Where you are in the leader track
You can read the GMV equation (a1), speak the metric dictionary (a2), read a funnel (a3), run a weekly review (a4), and judge benchmarks and seasons (a5). The last piece is making the numbers themselves trustworthy - because every skill above collapses if two dashboards disagree about what "conversion rate" means. That is governance, and it is a leadership job, not an analyst hobby.
One definition per metric, written down 15 min live
Here is the fight, verbatim from a thousand companies: marketing opens GA4 and reports conversion at 3.9%. Product opens Shopify and reports 2.7%. Voices rise. Someone questions the tracking. Someone questions the other team. Both numbers are correct - GA4's purchase journey counts users, Shopify's online store conversion rate counts sessions. Same store, same month, two denominators. The meeting was not wasted by a wrong number; it was wasted by an unwritten definition.
LiveThe cure costs one page6 min▶
Governance sounds heavy; the minimum viable version is one page per metric with four lines:
- Name - the exact phrase people say in meetings ("conversion rate").
- Definition with denominator - purchase sessions ÷ total sessions. The denominator is the line that ends fights.
- Source of truth - which system's number wins when tools disagree (here: Shopify for CR, with GA4's user-based CR reported as a separate, differently named metric).
- Owner - one person's name. Not a team. A name.
The Shopify-vs-GA4 split is the canonical example because it happens at essentially every store running both tools. Shopify's online store conversion rate divides purchase sessions by total sessions; GA4's purchase journey counts users through a closed funnel. At Mango Lane that is 2.7% vs ~3.9% - a 44% gap with zero business meaning. Companies that never write this down re-fight it monthly.
Self-studyWhat else belongs on the spec page3 min read▶
Beyond the four core lines, the spec pages that actually end arguments also declare:
- Exclusions and filters - does AOV exclude gift cards? Does CR exclude internal traffic and bots? Are cancelled orders in or out? (Recall a1: GMV vs revenue has no universal standard - always ask what is netted out.)
- Period and timezone - a "day" that ends at midnight UTC and a "day" that ends at local midnight produce different weeks. Pick one, write it down.
- Refresh cadence - if the dashboard lags the warehouse by a day, say so on the tile, or every Monday starts with "why is Sunday missing".
- Related-but-different metrics - the spec for session CR should name visitor CR as a separate metric, so nobody "corrects" one into the other.
None of this is analytics sophistication. It is the same discipline as a contract: the clauses exist so nobody has to argue from memory.
North star + guardrails 12 min live
A north star metric - for most stores GMV or orders - gives everyone one number to move. Useful, and dangerous: any single number can be moved in ways that hurt the business. Deep discounts push GMV up while margin bleeds. Aggressive upsells push GMV up while returns climb and customers do not come back. The fix is never to abandon the north star; it is to pair it with guardrail metrics that catch the cheap ways to move it.
LiveThe pairing table6 min▶
Every north star gets two named guardrails, reviewed on the same page, at the same cadence:
| North star | Cheap way to move it | Guardrails that catch it |
|---|---|---|
| GMV | Deep discounts on everything | Gross margin + discount rate |
| GMV | Aggressive upsells and bundles | Return rate + repeat purchase rate |
| Orders | Push cheap add-on items | AOV + gross margin |
| Conversion rate | Buy low-intent discount traffic out of the denominator's sight | Traffic quality (new-visitor share) + AOV |
The point is symmetry: whoever proposes moving the north star also names the guardrail their tactic would strain. That one norm removes most of the gaming before it starts.
Self-studyThe Goodhart pattern, lightly3 min read▶
There is an old observation - usually summarized as Goodhart's law - that when a measure becomes a target, it stops being a good measure. You do not need the academic version; you need the working version: when a driver becomes a target, watch it start lying.
- Target the CR and someone will trim low-intent traffic out of the denominator - CR "improves", orders do not.
- Target ATC rate and the button gets bigger while checkout completion quietly falls.
- Target GMV alone and discounts do the rest.
