Business and product metrics arranged by what feeds what. Every card carries a formula, the parent metric it rolls up into, and a worked example with numbers.
01
Money: from GMV to free cash flow
One ladder. Each line comes from the one above by subtracting a named cost. If a metric has no slot here, it is not a money metric.
GMVTurnover through the platform at buyer prices. Sellers' money, not yours.
less cancellations, refunds, fraud
Net GMVThe turnover that survived to fulfilment.
× take rate, or revenue recognised under your own model
RevenueCompany revenue. Gross for a reseller, net for an agent or marketplace.
less COGS: cost of goods, hosting, licences, delivery
Gross ProfitGross profit. Divided by revenue it becomes gross margin.
less other variable costs: payments, support, last mile
Contribution MarginWhat one order leaves behind before marketing.
less acquisition marketing
CM after CACCohort profit after paying for the cohort. This is where growth and bought growth separate.
less OPEX: engineering, brand, G&A
EBITDAOperating result before depreciation, interest and tax.
less depreciation, interest, tax
Net IncomeNet profit. The line a shareholder sees.
± working capital, less CAPEX
Free Cash FlowCash in the bank. Profit can be paper, cash cannot.
GMV
Gross Merchandise ValueEverything that crossed the counter, counted whole, before any deductionΣ price × quantity across paid orders
Turnover before deductions. A measure of the platform's scale, not its income.
→ rolls up into Net GMVShow exampleOrders in the month 1,200
Average order value $350
GMV = 1,200 × 350 = $420,000Orders later refunded are still in GMV. The deduction happens one line below.
Take rate
Commission rateOf the whole stack, the platform keeps the top coinRevenue ÷ GMV
The share of turnover the platform keeps. A rising take rate with falling GMV means sellers are leaving.
links GMV to RevenueShow exampleRevenue $42,000
GMV $420,000
Take rate = 42,000 ÷ 420,000 = 10%Use net GMV, otherwise refunds inflate the base and understate the rate.
Revenue
Top lineThe flow splits: a share stays with the platform, the rest goes to the sellerNet GMV × take rate, or Σ recognised revenue
The top line. In an agency model it is the commission, not the turnover.
→ rolls up into Gross ProfitShow exampleGMV $420,000
Refunds 5% −$21,000
Net GMV $399,000
Revenue = 399,000 × 10% = $39,900A reseller on the same data reports $399,000 of revenue and a huge COGS. The model moves the number tenfold.
COGS
Cost of Goods SoldThree streams leave the jar before any profit existsdirect cost of delivering the product
In SaaS: hosting, CDN, licences, the infrastructure that keeps the product running.
subtracted from RevenueShow exampleHosting and CDN $6,200
Licences and APIs $3,100
Product support $2,700
COGS $12,000
Gross Profit = 40,000 − 12,000 = $28,000
Gross Margin
Gross marginThe revenue column: the hatched base is cost, what stays above is margin(Revenue − COGS) ÷ Revenue
The ceiling of the model. SaaS holds 70-85%, marketplaces lower, retail lower still.
derived from Gross ProfitShow example(40,000 − 12,000) ÷ 40,000 = 70%Every new dollar of revenue brings 70 cents to cover everything else.
CM1 / CM2 / CM3
Layers of contribution marginThree cuts: each level takes its own slice offCM1 = Revenue − COGS
CM2 = CM1 − other variable costs
CM3 = CM2 − marketing
The numbering is internal. Agree it once and write it down, or two reports will never match.
→ rolls up into EBITDAShow exampleRevenue $40,000
− COGS $12,000
CM1 $28,000
− payments, support $8,000
CM2 $20,000
− marketing $12,000
CM3 $8,000
EBITDA
Earnings before interest, tax, depreciation, amortisationFour steps down: what you can stand on is left at the very endCM3 − OPEX
A proxy for operating efficiency. Not cash flow.
→ rolls up into Net IncomeShow exampleCM3 $8,000
− engineering $4,500
− brand and G&A $2,000
EBITDA $1,500EBITDA margin = 1,500 ÷ 40,000 = 3.75%.
