D2C Command Center — history grows with every report run

17–23 Aug still contains 23 Aug, which GA4 had not finished processing at pull time (engagement 7.9% vs ~98%), so that week is understated if anything.

The 20–26 Jul row is still comparison-derived (marked *) and the 3–9 Aug sales cells are still pending — both need the Shopify API token. Google cost for 10–16 Aug is recovered from the summary-card delta, not read directly. Google ROAS for 17–23 Aug (4.44) is provisional — Google was still projecting conversion value at pull time. MER = Shopify total sales ÷ (Google cost + Meta spend); both weeks are computed from exact figures.

Cross-check against Shopify for the same 1–28 Aug window: GA4 revenue ₹1,01,96,355 vs Shopify total sales ₹1,13,04,669 (GA4 reads 9.80% lower); GA4 sessions 8,65,754 vs Shopify 8,97,358 (3.52% lower). Both gaps are in the expected direction — Shopify total sales includes shipping and fees GA4 does not count, and the two define a session differently. Shopify remains the sales source of truth; GA4 is the channel lens.

A conversion dip, not a traffic dip — sessions were almost flat while purchases and add-to-baskets both fell ~17%, and AOV rose. One day is noise; worth watching whether add-to-basket keeps falling on flat traffic. 28 Aug is excluded: engagement 8.74% vs a settled ~98% means GA4 had not finished processing it.

Viewed by Session source/medium instead of channel group, every session has a named source. Revenue reconciles to the paisa (₹26,85,772.52 vs ₹26,85,773 for 27 Jul–2 Aug) and Direct matches exactly at 40,745 — same data, different lens. The UTMs are firing. GA4 simply cannot map these mediums to any channel rule, so they fall out of every bucket.

Fix, in two moves. (1) Now, and retroactive: add rules to the custom channel group in GA4 Admin → Data display → Channel groups for adyogi/fb, adyogi/ig, whatsapp and programmatic — custom groups reprocess history. (2) Forward: change the UTMs at source to utm_source=facebook, utm_medium=paid_social, so even GA4's default grouping works.

Revenue WoW by channel, 10–16 → 17–23 Aug: Direct ▲51.8% · Unassigned ▲28.4% · Referral ▲30.6% · Marketing Ads ▲0.2% · Organic ▼10.9%. Growth is coming from everything except classified paid, which is flat.

Reconciled exactly. Placed sums to ₹1,32,87,517.15 — zero difference against the independent shopify_d2c_pull.py run, on an identical 9,638 orders. Only 2 orders of 9,638 are unattributed (99.98% coverage); 9,246 came from the order's own UTM in note_attributes, 359 from tags, 31 from Shopify's visit journey. 537 orders (5.57%) carry a refund.

The RM question, answered. Engage 360's 25.11 over-claims by 2.95× — the scepticism was right. But the conclusion survives intact: RM's true ROAS is 8.51, still 122% of its 7.0 plan, it produces the cheapest orders on the site at ₹149.82 against a ₹277.64 blend, and those orders stick — 4.4% returns, second only to Meta. It does all that on 41.7% of a ₹3L budget. Its ceiling is triggerable audience, not budget: spend the rest on more triggers, not higher bids.

Meta is half the business and the highest-quality half. ₹62,92,424 NET, 50.5% of revenue, on 71.1% of paid spend. Its 3.0% return rate and 2.4% cancellation rate are the best of any major channel, and it over-claims by only 1.25× — much the most honest of the three panels. True ROAS 3.31 against the 4.13 shown.

Exchanges are bigger than they look: 359 orders, ₹4,26,875 NET — 3.7% of all orders. They are replacements, not new demand, so they are bucketed separately and must not be credited to any channel. Site-wide, 331 orders were cancelled (3.4%) worth ₹4,42,090, and 537 orders (5.57%) carry a refund totalling ₹6,99,627 — 5.3% of everything placed.

Source: pulls/2026-09-02_revenue_aug.json and pulls/2026-09-02_revenue-by-source.md. Four classification defects, a date-quoting bug that had dropped 373 orders, and a falsy-zero bug that counted fully-refunded orders at full value were all found and fixed during this run.

Meta CPP trajectory: Jun ₹335 → Jul ₹279 → wk 27 Jul ₹238 → wk 3–9 Aug ₹335 → wk 10–16 ₹347.93 → wk 17–23 ₹348.45. Three straight weeks parked ~16% above the ₹300 ceiling, and the last WoW move was +0.15% — it is not correcting on its own. Meta absorbed 25% more spend at unchanged efficiency (spend ▲25.3%, purchases ▲25.1%), so the binding constraint is the CPP target, not audience headroom. Meta CTR fell 5.2% WoW while CPM also fell 1.5% — costs eased and engagement still dropped, which is creative wear, not auction pressure. Google went the other way: 5.2% more cost for 4.6% fewer clicks.

