The last page of the roadmap
Session 1 drew the money map and flagged that take rate moves by almost half depending on whether the retainer is inside it. Session 2 put hours removed first. Session 3 got eight teams reading one table. Session 4 showed which grains Serai can actually hold. Session 5 turned the estate into three products, each of which publishes numbers other people will repeat. Tonight is the discipline under all of them: a registry, so that when someone outside the building reads GMV, take rate or active brands, the number means one thing and can be traced to the rows that made it. Serai and every figure below are constructed for these pages; the method is what transfers.
★ What you walk out with tonight. Three registry entries, for GMV, take rate and active brands, each with a definition in words, a formula, a grain, its source tables, one named owner, a version and the one question an auditor would ask first. Plus the sentence that ends every argument about which number is right: "the number is whatever version 1.0 says, and if we change it, we restate every quarter we have reported."
Three numbers everybody quotes and nobody defines
6 min liveTake one Serai quarter, constructed, and ask three people for its GMV. Finance, the commercial team and the brand lead each give a different number, and each is right about the rule they used. Then divide the fees by each GMV and count the active brands three ways. Nine numbers, one quarter, no mistakes anywhere, and that is exactly the problem.
LiveThree GMVs from the same orders2 min▶
Start from the orders placed on the stores Serai runs in the quarter. Count the item value plus the shipping the buyer paid and you get $48,600,000. Take the shipping out and you get $45,000,000, the figure the bench and the earlier sessions use. Keep only what actually settled, net of cancellations and returns, and you get $39,200,000.
Each one answers a real question. The first is closest to what buyers spent. The second is what Serai's 2 percent share is calculated on in its constructed contracts. The third is the money that actually reached the brands. The gap between the first and the third is $9.4 million, about a fifth of the widest number, and on a campaign quarter with heavy returns it grows.
LiveThree take rates, and a fourth2 min▶
Serai's constructed fees for the quarter are the retainers, $864,000, plus the GMV share, $900,000: $1,764,000. Divide by each GMV and the take rate is 3.63 percent, 3.92 percent or 4.50 percent. Leave the retainer out, on the argument that it is not a share of anything, and it is 2.00 percent.
Notice the direction. The more generous the GMV, the smaller the take rate looks, which is why a pitch to brands and a pitch to investors are tempted to use different ones. That temptation is the reason the definition has to be written down before either pitch is made.
Self-studyThree brand counts, and why "active" is the most argued word in a board pack2 min▶
48 brands are under contract. 44 placed at least one order in the quarter. 41 reached at least $10,000 of settled GMV. A brand that signed in the last week of the quarter and sold nothing is in the first count only; a brand with one small order is in the second but not the third.
Which is right depends on what the reader is using it for. A reader judging the sales team wants contracts. A reader judging whether the brands stay wants the ones that actually trade. A reader modelling revenue wants the ones that trade enough to matter. None of them is wrong, and a count published without its test is a number the reader will fill in with whichever meaning suits them.
A listed marketplace writes its rule down. Sea Limited's annual report defines Shopee's GMV as the value of orders on the marketplace, says its calculation "includes shipping and other charges", and counts an order "regardless of whether the transaction is settled or if the item is returned". That is the top row of the figure, chosen deliberately and published with the number. The lesson is not which rule Sea chose; it is that a listed company states the rule next to the number, so a reader never has to guess.
The registry: one definition, one owner, one lineage
6 min liveA metric registry is a short list of the numbers the company quotes outside the building, each written down once, with the same fields every time. It is not a dashboard and not a data catalogue. It is the page anyone can open to find out what a number means, who answers for it, and which rows it came from.
LiveThe eight fields every entry carries3 min▶
| Field | What goes in it | Why it is there |
|---|---|---|
| Name | A machine-safe name that says the rule: gmv_placed_ex_ship, not gmv | The name itself stops the wrong version being pasted in |
| Definition in words | One or two sentences a board member can read | The formula is for the analyst; the words are for everyone else |
| Formula | The calculation, with every filter spelled out | Two people computing it must get the same number |
| Grain | What one row is (order line, brand x quarter) and how it rolls up | Most disagreements are two grains, not two errors |
| Source tables | The conformed tables it reads, and through them the exports | This is the first link of the lineage |
| Owner | One named person, not a team | Someone answers the question when it is asked |
| Version | v1.0, then v1.1 when anything changes | Every reported number can say which rule made it |
| Change log | What changed, why, when, and the effect on reported periods | The history a careful reader asks for first |
Three headline metrics need three entries. Most enablers end up with ten to twenty entries once the brand-facing products from session 5 publish their own numbers; start with the three everybody quotes.