None of this requires bad actors - just smart people optimizing what you told them to optimize. Guardrails are how leadership says "move this number honestly".
Who owns the definitions 8 min live
A definition nobody owns drifts. A tracking change ships, a filter gets added, a dashboard gets rebuilt - and six months later "conversion rate" quietly means something new, and every trend line that crosses the change date is fiction. Ownership is three small rules, enforced from the top.
LiveThree rules, one demand5 min▶
- A named owner per metric. One person answers for the definition of CR, one for AOV, one for GMV. When two dashboards disagree, the owner adjudicates - not the loudest voice.
- Definitions change only with a changelog entry. Date, what changed, why, and which historical numbers are affected. A metric whose history cannot be trusted is worse than no metric.
- Dashboards cite the spec. Every tile links to (or names) the definition it implements. If a dashboard cannot say which spec it follows, it is an opinion with colors.
Self-studyFive questions that expose a shaky metric3 min read▶
Ask these about any number that is about to drive a decision. A well-governed metric survives all five in under a minute; a shaky one dies on the second:
- 1 · Who owns this definition? If the answer is "the dashboard", nobody does.
- 2 · What is the denominator? Sessions, visitors, orders, PDP sessions? The a2 reflex, forever.
- 3 · When did it last change? If nobody knows, trend lines across unknown change dates are unreliable.
- 4 · What would game it? If nobody can name a cheap way to move it, nobody has thought about it hard enough.
- 5 · Which guardrail watches it? A target without a guardrail is an invitation.
Governance drill: three incidents ★ 8 min · no computation
Three incidents, described exactly as they surface in real reviews. For each: what governance catches, and which rule did the catching.
LiveRun the drill8 min▶
1 · Two dashboards disagree on AOV. Finance's AOV runs a few dollars higher than the store dashboard's. Nobody changed anything. The spec-sheet answer: one of them excludes gift cards - Shopify's AOV does, by definition (revenue ÷ orders excluding gift cards and post-order adjustments). Both are computable; the spec says which one is the AOV. Rule that caught it: one definition, declared exclusions.
2 · CR "improved" right after a tracking change. A new consent banner ships, and session CR jumps the same week. Nothing about shoppers changed - the tracking now records fewer low-intent sessions, shrinking the denominator. That is definition drift, not business improvement. Rule that caught it: changelog entry on the change date, flagged in the review before anyone took credit.
3 · The team hits its GMV target - via discounting. GMV lands exactly on plan; celebration is scheduled. The guardrail page shows gross margin down and discount rate up: the target was reached by selling dollars for ninety cents. Rule that caught it: north star reviewed only alongside its guardrails, on the same page.
Try it yourself - this week ◐ 15 min total
- List your top 5 metrics - the ones that actually appear in your weekly review.
- For each, write two things: a named owner, and the denominator (or declared exclusions, for value metrics like AOV).
- Mark which metrics have neither. That short list is your governance backlog - and the first thing to hand your analysts.
- Pick your north star and write its two guardrails next to it. If you cannot name a cheap way to game the north star, ask the team - they can.
Sources this session draws on
Governance is less about citations and more about craft - but the load-bearing facts on this page come from these, with honesty labels:
Three questions before you go 🎯 ◐ 90 seconds
1 · Marketing reports CR at 3.9% (GA4), product reports 2.7% (Shopify). Both are correct. What ends this fight permanently?
The numbers differ because the denominators differ - users vs sessions. A written spec (definition + denominator + source of truth + owner) turns the monthly fight into a ten-second lookup.
2 · The team hits its GMV target through deep discounting. Which guardrail catches it?
Deep discounts move GMV up and margin down. That is why a north star is only reviewed alongside its guardrails: margin and discount rate for discount-driven GMV; return rate and repeat purchase rate for upsell-driven GMV.
3 · CR "improves" the same week a tracking change ships. Best first read?
A tracking or consent change alters which sessions get counted, shrinking or growing the denominator with zero shopper change. This is exactly what the changelog rule exists for: any metric change gets a dated entry before anyone takes credit for the "improvement".