Where you are
You know the store (b1) and you can compute UV, session CR, AOV, and GMV by hand (b2). This session turns those isolated numbers into a structure: a tree where GMV sits at the root and every number under it is a lever someone at Mango Lane owns. In b4 the tree grows the full funnel; in b5 you use it to diagnose a real drop. This is the course's centerpiece skill.
GMV is a product, and products pass through 10 min live
Mango Lane's month: 360,000 sessions x 2.7% session CR x $68 AOV = $660,960 GMV. Because the top line is a pure product, a percentage change in any single driver flows to the top line as roughly the same percentage. Sessions down 10%, everything else flat? GMV down 10%. Conversion down 10% and AOV down 5%? GMV down about 15%. That clean pass-through is the entire trick behind fast diagnosis.
LiveThe pass-through rule, precisely4 min▶
For the small moves you see week to week, percentage changes in a product roughly add. If sessions fall 10% and CR falls 5%, GMV falls about 15% (exactly: 1 − 0.90 × 0.95 = 14.5%). Two consequences:
- Diagnosis is subtraction. GMV fell 12%? Find drivers whose percentage falls sum to about 12 and you have explained the drop.
- Percentages, never absolutes. "Sessions fell 36,000" and "AOV fell $7" cannot be compared. "Sessions fell 10%" and "AOV fell 10%" hit GMV identically. Always convert to percentage change per driver first.
Self-studyWhere this idea comes from2 min read▶
Breaking a top-line ratio into multiplied levers is the DuPont analysis move from the 1910s (return on equity = margin x turnover x leverage), and it never stopped working. Ecommerce inherited it as GMV = traffic x conversion x AOV. If driver trees themselves fascinate you - additive trees, flow bridges, leading indicators - the sibling course learn-metric-decomposition-with-phoebe teaches the pattern across six industries. This course keeps the focus on the ecommerce numbers themselves.
Tree one: the simple GMV tree ★ 8 min · everyone builds
Here is Mango Lane's simple tree as a live simulator. Every leaf is editable and GMV recomputes instantly. Read the baseline, then run the drills below.
{
"unit": "$",
"root": {
"label": "GMV", "op": "x",
"children": [
{ "label": "Sessions", "value": 360000, "unit": "sessions" },
{ "label": "Session CR", "value": 0.027, "pct": true },
{ "label": "AOV", "value": 68, "unit": "$" }
]
}
}
Read the baseline: 360,000 x 2.7% x $68 = about $661K. Match it to the b2 numbers you computed by hand - same store, same month.
Drop Session CR from 2.7 to 2.43 (exactly −10%). Watch GMV fall 10%. That is pass-through, live.
Reset. Now raise AOV from 68 to 74.80 (+10%). GMV rises 10%. Levers work in both directions.
Press Simulate a drop. One driver falls at random and its path lights coral. Say the diagnosis out loud: "GMV fell X% because [driver] fell Y%."
Tree two: open the conversion black box ★ 10 min · everyone builds
"Session CR fell" is better than "GMV fell", but CR is still a black box - it hides the whole shopping journey. Replace the single 2.7% with the chain that produces it: PDP penetration (40% of sessions reach a product page) x add-to-cart (25% of those) x checkout start (45% of those) x checkout completion (60% of those). Multiply the chain: 0.40 x 0.25 x 0.45 x 0.60 = 2.7%. Same number, four levers deep.
{
"unit": "$",
"root": {
"label": "GMV", "op": "x",
"children": [
{ "label": "Orders", "op": "x", "unit": "orders", "children": [
{ "label": "Sessions", "value": 360000, "unit": "sessions" },
{ "label": "PDP penetration", "value": 0.40, "pct": true },
{ "label": "Add-to-cart rate", "value": 0.25, "pct": true },
{ "label": "Checkout start", "value": 0.45, "pct": true },
{ "label": "Checkout completion","value": 0.60, "pct": true }
] },
{ "label": "AOV", "value": 68, "unit": "$" }
]
}
}
Read the middle node: Orders = 9,720. The funnel chain and the flat 2.7% CR are the same fact at two zoom levels.