FCF, Burn, Runway
Cash flow, burn, runwayThe bag feeds the months until it runs out: what is left is the runwayRunway = cash on hand ÷ monthly net burn
Runway is measured in months. Below 12 it stops being a metric question and becomes a plan question.
the last line of the money ladderShow exampleCash on hand $6,000,000
Net burn per month $500,000
Runway = 6,000,000 ÷ 500,000 = 12 monthsA 20% rise in burn costs two months of runway with no change in revenue at all.
02
Subscription: how MRR moves
In subscription, revenue is not a number but a balance of five flows. Read the composition of the movement, not the total.
Opening MRRThe base you start from.
plus New MRR and Expansion MRR: upgrades, seats, overage
less Contraction MRR and Churned MRR
Net New MRRNet growth for the period. Negative means marketing is pouring into a leaking bucket.
accumulated month over month
MRR → ARRARR = MRR × 12. A yearly projection, not a fact.
MRR
Monthly Recurring RevenueThe same coin lands every monthΣ normalised monthly value of active subscriptions
Annual plans divide by 12. One-off payments never belong in MRR.
Annual Recurring RevenueTwelve monthly bars under one bracketMRR × 12
The investor's language. Valuation usually starts as a multiple of ARR.
derived from MRRShow exampleMRR $41,000 × 12 = ARR $492,000
At a 6x multiple the valuation is ≈ $2,950,000
NRR
Net Revenue RetentionThe same cohort outgrows its own starting line(start + expansion − contraction − churn) ÷ start
Measured on a fixed cohort, new customers excluded. Above 100% means growth without acquisition.
the quality of MRRShow exampleCohort a year ago $100,000
+ expansion $18,000
− contraction $4,000
− churn $9,000
Total $105,000
NRR = 105,000 ÷ 100,000 = 105%New customers stay out, otherwise the metric collapses into plain revenue growth.
GRR
Gross Revenue RetentionThe bar can only lose height: the start line is the ceiling(start − contraction − churn) ÷ start
Capped at 100%. Shows the leak raw, with no upgrade makeup.
the pair to NRRShow example(100,000 − 4,000 − 9,000) ÷ 100,000 = 87%NRR 105% with GRR 87% means the base leaks while large accounts pay more. Concentration risk.
SaaS Quick Ratio
Growth efficiencyScales: new and expansion against contraction and churn(New + Expansion) ÷ (Contraction + Churned)
How much new MRR arrives per dollar lost. Below 1 the business is shrinking.
a read on Net New MRRShow exampleNew $15,000 + Expansion $6,000 = $21,000
Contraction $2,000 + Churn $5,000 = $7,000
Quick Ratio = 21,000 ÷ 7,000 = 3.0
Bookings, Billings, Revenue
Three different numbersOne contract, one invoice, revenue spread across the monthsBookings = contracts signed
Billings = invoices issued
Revenue = revenue recognised
Mixing them in one report is the most common way to fool yourself a quarter ahead.
feed Revenue with a lagShow exampleJanuary: a 12-month contract, $120,000
Bookings in January $120,000
Billings in January $120,000
Revenue in January $10,000
Deferred revenue $110,000
03
Unit economics: a single customer
The whole business folded into one user. If it is negative here, scaling only speeds up the loss.
ARPU
Average Revenue Per UserOne coin split across the whole crowd, free users includedRevenue ÷ active users
Counted over everyone, free users included. Falling ARPU with rising revenue means the base is being diluted.
→ rolls up into LTVShow exampleMonthly revenue $40,000
Active users 60,000
ARPU = 40,000 ÷ 60,000 = $0.67
ARPPU
Average Revenue Per Paying UserOnly the marked figures hold a coin: the rest are freeRevenue ÷ paying users
The pair to ARPU. The gap between them is the paid conversion rate.
Lifetime ValueA stream of coins from one customer, thinning month by monthARPPU × gross margin ÷ churn rate
Count it in margin, not revenue. Cap the horizon at 12 or 24 months.