LiveLineage: a number you can walk back to its rows3 min▶
Lineage is the chain in the figure, followed backwards. The $45,000,000 on slide 4 names its entry, gmv_placed_ex_ship version 1.0. The entry names its source, the conformed order_lines table. The table was built from three marketplace exports whose price columns are all called something different: item_price, paid_price, sale_price. Analyst session 2 conforms them into one schema; that conforming step is part of the lineage, because a mapping error there changes the board-pack number without anyone touching the registry.
The test of a lineage is simple: pick any number in last quarter's board pack and ask how long it takes to get to the rows. If the answer is "a few minutes, by following the names", you have lineage. If it is "ask the analyst who built it", you have a person, and people leave. Analyst session 7, The metric registry as code, writes these entries as files with tests that fail when a definition drifts. The general practice of metric governance, beyond an enabler, is taught in the ecommerce metrics course, session 6.
Self-studyWhat owning a metric actually means2 min▶
The owner does not compute the number every quarter. The owner decides what it means, approves any change to the definition, and answers when an outside reader asks why it moved. That is a business role, so for GMV and take rate it usually sits in finance and for active brands in the commercial team, with the data team as the custodian that keeps the formula and the lineage true.
- One person, by name. "Finance" cannot be phoned when a reader asks a question on a Friday.
- The data team is not the owner. It keeps the entry running; it does not decide whether retainers belong in a take rate.
- Ownership survives a resignation. When the owner leaves, the entry gets a new name the same week, and the change log says so.
Two honest tables, one quarter, different totals. On the public Olist dataset (Olist, "Brazilian E-Commerce Public Dataset by Olist", Kaggle, CC BY-NC-SA 4.0), the order lines placed from April to June 2018 add up to R$ 2,858,289.74 of items, or R$ 3,332,156.97 with shipping. The payments table for the same orders says R$ 3,332,386.55, R$ 229.58 more. Neither table is wrong; they record different things. Count active sellers in the same quarter and you get 1,680 with any order, 1,653 with a delivered order, and 552 with at least R$ 1,000 of delivered items. A registry entry is where you write down which table is the source, so the R$ 229.58 is explained once instead of rediscovered every quarter. The analyst track recomputes every one of these figures.
Month one, not month twenty
4 min liveA registry written the month the company decides it might one day raise money from the public is a few afternoons of work. A registry written the month before a listing document is due is a migration: every past quarter re-derived under one rule, every deck that used a different one explained. The questions a serious reader will ask are public, so there is no reason to wait for them.
LiveThe three questions, as a checklist you run on every entry2 min▶
Run each registry entry through the card. Can you state the definition and how it is calculated in words a non-analyst reads once? Can you say why the number is useful to someone deciding whether to invest, not just to the team that likes it? Can you say how management actually uses it, in a meeting that happens, with a decision that follows?
The third question is the one that catches most metrics. A number that nobody inside uses to run the business is a number that exists to be quoted, and a careful reader can tell. If an entry fails it, either find the decision it informs or stop publishing it.
LiveWhen a definition changes: new version, prior periods restated2 min▶
Definitions do change, for good reasons. Suppose Serai decides settled GMV is the fairer base for take rate. The move from 3.92 to 4.50 percent on the same quarter is not growth; it is a new rule. So the change is a new version, v2.0, with a change-log line that says what changed, why, and the effect on amounts already reported. Every prior quarter is recomputed on the new rule, and for a while both are shown side by side.
The failure mode is the quiet one: apply the new rule from next quarter, leave the old quarters as they were, and let the chart show a jump that is really a definition. The frozen definition is applied to every period reported, or the trend line is fiction.
Self-studyWhy the registry comes before the pitch, not after it2 min▶
- In month one it is a decision. Three entries, a few afternoons, argued out among four people who all remember why.
- In month twenty it is an archaeology project. Eight quarters of board packs, three versions of GMV in circulation, and the analyst who built the first dashboard has left.
