Where you are
b5's workflow ended at step 4 with a pointer: "cut by segments - see b6". This is b6. You learn the three standard cuts every analyst runs, the mix effect that makes blended numbers lie while every segment tells the truth, and the discipline of cutting in a fixed order and stopping. Next session (b7) packages all of it into the weekly review pack.
The tree says which lever - segments say where 12 min live
"Session CR fell 8%" is a lever. But Mango Lane's CR is not one number - it is a blend of many populations converting at different rates. Three cuts cover most diagnoses: channel (organic, paid, CRM, social), device (mobile, desktop), and new vs returning.
| Cut | The question it answers | What typically breaks it |
|---|---|---|
| Channel | Did one traffic source break? | Paused campaigns, search algorithm updates, broken email links |
| Device | Did one platform break? | App releases, mobile checkout changes, responsive-layout bugs |
| New vs returning | Did acquisition quality shift? | New audiences, discount-hunting promo traffic, an aging customer base |
Here is the device cut at Mango Lane's baseline - the same month you already know.
| Device | Sessions | Session CR | Orders |
|---|---|---|---|
| Mobile | 252,000 | 2.4% | 6,048 |
| Desktop | 108,000 | 3.4% | 3,672 |
| Total | 360,000 | 2.7% blended | 9,720 |
LiveRead the device table like an analyst5 min▶
Three things to notice before we break it:
- The cut reconciles exactly. 6,048 + 3,672 = 9,720 orders; 252,000 + 108,000 = 360,000 sessions; and 9,720 / 360,000 = 2.7% blended. If your segments do not sum back to the total, you have a tracking gap, not a diagnosis.
- Mobile converts lower - everywhere, not just here. Littledata's 2025 Shopify benchmark puts mobile CR at 1.2% vs desktop 1.9%. Smaller screens, more window-shopping, more checkout friction. Mango Lane's 2.4% vs 3.4% follows the same pattern from a healthier baseline.
- The blended 2.7% sits between the segments, weighted by traffic. Mobile is 70% of sessions, so the blend leans toward mobile's 2.4%. Hold that thought - it is about to matter a lot.
Self-studyWhy exactly these three cuts3 min read▶
You can slice by dozens of properties - category, price band, region, browser, hour of day. The three standard cuts earn their place because each maps to a distinct cause family with a distinct owner:
- Channel = acquisition machinery. Campaigns, budgets, algorithms, email sends. Breaks are frequent, sharp, and owned by marketing.
- Device = the product surface. Code ships per-platform, so bugs land per-device. Owned by engineering and product.
- New vs returning = audience quality. Shifts are slow and compositional. Owned by acquisition strategy and CRM together.
- Everything else is situational. Cut by category when merchandising changed, by region when logistics changed. Add a cut when you have a hypothesis, not because the dropdown offers it.
The mix effect: the trap of the session ★ 15 min · everyone computes
Next month, paid marketing pushes app installs hard and the traffic mix shifts: 280,000 mobile / 80,000 desktop sessions. Here is the trap: both segment CRs hold exactly - mobile still 2.4%, desktop still 3.4%. Not a single shopper behaves differently. Now compute the blend by hand.
Mobile orders: 280,000 x 0.024 = 6,720. Desktop orders: 80,000 x 0.034 = 2,720.
Total orders: 6,720 + 2,720 = 9,440 across the same 360,000 sessions.
Blended CR: 9,440 / 360,000 = 2.62%. Last month it was 2.70%.
Read the result out loud: blended CR "fell" 3% with zero behavior change. More traffic simply moved to the lower-converting device. That is the mix effect - pure composition, no conversion problem anywhere.
This pattern has a famous name. It is a Simpson's-paradox-style reversal: every part can hold steady (or even improve) while the blend moves the other way, purely because the weights shifted. It runs in reverse too - a store whose blended CR is "improving" may just be acquiring less low-intent traffic. Either direction, the fix is the same: look at segment rates AND segment mix before crediting behavior.
Self-studySplit any blended move into rate and mix4 min read▶
The general two-step decomposition, usable on any blended metric:
Counterfactual blend: apply this period's mix to LAST period's segment rates. For our shift: (280,000 x 0.024 + 80,000 x 0.034) / 360,000 = 2.62%. This is what the blend "should" be if only the mix moved.
Mix piece: counterfactual minus last period's actual. 2.62% - 2.70% = -0.08 points from mix.
Rate piece: this period's actual minus the counterfactual. Here 2.62% - 2.62% = 0.00 points from behavior. The entire move was composition - and now you can say so with arithmetic, not intuition.
When both pieces are nonzero (the usual real-world case), the split tells you how much of the story belongs to acquisition (mix) and how much to the site experience (rate) - which is exactly the line between the marketing team's problem and the product team's problem.
New vs returning - mix in disguise 8 min live
Shopify's returning customer rate = returning customers / (returning + first-time customers) in the period. Returning customers convert better and spend more - they already trust the store, know their size, have a saved card. Which sets up the same trap at a slower speed.