Cut Checkout completion from 60 to 48 (−20%). GMV falls 20% - a checkout bug passes straight through, however healthy the rest of the funnel is.
Reset. Cut PDP penetration from 40 to 36 (−10%). Same −10% GMV. A navigation problem and a payment problem hit the top line by the same rule - but land on different teams' desks.
Simulate a drop and name the guilty gate, not just "conversion". That precision is what b4 measures and b5 exploits.
Why the deep tree earns its keep: industry-wide, roughly 70% of carts are abandoned (Baymard Institute's running average across 50 studies is 70.22%). A store watching only "conversion rate" sees a fuzzy 2.7%; a store watching the four gates sees exactly which one leaks worse than its benchmark - and Mango Lane's 60% checkout completion is actually strong (typical Shopify stores complete around 45%, per Littledata's 2025 panel).
Tree three: GMV as a sum of channels ★ 8 min · everyone builds
Trees do not have to multiply. The same $660,960 is also the sum of what each traffic channel brought in: organic + paid + CRM + social. The multiply tree answers "which lever?"; the additive tree answers "which slice?". Real diagnosis (b5, b6) uses both.
{
"unit": "$",
"root": {
"label": "GMV", "op": "+",
"children": [
{ "label": "Organic", "value": 264384, "unit": "$" },
{ "label": "Paid", "value": 198288, "unit": "$" },
{ "label": "CRM", "value": 132192, "unit": "$" },
{ "label": "Social", "value": 66096, "unit": "$" }
]
}
}
Read the shares: organic 40%, paid 30%, CRM 20%, social 10%. They sum to $660,960 exactly - the same month as trees one and two.
Cut Paid by half (198,288 to 99,144). GMV falls 15% - a 50% fall in a 30% slice moves the total by 0.5 x 30%. In an additive tree, impact = the branch's percentage move x its share.
Compare the rules: multiplicative pass-through (full percentage) vs additive weighting (percentage x share). Knowing which tree you are reading is half the diagnosis.
Self-studyWhich tree when3 min read▶
| Question | Tree | Rule |
|---|---|---|
| Which lever moved? (rate vs volume vs value) | Multiplicative (trees 1-2) | % change passes straight through |
| Which slice moved? (channel, device, category) | Additive (tree 3) | % change x that slice's share |
| Both - lever first, then slice | b5's diagnosis walk | multiply tree aims you, additive tree locates it |
Try it yourself - this week ◐ 20-30 min total
- Rebuild tree one from memory on paper: three leaves, their baseline values, their owners. If you cannot, replay Demo 1.
- In the funnel-deep tree, find two different single-leaf edits that each cut GMV by exactly 10%. Notice they land on different owners.
- Take your own store or product's top-line number and write it as a product of 3-5 levers. Name each lever's owner. An unowned lever is an unwatched lever.
- Bookmark this page - the simulators are your scratch pad for every later session. Paste your own numbers in whenever you like.
Sources this session draws on
The tree is a craft pattern, not a certificate topic; the numbers inside it are the platform-defined metrics from b1-b2. This page draws on:
Three questions before you go 🎯 ◐ 90 seconds
1 · Sessions fall 10%, everything else flat. In GMV = Sessions x CR x AOV, GMV falls...
In a multiplicative tree, a single driver's percentage change flows to the top line at the same percentage. That pass-through is what makes the tree readable at a glance.
2 · Why replace the flat 2.7% session CR with the four funnel gates in tree two?
0.40 x 0.25 x 0.45 x 0.60 = 2.7% exactly. The deep tree adds no new data - it adds addressability: a navigation problem and a payment bug stop looking identical.
3 · In the channel tree, Paid (30% of GMV) falls 50%. GMV falls...
Additive branches are weighted by share: 50% x 30% = 15%. Multiplicative levers pass through in full; additive slices pass through by weight. Know which tree you are reading.