DataSlice ru🍕 Telegram (in Russian)

Metrics and their hierarchies

Start with Basics, then open Industries: ready-made metric trees for 16 verticals with real-company breakdowns.

What a metric is and why hierarchies exist

A metric is a number measuring the state of a product or business: revenue, conversion, retention, response time. One number on its own means little — what matters is how metrics connect and which one leads.

That's why metrics are arranged into a hierarchy. At the top sits a single North Star that best reflects the product's value. Below it are the drivers it's made of and the operational levers teams move directly. At the foundation are counter-metrics that stop anyone from gaming the headline number at quality's expense (hello, Goodhart's law).

Metrics get laid out in two different ways — as a tree and as a pyramid. They are not the same thing, and the difference is visible below.

Tree versus metric hierarchy

A tree breaks one metric into factors and terms (Profit = Revenue − Costs). You follow cause-and-effect levers — which gear to turn.
A hierarchy (pyramid) groups metrics by abstraction level and audience. Not "what it's made of" but "who cares about what" — to align metrics across team levels.
Business
revenue, base growth — for leadership
Product
conversion, average order, retention
Interface
CES, CSI, task success
Platform
response time, errors, load

North Star Metric: one metric at the top

A North Star Metric is the single metric that best captures the value users get from the product. It sits at the top of the tree (level L0), and everything else is its decomposition.

The point of an NSM isn't to measure everything with one number — that's impossible. The point is focus: a company with twenty equal priorities has none. The NSM answers "if we could move only one number, which one?".

One caveat: "one metric" doesn't mean "the only metric". Counter-metrics always live next to it — guardrails that stop you from gaming the headline number at the product's expense.

Three types of North Star

Amplitude's classification: the NSM type follows from what users actually value the product for. Picking the type matters more than picking the formula — it shapes the entire tree.

Transaction
What it measures
The number (or value) of completed transactions

Value appears at the moment of the deal: the order arrived, the ride happened, the night was booked. Count completed transactions, not created ones: a completed one is where demand, supply and operations all met.

Examples: Airbnb — booked nights · Uber — completed trips · Ozon and Wildberries — GMV · DoorDash — completed orders
When this is you: Your product is this type if you match two sides or sell goods, and money arrives with every deal — as commission or margin.
Covered in industries:

What makes a North Star good

It reflects user value
It grows when the user is better off, not when you squeezed them harder. Revenue is a poor NSM: it also grows from raising prices.
It leads revenue
Money arrives with a lag; the NSM should move earlier and predict it. A metric that moves in lockstep with revenue is useless for steering.
It's measured often
Daily or weekly. A metric computed once a quarter gives no feedback for decisions.
Teams can move it
There must be a visible chain from a team's action to the metric. Otherwise it's a metric for the report, not for the work.
It's not a vanity metric
Not "total registrations": cumulative counters always go up, even as the product dies.
It works with counter-metrics
Every NSM can be gamed against the product. If you can't think of a guardrail for it, you just haven't thought hard enough.

Input metrics vs output metrics

An NSM is an output: a result nobody can touch directly. You can't "do" completed orders — you can improve delivery time, selection and conversion, and orders grow on their own.

Input metrics are what a team moves directly this week. In a metric tree they live at the bottom: L4–L5. The distinction is practical — you set goals on outputs and plan work on inputs.

Decomposing a North Star down to level 6

The tree below breaks a marketplace's GMV down from the North Star (L0) to operational levers (L5). Click any node to open its metric card with a formula and SQL. Notice how the nature of the nodes changes on the way down: measurable results at the top, things teams do with their hands at the bottom.

Click a node to open the metric card with its formula and SQL.
Counter-metrics: Return rate · Promo margin · Seller churn

Goodhart's law: why one metric is never enough

When a measure becomes a target, it ceases to be a good measure. That's not theory but daily practice: give support a response-time goal and you get fast useless replies; give the feed a watch-time goal and you get clickbait.

So an NSM never lives alone. At the bottom of the tree sit counter-metrics (guardrails): return rate next to revenue, cancellations next to orders, opt-outs next to pushes. They're not meant to grow — they're meant not to get worse.

More in the lesson on Goodhart's law →

See it on real trees

Every industry in the "Industries" tab is an NSM of one of the three types, decomposed down to level 6.

Attention
Transaction
Productivity

Ready-made frameworks

AARRR (Pirate Metrics)
Acquisition → Activation → Retention → Referral → Revenue
The user lifecycle funnel. Handy for products and startups: where people leak out on the way to value and money.
HEART (Google)
Happiness · Engagement · Adoption · Retention · Task success
For product quality and UX: each letter gets its own goal, signal and metric.
Profit Tree
Profit = Revenue − Costs → then break each branch into levers
The financial tree for unit economics: it shows which lever moves profit and how.

Building a tree for an unfamiliar product

Identify the archetype by how the product makes money — and the tree's shape follows.

A glossary of metrics by type

The basic vocabulary: which metrics exist and how they're computed. Grouped by lifecycle stage — acquisition → engagement → retention → revenue — the same logic as the AARRR frame (Acquisition → Activation → Retention → Referral → Revenue).

Acquisition
CAC
marketing spend / new customersthe cost of acquiring one customer.
Conversion
target actions / visitsthe share of visitors who reach the target step.
CTR
clicks / impressionsthe click rate of an ad or element.
ROAS / ROMI
ad revenue / ad spendthe return on advertising spend.
Engagement
DAU / MAU
daily / monthly uniquesthe active audience; their ratio is stickiness.
Frequency
sessions / userhow often a person comes back.
Depth
actions (or minutes) / sessionhow intensely the product is used per visit.
Retention
Day-N retention
returned on day N / cohortthe share still around after 1/7/30 days.
Churn
lost in period / base at startchurn; the flip side of retention.
NRR
cohort revenue now / a year agonet revenue retention, counting expansion and churn.
Monetization
ARPU / ARPPU
revenue / all (or paying) usersaverage revenue per user or per paying user.
LTV
ARPU × average lifetimehow much a customer brings over their lifetime.
LTV / CAC
LTV ÷ CACacquisition payback; healthy above 3.
GMV
the value of all dealsgross volume for marketplaces and platforms.
Quality & counter-metrics
NPS / CSAT
survey indiceswillingness to recommend / satisfaction.
Complaint / return rate
complaints (returns) / actionsa quality guardrail: it catches gaming of the headline metric.
p95 / p99 latency
response-time percentileshow bad it is for the unluckiest — technical reliability.
Where statistics comes in: a metric went up — check it wasn’t by chance (A/B and the t-test); revenue means lie because of long tails — use the median (distributions and skew); optimizing a single number — keep counter-metrics (Goodhart’s law).
Breakdowns like this live in the channel

Data analytics in plain words: how to count metrics, how not to fool yourself in an A/B test, what gets asked in interviews and how this site gets built. The channel is in Russian.

Open the Telegram channel