A two-year-old Google Ads account, ₹62.9 lakh spent, 30 campaigns, 80% on Shopping. I took it over on 1 August 2026. Three campaigns launched, a conversion fix that cut store CPC 66%, 24 build documents, and a measurement audit that found ₹12.56 crore of value that never existed. This is the full record — the wins, the data, the mistakes, and what was handed over.
Run by a previous team for 25 months before me. None of the history below is my work — it is the starting line.
Shopping took four rupees in five and returned under two-thirds of the conversions. Search and PMax did the opposite on 9% of the money.
Shopping CPA against Search CPA. 15.6× the cost on 80× the budget. The money sat on interruption, not captured intent.
Google was the smaller channel. Every Google decision below sits inside a business that is 78% Meta.
{ignore} ValueTrack in a live URL₹32,289 of untraceable clicksMEDPitch to last build in 37 days. Three campaigns live, eleven asset and ad groups written, twenty-four documents shipped.
"Not built" means specced and handed over but not published before close. Execution after 3 September is not recorded here.
The only full month on the account. The store grew. The account grew four times faster.
₹14,900/day to 9 August. ₹533 on the 10th when the balance ran dry — about ₹32,000 of delivery lost. Then ₹33,000–40,000/day to month end. Balance exhausted again on the 31st.
| Campaign | Spend | Conv. | CPA | Call |
|---|---|---|---|---|
| S_Tofu_18july Shopping | ₹4,49,901 | 235.36 | ₹1,912 | HOLD |
| Stand_14Jan Shopping | ₹2,60,342 | 1,231.55 | ₹211 | FIX |
| Demand Gen_15jan | ₹60,982 | 35.00 | ₹1,742 | WATCH |
| Search | Brand | Bangalore · mine | ₹57,249 | 23.77 | ₹2,408 | CUT |
| Sales_PMax_CatchAll | ₹38,806 | 120.96 | ₹321 | FIX |
| PMax | Store | Bangalore · mine | ₹27,025 | 961.97 | ₹28 | SCALE |
| Demand Gen | Remarketing · mine | ₹9,999 | 0.00 | — | WATCH |
| Brand_3june | ₹3,093 | 84.98 | ₹36 | SCALE |
Channel split: Shopping ₹7,10,243 · PMax ₹81,602 · Demand Gen ₹73,050 · Search ₹60,336. PMax Store conversions are local actions, not sales.
Dashed = conversion data defective (Section 07).
The first campaign I launched read zero for 17 days. The problem was not the campaign. It was which conversion Google was told to chase.
No local action was set to Primary, so Smart Bidding ran blind and bought anonymous clicks at ₹38.60. One goal change on 21–25 Aug. CPC fell to ₹13.13, and to ₹10.13 across 18–25 Aug.
Aug 1–17 against Aug 1–25, asset group Store & Showroom 01. The highest-value single edit made to the account in the engagement.
Bars show change on the right scale: CTR and clicks up, CPC down. 34.06% conversion rate describes local actions — directions, calls, visits — never ₹38 per sale.
local actions on ₹27,025 — 3% of budget.
One month produced 2.5× the local actions of the account's whole prior life.
All three new groups landed at ₹20–29. The inherited groups ran 10× higher.
Strip out every Google conversion number. Use only Shopify revenue and the billing screen. This is the honest read of what the extra money bought.
₹6,40,702 of extra spend bought ₹19,14,731 of extra gross sales and 272 extra orders — ₹2,356 per incremental order. After the 19.8% reversal rate: 2.24× net.
Sessions up 53%, orders up 16%. The new traffic converted a quarter worse. That is buying wider, not better.
Google's clean-period data: 2.73×. Shopify incremental: 2.99×. Within 10% of each other with nothing in common.
Above ~45% gross margin the scale-up paid. Below 33% it lost money. COGS was never supplied.
Revenue tripled in a year. Efficiency fell every quarter while it did. My seven weeks sit at the far right of every chart below.
July 2026 estimated from 17 days on the within-month revenue curve, not linear. Meta spend for August 2026 was never supplied, so August does not appear here.
Dashed line: the 3.50× guard rail I set on 21 Aug. The Stage 3 doc used "hold below 15.0" — Shopify revenue over Google spend only, a trigger that could never fire. Corrected in four documents.
July 2026, restated — the worst month in two years, the month before I started.
Google share and MER correlate weakly: months with Google ≥20% of spend averaged 4.01×, the rest 3.60×. r = 0.28 — a reason to test, not a proof.
Gross peaked at ₹1.22 Cr in Feb 2026. August's ₹1.07 Cr is second-best — on more than double February's Google spend.
August at 9.82× is the floor of the year, against a 19.0× average.
From ₹1,214 in July 2024 to ₹10.88 L in August 2026. Trailing twelve months ₹71.22 L against ₹8.18 L before — 8.7×. August alone exceeds the account's first fifteen months combined. Still funded by manual ₹15,000 netbanking top-ups.
| Month | Revenue | Orders | Return % | Meta | Google % | MER | CAC |
|---|
Google reported ₹12.56 crore of conversion value in August. The whole company — every channel — took ₹1.07 crore. Most of the engagement's analysis went into finding out which numbers could be trusted.