LiveThe store that "improved" by aging4 min▶
A store two years into its life sees CR climb quarter after quarter and credits its UX roadmap. Cut by new vs returning and both segment CRs are flat - the store is simply aging into a returning-heavy base, and returning customers convert better. Mix again, on a quarterly clock instead of a weekly one.
A quick illustration with made-up round numbers (not Mango Lane's): say new visitors convert at 2.0% and returning at 4.5%, both flat all year. Returning share grows from 20% to 35% of sessions as the base matures. Blended CR moves from 2.0 x 0.80 + 4.5 x 0.20 = 2.5% to 2.0 x 0.65 + 4.5 x 0.35 = 2.875% - a 15% "improvement" in which not one visitor behaved differently.
- Why it matters: if the CR gain is mix, then acquisition is quietly stalling while the dashboard celebrates. The blend hides exactly the problem you most need to see.
- The check is always the same: segment rates flat + mix shifted = composition story. Segment rates moved = behavior story. Run the check before writing the narrative.
Self-studyChannel cuts and the additive rule3 min read▶
The channel cut uses the additive tree from b3: Mango Lane's GMV splits Organic 40% / Paid 30% / CRM 20% / Social 10%. The rule from b3 applies directly: additive impact = the branch's move x its share.
- Paid falls 20%? Top-line impact = 20% x 30% share = 6% of GMV.
- Social falls 20%? Impact = 20% x 10% = 2%. Same percentage fall, a third of the damage - share is what converts a channel move into a GMV move.
- This is why the channel cut comes first in the discipline below: one broken source with a big share explains a big drop fast.
The cut discipline ★ 8 min · everyone works
Segments are powerful and bottomless - you can cut by channel x device x cohort x category until every cell has three sessions in it and everything looks broken. The discipline: a fixed order, and a stopping rule.
Cut 1 - channel first. Did one traffic source break? A paused campaign, a Google update, a broken email link. Channels break independently and often explain the whole move in one look.
Cut 2 - device second. Did an app release or a mobile checkout change break one platform? Releases ship per-platform, so damage shows up per-device.
Cut 3 - new vs returning third. Did acquisition quality shift? New-heavy traffic converts lower - a mix story wearing a behavior costume.
Stop at the first cut that isolates the fall. If the entire drop sits in paid mobile traffic, you are done cutting - write the finding (b5's template). Deeper cuts on thin data are noise-chasing, not thoroughness.
What noise-chasing looks like from the inside: a team spends three days investigating why "CRM desktop tablet-landscape sessions" converted 40% worse last week - a cell with nine orders in it. Meanwhile the actual answer to the store-wide drop was sitting one level up: the whole paid channel fell 20% when a campaign hit its budget cap. Fixed cut order plus the stopping rule would have found it in the first ten minutes. Depth feels like diligence; usually it is procrastination with a pivot table.
Self-studyThe thin-slice rule3 min read▶
A segment with a handful of daily orders swings wildly for free. Social desktop traffic at Mango Lane might convert 12 orders one day and 21 the next - a "75% surge" that is pure small-number noise. The working rule:
- Do not diagnose weekly moves on slices under roughly 100 conversions. Below that, week-to-week swings of 20-30% are what randomness looks like, not what problems look like.
- Widen before you conclude: a thin slice that looks broken this week deserves a 4-8 week window before it deserves a finding.
- This is the stopping rule's twin: the cut order stops you going too deep; the thin-slice rule stops you trusting what you find when you do.
Try it yourself - this week ◐ 20-30 min total
- Compute your store's blended CR two ways: once overall (orders / sessions), and once as the mix-weighted sum of device CRs (mobile sessions x mobile CR + desktop sessions x desktop CR, divided by total sessions). If the two disagree, you have found a tracking gap - go find which sessions or orders are missing a device label.
- Pull last quarter's new vs returning split. Is your returning share growing? If so, how much of your CR trend is mix?
- Rerun Demo 1's arithmetic with your own device numbers: hold both CRs fixed, shift 10% of sessions from desktop to mobile, and see how much your blended CR "falls" with zero behavior change.
- List your thin slices: which segments in your store are under 100 weekly conversions? Write them down so future-you does not diagnose them on a Tuesday.
Sources this session draws on
The mix arithmetic is fully shown on this page - you verified it by hand. The benchmark context comes from platform docs and panel data. This page draws on:
Three questions before you go 🎯 ◐ 90 seconds
1 · Mobile CR holds at 2.4% and desktop at 3.4%, but mix shifts from 252K/108K to 280K/80K. Blended CR...
(280,000 x 0.024 + 80,000 x 0.034) / 360,000 = 9,440 / 360,000 = 2.62%. The blend fell about 3% with zero behavior change - pure mix. Check the mix before crediting behavior.
2 · A lever fell and you are cutting to find where. The disciplined order is...
Channel first (sources break independently and often explain everything), device second (releases ship per-platform), new/returning third (acquisition quality). Stop at the first cut that isolates the drop - deeper cuts on thin data chase noise.
3 · A segment with 40 weekly conversions shows a 25% CR jump this week. You should...
The thin-slice rule: do not diagnose weekly moves on slices under roughly 100 conversions. Small-number segments swing 20-30% for free. Widen the window to 4-8 weeks before concluding anything - and never cut a thin slice thinner.