→ rolls up into LTV/CACShow exampleARPPU $25 per month
Gross margin 70% → $17.50 of margin
Churn 5% per month
LTV = 17.50 ÷ 0.05 = $350Churn at 7% instead of 5% drops LTV to $250. The metric is brutally sensitive to its denominator.
CAC
Customer Acquisition CostMoney goes into the megaphone, one customer comes backacquisition spend ÷ new customers
Only new customers below the line. Media and growth-team salaries both go above it.
→ rolls up into LTV/CACShow exampleMedia spend $60,000
Growth team $18,000
Creative $2,000
Total $80,000
New customers 400
CAC = 80,000 ÷ 400 = $200
Blended vs Paid CAC
Blended and paidTwo funnels: the wide one counts everyone, the narrow one only paidBlended = spend ÷ all new users
Paid = paid spend ÷ paid-acquired users
Blended looks better and says less: organic hides what the channel actually costs.
a refinement of CACShow example1,000 new users, 400 of them paid
Blended = 80,000 ÷ 1,000 = $80
Paid = 80,000 ÷ 400 = $200A decision to scale a channel is made on paid, never on blended.
LTV / CAC
Value to cost ratioTwo columns side by side: what a customer brings against what they costLTV ÷ CAC
Below 1 the business pays for the right to operate. Above 5 usually means underinvestment in growth.
the verdict of unit economicsShow exampleLTV $350 ÷ CAC $200 = 1.75
A healthy SaaS benchmark is 3.0To reach 3.0 you either cut churn to 2.9% or CAC to $117.
CAC Payback
Payback periodCumulative margin crosses the CAC line in month elevenCAC ÷ monthly margin per customer
Measured in months and hits runway directly. It matters more than LTV/CAC when cash is short.
→ drives Free Cash FlowShow exampleCAC $200 ÷ $17.50 margin per month
= 11.4 monthsAt 5% churn half the cohort leaves before it pays back. Payback and retention are read together.
Magic Number, Burn Multiple
Efficiency of the dollar investedHow much new ARR came out per dollar put inMagic = ΔARR ÷ prior-quarter sales spend
Burn Multiple = net burn ÷ Net New ARR
Magic above 0.75 means press the accelerator. A Burn Multiple below 1.5 is acceptable.
the joint between growth and moneyShow exampleQuarterly ΔARR $900,000
Prior-quarter S&M $1,200,000
Magic = 0.75
Annual net burn $2,000,000
Net New ARR $1,500,000
Burn Multiple = 1.33
04
The funnel: AARRR
Five stages, each with its own metrics. Always work the stage where the leak is biggest in people, not in percent.
Share who reached valueThrough the door, three steps, to the star: that is activationusers who did the key action ÷ signups
The key action is chosen by correlation with day-30 retention, not by team taste.
→ rolls up into RetentionShow example10,000 signups
4,000 reached the first export
Activation = 40%
D30 among activated 38%
D30 among the rest 4%The 38% versus 4% gap is what makes that action the key one.
TTV
Time to ValueThe stopwatch measures the road from the start to the starmedian time from signup to the key action
Median, not mean. The mean breaks on the tail of people who came back six months later.
a driver of Activation RateShow exampleMedian TTV 18 minutes
Mean TTV 9 days
TTV under 10 min → D30 45%
TTV over 1 day → D30 12%
Step conversion
Per step and end to endA funnel of four levels, each narrower than the one aboveStep CR = exits ÷ entries
End-to-end CR = product of steps
End-to-end conversion hides where the break is. Keep both.
the structure of the funnelShow example100,000 visits
→ 10,000 signups 10%
→ 4,000 activations 40%
→ 600 payers 15%
End to end = 0.6%
People lost: 90,000 at the first stepThe worst percentage is step three, but step one is the fix: 90,000 people are lost there.
K-factor
Viral coefficientOne brings two, and the two bring the next onesinvites per user × invite conversion
Above 1 growth sustains itself. In reality it is almost always below.