Google claimed 11.8 times the company's revenue. Stand_14Jan booked ₹12.33 Cr between 1 and 4 August — 99.7% of its month — on ₹53,896 of spend.
Strip those four days and the account converts at about 13 a day across five campaigns. That is the real run rate.
Two traps left alone on purpose. Google's "Fix it" button would promote add-to-cart to primary — a remarketing campaign would then buy abandonment. "Set up a Begin checkout conversion" would add a third GA4 action to a duplication problem.
Coverage fell from 41.9% in December to 2.9% in February, sat at 2–4% until May, and reached 57.0% in August. Orders barely moved. Every conversion-rate trend between February and May is void, and August's checkout "improvement" cannot be separated from the tracking recovering.
Shopify reports a 0.01% conversion rate: it counted 68 checkouts against 1,960 orders, because 96.5% complete in third-party checkouts that emit no events. The true rate is 0.466%.
The account ran at 74% of its budget cap. The best campaign used 29% of its own. Budget was never the constraint — allocation was.
Label = cost per conversion. Demand Gen spent 98.5% of its cap for 29 conversions. Sales_PMax_CatchAll used 28.7% at ₹217.
Sales_PMax_CatchAll against Shopping. 103 conversions, 17.2% of the account, on 3.2% of spend. Shopping took 80.7% of spend for 33.8% of conversions.
The client tracker's TOTAL cell read ₹9,77,600 — that was the cap, not spend. Restated, not silently corrected.
| Campaign | Type | Cost | Conv. | Value | ROAS | Cost/conv |
|---|---|---|---|---|---|---|
| S_Tofu_18july | Shopping | ₹18,519 | 6.15 | ₹40,097 | 2.17 | ₹3,011 |
| Stand_14Jan | Shopping | ₹16,404 | 3.87 | ₹18,555 | 1.13 | ₹4,243 |
| Sales_PMax_CatchAll · 96.4% of cap | PMax | ₹5,759 | 4.00 | ₹20,335 | 3.53 | ₹1,440 |
| Demand Gen_15jan | Demand Gen | ₹4,025 | 2.00 | ₹4,600 | 1.14 | ₹2,012 |
| Search | Brand | Bangalore | Search | ₹4,020 | 1.67 | ₹9,505 | 2.36 | ₹2,404 |
| Demand Gen | Remarketing (C04) | Demand Gen | ₹3,599 | 0.00 | ₹0 | 0.00 | — |
| PMax | Store | PMax | ₹2,406 | 9.00 local | ₹0 | — | — |
| Brand_3june | Search | ₹37 | 2.00 | ₹6,500 | 175.50 | ₹18.52 |
| Account | ₹54,783 | 18.69 purchases | ₹99,601 | 1.82 | ₹2,931 |
Two days is not a CPA window. What is decisive at this grain: CatchAll now spends ₹2,880/day against a ₹2,987 cap, and Shopping takes 64% of spend at 1.68 ROAS. Google claims 17.0% of Shopify's ₹5,86,733 on ~20% of blended spend — coherent.
Brand_3june (inherited) vs Search | Brand | Bangalore (mine). Brand Broad 01 was buying non-brand traffic. Fix handed over: pause, head-term negatives, rebuild on phrase — ₹41,200/month.
The first auction insights the account ever produced — 26 months had been below Google's 10% reporting floor.
Two full exports: 8,60,452 unique terms all-time, and 5,05,231 for August alone. The waste was large. The sellers fit on three pages.
The two campaigns recommended from week one — Brand_3june and PMax Store — waste 11% and 22%. Everything else 86–94%.
Brand: 207 terms, 1.3% of spend, 24.8% of all conversions at ₹46. Competitors: ₹3.62 L at ₹1,215. Buckets overlap.
The remaining ₹4.4 L of August waste is a feed and structure problem. Negatives will not touch it.
mr jewels indiranagar · own store, own postcode₹3,167₹117/click₹2.41 crore reversed in twelve months. 8,396 SKUs in the feed, 478 that actually move. The ads pointed at necklaces; the business is built on kadas.
Reversed over twelve months — 19.8% of gross. About ₹21.2 L in August, twice the Google bill.
A quarter recovered is ₹5.3 L a month at zero acquisition cost.
fastrrv3 carries 75.8% of sales at 23.8% reversal — 3.6× the native Online Store. It accounts for ₹2.31 Cr of the ₹2.41 Cr.
Every line above 40% is a deity or animal pendant — they top sell-through at 100% and come back. ₹1,19,134 of August ad spend went to products with a 30%+ reversal history.
Bar = gross sales. Label = return rate. Necklace is 20.4% of revenue with the worst returns at 26.9%; brooch the best at 11.7%. Total ₹12.65 Cr, 22,242 units, ASP ₹5,688.
₹1,80,256 spent on 1,861 SKUs that returned nothing — 24.9% of Shopping. 5,200 SKUs missing age_group and gender; one feed rule releases ~5,656 products.