- Every deck in between is a liability. A pitch that quoted 48 active brands and a board pack that quoted 41 are both on file, and a reader who finds both will ask which one was true.
- It fits on the roadmap. On the one-page roadmap the registry is a small, early line item that needs nobody's permission, which is exactly the shape session 2 said to put first.
The questions are already public. The US Securities and Exchange Commission's February 2020 guidance on management's discussion and analysis (Release No. 33-10751) says the Commission "would generally expect" a key metric to come with a clear definition and how it is calculated, a statement of why it provides useful information to investors, and a statement of how management uses it. It adds that when the calculation method changes, a company should consider disclosing the differences, the reasons and the effects on amounts previously reported, and it reminds companies that such metrics fall under their disclosure controls. Read it as one public example of the questions a serious reader asks, which makes it a ready-made checklist for a registry. It says nothing here about what any particular exchange requires.
Register your three headline metrics
★ 20 min · everyone buildsNo code, no tool: a sheet with eight rows and three columns is enough. Fill it for your own company, or for Serai using the constructed entries below. The hard part is not the typing; it is the two or three arguments you will have about which rule wins, and those arguments are the point.
Collect every version already in circulation. Open the last pitch deck, the last board pack and the last brand review. Write down every GMV, take rate and active-brand count you find, with the document it came from and the rule it used, or "not stated". For Serai that list is the nine numbers from Part 1, plus the 2.00 percent take rate. Most companies find at least two versions of each; that list is the case for doing the rest.
Register GMV: pick one rule, write it in words first. Choose the version your fees are actually calculated on, name it so the rule is in the name, and publish the others beside it under their own names rather than deleting them. Then write the question an auditor would ask first, and your answer.
Register take rate, and settle the retainer argument once. Decide whether the retainer is inside the take rate, and write down why. Whichever you choose, publish the other version under its own name, so nobody can quote one as if it were the other.
Register active brands, with the test in the name. "Active" is the word that gets stretched most, so the entry says exactly what a brand must do in the quarter to count. Pick the test that matches how the number is used, and publish the others beside it.
Run the card, start the change log, freeze it. Put each entry through the three questions from Part 3 and add the line about how management uses it. Then write the first change-log line and agree, out loud, that every reported period uses v1.0 until a new version is logged and prior periods are restated.
Before the analyst track
30 min- Finish step 1 for your own company: every GMV, take rate and active-brand count in your last pitch deck, board pack and brand review, each with the rule it used or "not stated". Count how many versions of each you found.
- Fill the eight fields for your three headline metrics and put a person's name, not a team's, in every owner field. Send each entry to its owner and ask them to change one word they disagree with.
- Pick one number from last quarter's board pack and trace it back to the export rows. Time it. If it takes longer than an afternoon, write down where the trail went cold; that is the first gap in your lineage.
- Run each entry through the three questions and the method-change line. For any metric where you cannot answer how management uses it, decide whether to find the decision it informs or stop publishing it.
Three questions
Clear all three and Leader session 6 gets stamped in your passport.
1 · Serai's quarter shows GMV of $48.6M, $45.0M and $39.2M. How can all three be true?
Nobody made a mistake. Shipping in or out and placed or settled are both defensible choices, and each number answers a real question. The trouble starts only when one is quoted without its rule, which is what the registry prevents.
2 · What must a registry entry carry?
The words are for readers, the formula for analysts, the grain and sources for the lineage, the owner for the question on a Friday, and the version and change log for every number already reported. A team is not an owner, and a chart is not a definition.
3 · Serai decides take rate should be measured on settled GMV instead of placed GMV. What should happen?
The move from 3.92 to 4.50 percent on the same quarter is a new rule, not growth. B turns a definition into a fake jump on the trend line; C refuses a legitimate improvement; D is the problem the registry exists to end.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Leader session 6 cheat sheet · pin this
That closes the Leader track. You have the money map, the hours-first rule, one table for a new channel, the grains you can hold, three products and now the registry under all of them. Lay them side by side on the playbook's roadmap on one page, and you have something a chief executive can fund. The Analyst track builds the same estate in Python and SQL, starting with Seven domains, one enabler estate; its session 7 turns tonight's three entries into code with tests that fail when a definition drifts.