Acquisition mixThree streams of different width enter the same doorshare of new users, CAC and retention by channel
A channel is judged on three: volume, price, cohort quality. Two out of three is an incomplete decision.
a cut of AcquisitionShow exampleSEO 40% of new, CAC $30, D30 38%
Paid social 35% of new, CAC $240, D30 14%
Referral 15% of new, CAC $12, D30 41%
Other 10% of new
LTV/CAC on paid social = 350 ÷ 240 = 1.46
05
Engagement: how the product is used
The layer between activation and money. It answers whether a habit exists at all.
DAU / WAU / MAU
Daily, weekly, monthly activesThree nested circles: the day inside the week inside the monthunique users with an activity event in the window
Write the definition of activity down. "Opened the app" and "completed an action" differ by multiples.
the base of every product metricShow exampleApp opened: DAU 12,000
Key action done: DAU 4,300
MAU 60,000, WAU 38,000, DAU 12,000Same product, two definitions, a 2.8x difference.
Stickiness
StickinessA month of squares with six marked: that is how many days people returnDAU ÷ MAU
How many days a month the typical user comes back.
derived from DAU/MAUShow example12,000 ÷ 60,000 = 20%
20% × 30 days ≈ 6 days a month
Messengers 50-60%
Social 30-40%
Utilities 10-20%
Sessions
Count, length, depthShort visits along one day, the clock counts eachsessions per user, median length, actions per session
A long session is not automatically good. A utility wants it shorter, content wants it longer.
a cut of DAUShow exampleSessions per active per day 2.4
Median length 3 m 10 s
Actions per session 7
After simplifying export:
length 1 m 50 s, sessions 3.1Sessions got shorter and more frequent. For a utility that is an improvement.
Feature Adoption
Feature uptakeIn the crowd few use the feature, and those few use it oftenbreadth = feature users ÷ actives
depth = uses per feature user
Two numbers per feature. Narrow and deep is a niche tool, wide and shallow is decoration.
explains RetentionShow exampleCSV export
breadth 9,000 ÷ 60,000 = 15%
depth 12 times a month
D30 with the feature 46%, without it 19%
Power User Curve
Distribution of active daysThe histogram dips in the middle and rises again on the righthistogram of users by 1-30 active days in a month
A smile with a right-hand hump means a habit core. A left peak only means the product is one-off.
a cut of MAUShow example1-3 days 60% of users, 18% of actions
4-10 days 28% of users, 34% of actions
20+ days 12% of users, 48% of actions12% of users produce nearly half the activity. Their churn hurts more than anyone else's.
Core Action
Core actionOne big button, pressed again and againcompletions per unit of time
The thing the user came for. It usually becomes the North Star.
→ rolls up into North StarShow exampleExports per week 85,000
Per active user 2.2
Actives doing at least one 41%
06
Retention and churn
Retention is a cohort curve, not a number. A flat plateau matters more than a high D1.
Retention D1 / D7 / D30
The classic cohortThe curve drops fast, then flattens into a plateau: that is the coreusers returning on day N ÷ cohort size
State the model: N-day, unbounded or rolling. The same data yields different numbers under each.
→ sets LTVShow exampleCohort of 5,000 signups
D1 2,100 = 42%
D7 1,150 = 23%
D30 800 = 16%
D60 760 = 15.2%
D90 750 = 15%The curve flattens near 15%. That plateau is the core LTV is built on.
Churn Rate
ChurnA bucket with a hole: what leaks out is the churnlost in the period ÷ base at the start
Separate behavioural churn (stopped using) from billing churn (cancelled).
the inverse of RetentionShow exampleBase on the 1st 1,600
Cancelled in the month 80
Churn = 80 ÷ 1,600 = 5%
Annualised ≈ 1 − (1 − 0.05)^12 = 46%5% a month sounds mild and means losing nearly half the base in a year.
Logo vs Revenue Churn
Customer churn and revenue churnTwo small customers leave on one side, one large on the otherLogo = lost customers ÷ customers
Revenue = lost MRR ÷ MRR
The gap shows who is leaving: small accounts or large ones.