Same-day and same-week ad conversion values are provisional — Google labels them "pending" / "calculating projections" and Meta attribution lags roughly a day.

Meta spend ÷ all 9,638 Shopify orders = ₹197.36 per order account-wide. Meta MER against August order value (₹1,32,87,517) = 6.99.

Verified twice, not eyeballed: channel rows sum exactly to each total, and the three event counts sum to 194,568 — precisely the "Key events (All events)" figure GA4 reported independently.

August delivered 9,638 orders against a 15,846 plan (60.8%) at a site conversion rate of 1.029%. Hitting 15,846 would have needed roughly 15.4 lakh sessions against the 9.37 lakh actually delivered — or the same traffic converting at 1.69%, a 64% improvement.

→ The plan was not reachable on August's traffic at August's conversion rate. Whoever sets September's target should choose which of the two they intend to change, because the plan implicitly assumed both. Meanwhile every basket metric beat plan: AOV 106.1%, ABS 107.2%, ASP 98.9%.

This also corrects something on this page. The earlier reading of the Google gap said free organic Shopping listings were being counted inside campaign results. That literal statement cannot be right: a campaign row can only carry conversions attributed to an ad interaction with that campaign, and free listings belong to no campaign. What survives, with its financial consequence unchanged, is that Google credits itself 2,657.40 website conversions where order-level Shopify attribution finds only 1,288 August orders whose last click carried a Google Ads UTM — 2.06× the orders that arrive by a paid Google click. The mechanism is view-through, engaged-view and data-driven attribution: the buyer saw or clicked a Google ad somewhere in the lookback window, then arrived via a free listing, organic search or direct, and Google books the whole order. Paid+free order attribution (2,656) and Google's website conversions (2,657.40) land 0.05% apart, which is consistent with those extra ~1,369 claimed conversions being predominantly free-listing orders — but Google does not publish which conversions were last-click, so that remains coherent rather than proven. Two candidate explanations are now dead (Google-hosted and YouTube-hosted are both exactly zero) and one is newly proven (store visits). Order-attributed paid Google ROAS stands at 2.24 and none of this moves it.

Where the channel revenue disagrees with the panels. Meta: attributed ₹62.92L vs the sheet's ₹75.78L and the panel's implied ₹78.5L. Google: attributed paid ₹17.30L vs the sheet's ₹34.32L — but the sheet's figure is within 2.3% of paid + free listings combined (₹35.10L), which suggests the sheet is already carrying both blended together. RM: attributed ₹10.64L vs the sheet's ₹10.89L — 2.3% apart, the closest agreement of the three.

Targets above are the sheet's. Actuals on these exact boundaries are in the table below — pulled 1 Sep from the in-house sync endpoint, one request per day.

Weeks sum exactly to the month total — checked by the script, not by eye. ABS and ASP appear here for the first time; the Shopify Analytics UI never exposed unit counts. Order value is pre-refund — see the reconciliation note below.

These are exact portal pulls, not derived. They are shown separately rather than forced into the sheet's week columns.

→ Action: this is the arithmetic behind purchases landing at 62.5% of plan while reached-checkout sits at 87%. Take these two series to GoKwik and ask what changed at the payment step since June. Worth asking about directly: a payment-method list slow to load or rendering empty, a failing or newly added gateway, COD being gated (prepaid share swung 53.5% → 46.0% in a week, so payment-mix rules are moving), or an EMI/offer widget erroring. No amount of Meta or Google spend fixes this.

Pull limits, stated plainly: GoKwik's custom date picker would not open after four attempts, so MTD 1–28 Aug is not available — these are the 7D and Yesterday presets. The 7D counts are rounded to the nearest 100 by GoKwik; 28 Aug counts are exact. Also worth doing: "Session Performance" inside GoKwik needs a Google Analytics connection that is not set up.

→ Action: the abandonment journeys attack precisely the checkout leak GoKwik just exposed — they recover people who reach payment and stall. Expanding those flows is the one lever that hits both problems at once. Deliverability is the constraint: only 64.4% of sends arrive, and the Welcome journey burns 3.88L sends for 266 orders (0.07%) while Cart Abandonment Known User gets 603 orders from a tenth of the volume.

Window caveat: Engage's blocks each carry their own date control and the Journey block could not be switched off "Last 30 Days" (≈30 Jul – 29 Aug). These are therefore not a clean Aug 1–28 read, and the % columns above compare a ~30-day actual against a 28-day pro-rated plan.

