absent where the gap is absent.

For Saturday, a drill-down by hour might show:

The extra cups are almost all between 11 and 3. Heat alone would not usually start and stop that sharply. A street fair outside from 11 to 3 would.

The shortcut

Ask "Where exactly is the difference?" before asking "why". Keep splitting until one slice holds most of the gap, then look for what was different about that slice.

The trap is correlation without checking: two things happening together does not prove one caused the other. Always ask what else changed at the same time.

Drills

Answer fast — trust the rule.

Remember

• To find why, break the number down and compare it to a normal case.
• Shortcut: ask "where exactly is the difference?" before "why?"
• Trap: treating two things that move together as cause and effect.