→ rolls up into GRRShow exampleLogo churn 5.0%
Revenue churn 2.1%
→ small accounts are leaving
The reverse case:
Logo 1.5%, Revenue 6.0%
→ one large contract walked out
Customer Lifetime
Expected lifetimeAn hourglass: the customer months run through it1 ÷ churn rate
Only valid when churn is stable. On a young base it overstates lifetime by multiples.
Resurrected usersSomeone who left comes back through the same doorreturning after a dormant spell ÷ dormant users
The fourth flow in the user balance: new, retained, churned, resurrected.
a flow into MAUShow exampleMAU last month 58,000
+ new 6,500
+ resurrected 1,900
− churned 6,400
MAU this month 60,000The 2,000 gain rests on resurrected users. Without them MAU would be flat.
Cohort table
The retention triangleA triangle of cells: rows are cohorts, columns are agerows are cohorts by entry date, columns are age
Reading down a column shows product effect, reading across a row shows cohort quality.
the instrument for RetentionShow exampleCohort M1 M2 M3
January 42% 24% 19%
February 44% 27% 22%
March 47% 31% 26%Column M3 rises from 19% to 26%: the onboarding changes are working.
07
Perception and support
Survey metrics do not replace behavioural ones. Their place is early diagnosis and segment cuts.
Customer SatisfactionA closed ticket and a star ratingsatisfied responses ÷ all responses
Tied to one interaction: a ticket, a purchase, a release.
a cut of qualityShow example430 ratings of 4-5 out of 500 = 86%
Survey response 500 ÷ 1,800 tickets = 28%At 28% response the unhappy answer more readily. The level is biased down, the trend still reads.
CES
Customer Effort ScoreThe same box: heavy on one side, light on the otherrating of "it was easy to get my task done"
Predicts repeat purchase better than NPS in utilitarian products.
a driver of RetentionShow exampleMean score 5.8 of 7
Share 6-7 64%
Share 1-3 11%
Churn among 1-3 responders 14%/mo
Churn among 6-7 responders 2%
PMF Score
The Sean Ellis testFour in ten would be very disappointed without the productshare answering "very disappointed" without the product
Threshold 40%. Ask active users only, otherwise the number means nothing.
the gate before scalingShow example300 active users answered
Very disappointed 132 = 44%
Somewhat 120 = 40%
Not disappointed 48 = 16%
Analyst segment 61%
Student segment 18%PMF exists in one segment, not in the product overall. That segment is what you scale.
Contact Rate
Support loadOut of the crowd, only a few reach supporttickets ÷ active users
The direct cost of poor UX. It rises with COGS and falls once the top three causes are fixed.
→ rolls up into COGSShow example1,800 tickets ÷ 60,000 = 3.0%
Cost per ticket $4.20
Support spend $7,560 a month
Top cause = 38% of tickets
Fixing it = −$2,870 a month
FRT, Resolution Time
First response and resolution timeA stopwatch between the question and the answermedian and p90 across tickets
Report percentiles. A support average always lies in the company's favour.
a driver of CSATShow exampleFRT median 2 h 10 m
FRT p90 11 h 40 m
Resolution median 9 h
Resolution p90 52 h
CSAT when FRT under 1 h 94%
CSAT when FRT over 8 h 61%
08
Marketplace and platform
A two-sided market is measured symmetrically: demand, supply and the quality of the match.
Liquidity
Supply liquidityTen listings on the shelf, three of them soldtransactions ÷ listings posted
The share of supply that finds a buyer. The core health metric of a marketplace.
→ rolls up into GMVShow example3,400 deals ÷ 12,000 listings = 28%
Core cities 41%
Everywhere else 19%Below 20% a seller does not recover the effort of listing and leaves.
Fill Rate
Demand coverageRequests come in, most get matched, some are left hangingfulfilled requests ÷ all requests
The share of demand that got an answer. Liquidity seen from the buyer's side.