Four blocks cleared on 1–2 Sep once the Shopify token and the sync endpoint were working. Two more are now buildable on request. The remaining three are reconciliation questions, not data gaps — the numbers exist on both sides and disagree, which needs a human decision about definitions rather than another pull.

→ Action: this is now measured, not inferred. Add the rules to the custom channel group in GA4 Admin → Data display → Channel groups — it applies retroactively and repairs history — then fix the UTMs at source to utm_source=facebook / utm_medium=paid_social. Until then, every channel-level budget decision is made on a chart that splits your largest paid platform in half.

→ Action: this is the cleanest growth week in the series — more traffic, better conversion, less discount. Establish what drove it before the next promo cycle buries the signal, and re-baseline September targets off this week rather than the August average.

→ Action: nothing about discounting, product mix or price realisation explains August. Three sources now reach the same conclusion from different directions: GA4 (sessions ▼14%), Shopify (orders ▼18%) and this feed (61% of the order plan while baskets beat plan). Before September targets are set, establish whether 15,846 orders was ever reachable at the traffic the site actually gets — or whether the plan was built on a conversion rate the checkout cannot currently deliver. The GoKwik payment step is losing half of everyone who reaches it.

→ Action: the month was lost at the top of the funnel, which puts it squarely on media. Meta spend is up but CTR is falling; Google is buying fewer clicks for more money. Fix reach before touching price or promo.

→ Action: Meta proved it absorbs 25% more budget at unchanged efficiency, so the binding constraint is the CPP target, not audience headroom. The cheapest fix in the account: Prospect_ASC_Volume_Remarke runs 5.03 ROAS on ₹1.74L MTD while PPM_Log takes ₹2.11L at 3.57. Moving a fifth of PPM_Log's budget to the remarketing line is a CPP improvement available this week.

→ Action: the creative refresh flagged mid-August has not landed. Also worth noting the account carries 9 unpublished edits, one on Engage_Dyna — the largest MTD spender at ₹3.26L. Ask Adyogi what is queued before it goes live mid-flight.

→ Action: ask GoKwik for a drop-off reason breakdown on 17–23 Aug. Also flagged: fulfilment jumped 68.6% (1,708 → 2,879) against a 14.8% rise in orders — that is a backlog being cleared, so any fulfilment SLA read across these two weeks is distorted.

→ Action: this will block account linking, which is exactly what the API automation plan needs. Create the passkey before starting the Shopify/Google API work.

→ Action: retire the "post-EOSS cooling" framing from planning conversations. The trough was one week, not a trend. Re-baseline August targets off the 17–23 Aug run-rate, not the 3–9 Aug one.

→ Action: work out what changed around 17 Aug — a CRM or WhatsApp push, brand demand, an app release, PR. Whatever it is, it is outperforming paid media and is currently unattributed and therefore unfunded.

→ Action: two moves. Add rules to the custom channel group in GA4 Admin → Data display → Channel groups (retroactive — it repairs history). Then fix the UTMs at source to utm_source=facebook / utm_medium=paid_social. Until then every channel-level budget decision is being made on a view that under-credits your largest paid platform.

→ Action: closing Direct's gap to organic's 0.64% is worth roughly ₹1.2L a month at current volume. Find what drove 17 Aug and whether it is repeatable. Note that part of "Direct" is misclassified paid and social, so some of this gain belongs elsewhere once the UTM fix lands.

→ Action: this needs no budget approval — the budget exists and is not being used. Drop the tROAS on PerformanceMax_All_Products, review the 97% exclusion on Slippers, and confirm the 94% / 97% exclusions on the two PMax campaigns are deliberate rather than feed-disapproval fallout. Separately: ₹1,66,555 of Google's reported conversion value is offline store visits and should come out of every D2C ROAS comparison.

→ Action: before reporting any same-day or previous-day GA4 figure, compare its engagement rate with the prior day. Single digits or teens against ~97% means the day is not ready. This is now written into the reporting SOP.

Locked sources: Shopify libertyshoes-online · GA4 “Liberty New - GA4” (289711881) · Google Ads 530-752-6249 · Meta “Liberty Shoes Online _ 1” (1476088395883569). Definitions: Conversion = Shopify orders ÷ sessions · MER = Shopify total sales ÷ (Google + Meta spend) · Meta CPP = spend ÷ omni purchases · Google ROAS = conv. value ÷ cost · AOV = Shopify's own tile (do not mix with total ÷ orders). Same-day ad values are provisional. This is the single D2C artifact — every run updates it in place; no new dashboards.