The fifth level is about the numbers you run a business with. How a metric differs from a plain number, how revenue grows, where visitors leave, who comes back, what a customer brings over their whole lifetime, and which products actually feed the store. Nine lessons — each ends with a decision you can make on Monday.
Every lesson is a template with demo data: ten minutes to complete, one skill that stays. Open it in the Templates section of Analytics Agent and press “Use this template” to take the lesson on your own data.
Not Every Number Is a KPI
Shows: three tests for a metric: does it move from your actions, do you understand what moves it, do you trust the data; three to five metrics, each with a goal and one owner; “empty” numbers are correct and useless — the customer counter grows in the worst month.
You get: you stop steering by numbers that cannot fall and pick a few that actually work.
Revenue and the Average Check
Shows: revenue = number of checks × average check; a two-factor breakdown to the cent shows what exactly moved — the checks, the check size, or both; the check splits again into “items × price”. One month grew while checks were falling — and the breakdown sees it before decisions.
You get: the question “why did revenue change” gets exact arithmetic instead of opinions.
Conversion: The Steps of a Customer
Shows: step conversion is those who got through divided by those who entered; the overall number hides the address of the leak, while a site sliced by devices finds a 39-point gap on one step; a point on the thinnest step is worth more than ten on the healthiest one.
You get: you know which step to fix first and what it will give — not “improve the site in general”.
Retention: Who Comes Back
Shows: the retention curve — the share of one group of customers month after month; the shape is always the same: a steep drop in the first months, then the plateau of the loyal core — the number to move; the acquisition channel decides who stays: organic holds 2.5× more than coupon.
You get: “will customers come back” stops being a guess — and you see which acquisition door feeds you for the long run.
Cohorts: Customers Grouped by First Purchase
Shows: everyone who started in the same month is a cohort; the overall average can hit records while nothing changes for anyone: the share of repeat orders reached an all-time high in the store’s worst month. Compare cohorts at the same age — month 3 against month 3.
You get: you see what really changed — the audience or the behavior — not just “everything grows / everything falls”.
Reading the Cohort Table
Shows: the triangle where “rows are lives, columns are peers”; the empty corner is youth, not missing data; small rows swing on their own — quote them only with a size; hot starts cool down: the year’s best start settled into the middle by month three.
You get: the cohort table is an argument of the “you cannot explain this by season” level — and you can read it.
What a Customer Is Worth Over the Years
Shows: customer value = average check × frequency × lifetime; acquisition pays off when cost is below value: one door returns 2.3× its cost in the first month, another returns 33.5 cents per dollar; young customers are “not worse — they are not finished yet”.
You get: you know how much you can really spend to acquire one customer — instead of a rule-of-thumb guess that is usually three times more optimistic than the truth.
Churn: Where Customers Go
Shows: “left” is a definition derived from your own data rhythm: three silent months mean churn if the longest gap in the store is 44 days; the two diseases — “left” and “gone quiet” — are broken down by revenue; wake the recently quiet first — bringing them back is worth 2.4× more.
You get: you catch churn earlier and know whom to win back and whom to let go.
ABC Analysis: The Vital Few
Shows: a sorted product list with your own threshold: A — the top up to 80% of value, B — up to 95%, C — the tail; the flat curve is the finding itself: here 25 of 40 products bring 80.9% of revenue; slices by revenue and by margin disagree on 11 of 40 products — the disagreement is the most interesting part.
You get: attention goes where the money is, automation goes to the tail; which value to rank by is your decision, not arithmetic.