→ rolls up into Net GMVShow example8,600 of 10,000 requests = 86%
Morning 94%
Peak 6-8 pm 71%The problem is not the volume of supply but its schedule.
Search-to-Fill
Search to transactionFrom the magnifier through the results to a single dealtransactions ÷ search sessions
The quality of search and matching, not the size of the catalogue.
→ rolls up into ConversionShow example3,400 deals ÷ 41,000 searches = 8.3%
Zero-result searches 11% of sessions
After synonym expansion: 9.6%
Supply / Demand
Balance of sidesA seesaw: many buyers, few sellersactive sellers ÷ active buyers
Shows the tilt of the market and which side the budget belongs to.
Repeat buyer shareThe same buyer returns to the same shopbuyers with 2+ orders ÷ all buyers
Whether demand returns without advertising. It lowers the load on CAC directly.
→ rolls up into FrequencyShow example2,900 of 9,800 buyers = 30%
Their share of GMV 58%
Their CAC $0
Concentration
Turnover concentrationOne wide block and a scatter of narrow ones share the same turnoverGMV share held by the top 10% of sellers
Dependency risk on a few players. Read it alongside their retention.
→ a risk to GMVShow exampleTop 10% = 420 sellers
Their GMV share 62%
Top 5 sellers 19%
Losing one of the top 5 = −$80,000 GMV
Time to First Transaction
Time to first dealDays on the calendar before the first handshakemedian from signup to first transaction
How fast a new participant reaches the market. The two-sided version of TTV.
→ rolls up into ActivationShow exampleBuyers median 2 days
Sellers median 6 days
Sellers with a deal in week one:
90-day retention 64%
Everyone else 17%
Leakage
Off-platform transactionsTwo sides shake hands outside the dashed boundary of the platformestimated share of deals done around the platform
How much take rate escapes the till. Measured by survey and indirect signals.
→ a leak out of RevenueShow exampleSurvey of 600 sellers: 14% of deals off-platform
Off-platform GMV ≈ $59,000 a month
Take rate lost ≈ $5,900
Signal: 23% of chats contain a phone number
09
Reliability and delivery speed
Technical metrics enter the product tree exactly where they move conversion and churn.
Uptime, SLO, Error Budget
Availability and error budgetA month of ticks with two gaps: what is left is the error budgetsuccessful requests ÷ all requests
error budget = 1 − SLO
99.9% is 43 minutes of downtime a month. The error budget is what releases spend.
→ hits ChurnShow exampleSLO 99.9% → budget 43 m 12 s a month
Incident 1 12 m
Incident 2 9 m
Spent 21 m = 49% of the budget
99.95% → 21 m 36 s
99.99% → 4 m 19 s
Latency p50 / p95 / p99
Response time percentilesA distribution with a long tail: p50 near the peak, p99 far to the rightpercentile of the response-time distribution
Users live in the tail. Average latency describes nobody's experience.
→ a driver of ConversionShow examplep50 180 ms
p95 1.4 s
p99 3.2 s
Mean 310 ms
1.2M requests a day
p99 = 12,000 requests slower than 3.2 sA 310 ms mean looks calm until the tail is converted into people.
Crash-free rate
Crash-free sessions and usersA phone with a crack: the share of sessions that make it to the endsessions without a crash ÷ all sessions
On mobile the bar is above 99.5% by session. Below that it shows up in reviews and store rating.
→ hits RetentionShow exampleCrash-free sessions 99.2%
Crash-free users 94.1%
Of 60,000 users, 3,540 saw a crash
Their D30 is 9%, everyone else 34%
Thresholds apply to the 75th percentile of real users, not to a lab run.
→ a driver of Acquisition and ConversionShow exampleLCP p75 3.1 s → above threshold
INP p75 140 ms → within
CLS p75 0.04 → within
After compressing the hero image LCP 2.2 s,
visit → signup conversion 10% → 11.4%
DORA: speed
Deployment Frequency, Lead Time for ChangesCommits travel down the pipe to production, the clock times the tripreleases per week, time from commit to production
The speed of feedback. Rare releases make every release risky.
→ sets Cycle TimeShow exampleDeploys per week 14
Lead time median 6 h
Lead time p85 31 h
Elite: daily or more, under a day
Low: monthly, over a month
DORA: stability
Change Failure Rate, Time to RestoreA row of releases, one turned back, with the time to restore beside itCFR = releases with a rollback ÷ all releases
MTTR = median time to restore
The pair to speed. Without them deploy frequency turns into incident frequency.
→ guards UptimeShow example56 releases in a month, 4 rolled back
CFR = 4 ÷ 56 = 7.1%
MTTR median 38 m, p90 2 h 15 m
Elite: CFR under 15%, MTTR under 1 h
Cycle Time
Task cycle timeFour stages on one line, review is the widestfrom work started to shipped, median and p85
Breaks into stages. Fix the longest one, not the most visible one.
→ rolls up into Lead TimeShow exampleDevelopment 1.4 days
Review 2.9 days
QA 0.8 days
Waiting to ship 1.1 days
Median 6.2 days, p85 14 days47% of the time the task sits in review. That is the bottleneck.
Escaped Defects
Defects found by usersThe sieve catches the bugs, one slips through to the userproduction bugs ÷ releases, and their share of all bugs
The guard metric for release speed. It rises first when tests are cut for deadlines.
→ guards Deployment FrequencyShow exampleBugs found in total 240
Found by users 41 = 17%
Per release 0.7
A quarter ago: 9% and 0.3 per release
10
North Star and building your own tree
The tree is built top down and driven bottom up. A team moves the bottom layer only.
Business goalProfit, market share, valuation. Changes once a year.
one metric that captures delivered value
North Star MetricNights booked, minutes listened, reports sent. Not revenue.
decomposes into a product or a sum
DriversUsually 3-5: base width, frequency, depth, quality.
each driver breaks into something controllable
Team input metricsWhat a team moves in a sprint: step conversion, TTV, feature adoption, p95.
North Star criteria
One star, three factors under it
Captures user value, not the fact of payment.
Leads revenue: it moves before the money does.
Controllable by the team within a quarter.
Decomposes into factors that do not overlap.
Resists gaming: you cannot raise it by making the product worse.
Show exampleNorth Star: exports per week 85,000
= weekly actives 38,000
× share who export 41%
× exports per exporting user 5.5Three factors, three owners, three different teams.
Levels and owners
A pyramid of four levels, each with its own owner
L0 money: CEO and board, horizon one year.
L1 North Star: product leadership, horizon one quarter.
L2 drivers: area owners, horizon one month.
L3 input metrics: the team, horizon one sprint.
L4 counters and events: analytics, refreshed daily.
Show exampleL0 EBITDA $1,500 a month
L1 85,000 exports a week
L2 38,000 actives, 41% share
L3 TTV 18 min, activation 40%, p95 1.4 s
L4 events export_clicked, export_done
11
Traps
Eleven ways to get the right number and the wrong conclusion.
Vanity metrics. Cumulative signups, downloads, followers. They only go up and change no decision.
Goodhart's law. A metric that becomes a target stops being a metric. Pair every target with a guard: speed plus quality, conversion plus refunds.
Means instead of percentiles. Load time, session length, order value: anything with a long tail is read at the median and p90.
A ratio with no denominator. Conversion rose to 8% because traffic halved. Always show both parts.
Survivorship bias. Retention measured on active users is flattering by construction.
Simpson's paradox. A metric rises in every segment and falls overall because the mix changed.
Mixed cohorts. A young base hides churn: the leavers have not had time to leave.
Last-click attribution. The closing channel eats the credit of the channel that introduced you.
Peeking in A/B tests. Stopping the moment significance appears inflates the effect. Fix sample size and duration before the start.
Local optimum. A string of small conversion wins will not save a product with no retention.
Definition drift. "Active user" changed quietly and the yearly chart became meaningless. Keep definitions in one place, each with the date it changed.
Nothing found. Try another word: churn, CAC, p95, take rate.