The Stage 3 draft sent four ad groups to studs, necklaces and kundan-jewellery — three of the worst pages on the site. Swapped before handover: brooch +73%, earrings +68%, choker-sets +119%.
Seven of ten are kadas or bangles. Only two appear in Google's top-20 spend.
A Bengaluru showroom selling to every metro. Mobile carries almost everything. The evening is the strongest block of the week.
Two peaks a day: midday and 9–11PM. 3–5AM carries 2.4% of impressions — excluded in the Stage 5 schedule.
Mobile: 96.3% of conversions, 95.1% of 35.6 L store sessions. All-time, desktop converted at 2× mobile (1.42% vs 0.70%) on 7.4% of spend.
Bengaluru is 9.0% of Indian sessions. Hyderabad, Mumbai and Delhi together send 2.6× Bengaluru. Eight metros = 45.9%. 7.4% of sessions come from outside India, led by 97,561 from the US.
Known for only ~4% of impressions — a signal, not a target. 45–54 leads at 24.6%; men are 45.5%. That is why no asset group was restricted to women: festive jewellery here is often bought for women by men.
Shopify can name the source of 3.4% of orders. That sliver is biased — and still points the same way as every other method in this study.
Separate scales — read the mismatch, not the heights.
An acquisition treadmill, not a retention business. Four in five customers are new, which is why reversals and first-order cost matter more here than anywhere.
A case study that only lists wins is a sales page. These are the decisions and readings that were wrong, what they cost, and how each was caught.
On 17 Aug I took Search | Brand | Bangalore from ₹2,000 to ₹3,100 a day. My AG04 spec, written the same day, said cap it at ₹150–200 — measured brand demand was about ₹64 a day. The surplus went into broad match buying non-brand clicks. The campaign was also on Smart Bidding, so the ₹12 ad-group caps did nothing: AG04 ran at ₹36.61.
August billing rose 143% while Stand_14Jan and Sales_PMax_CatchAll were reporting impossible value. CatchAll scaled 537% and Stand_14Jan 50% on numbers later shown to be defective. Only S_Tofu's scale-up was sound — its CPC fell to ₹4.67 on a 513% click increase.
"33 lifetime add-to-cart events, purchase tracking broken" came from the original audit and went into the Stage 4 plans, the C04 walkthrough and the blended review. It was read off goal cards that only count primary actions. Add-to-cart fired 26,737 times. A Shopify pixel rebuild and a 7 Sep scaling hard-stop were both built on it.
The top-100 search terms by cost showed zero conversions on every row. I read it as a bad export and drafted a blanket competitor exclusion. The full 5,05,231-term file showed those 70 competitor terms returning 112.5 conversions at ₹87 — among the cheapest in the account.
The first blended review scaled 17 days of July by 31/17. Returns post in one batch late in the month, so the first 17 days carry 66% of revenue, not 55%. July MER went out as 3.20× and was restated to 2.68× the same day.
The Festive & Gifting spec put Dhanteras on 26 Oct and Diwali on 28 Oct 2026. They fall on 6 and 8 November. The decision held — build early — but the runway was ten days longer than the spec claimed.
C04 Demand Gen remarketing went live with a cart audience and three ads. By 2 Sep: ₹11,971, 3,118 clicks, zero purchases, and a second ad group already specced on the same foundation. The 92 view-through conversions argued for patience; the missing stop-or-continue date did not.
Read the source, not the summary card. Conversion actions, not goal tiles. Full exports, not samples.
Verify the signal before scaling on it. Reconcile Google purchases to Shopify orders weekly; stop if Google claims more than 60%.
Restate in public. Every correction above is written into the document it corrects. Nothing silently fixed.
Every plan reached the same conclusion from a different direction: media can add 25–45 orders a day, but the missing 43 sit in the cart.
Shopping falls from 83% of purchase spend to 17%; PMax rises from 3% to 64%. Purchase media ₹8.13 L → ₹13.34 L. Only ₹5.44 L is new money.
CatchAll CPA modelled as ₹217 × (spend ÷ ₹25,821)^k, k 0.35–0.50 — 2.9× to 4.6× degradation assumed, not a flat extrapolation.
One campaign on the 478 movers across eight metros. Four asset groups: earrings 39.7% · kada 37.0% · rings 15.1% · men's brooch 8.2% of orders. Necklace excluded on 26.9% returns and a 1.09% page. Target ₹29,000/day → 25 orders base, 21 downside.
₹1,17,304 gross/day · ₹97,403 net of 17% returns · net ROAS 3.36×. Break-even at 30% gross margin. The margin figure was never supplied.
1,43,983 sessions added to cart in a year; 1,23,133 left. At 25% cart→order the store does 98.6 orders a day on today's traffic.
PMax Store logged 90 clicks-to-call; the call details report returned no rows. Specced a split build: Google forwarding numbers stay on Search call assets, Exotel numbers go on GBP, website, Meta and offline. Caller number, duration and answer rate for every source.
₹800/day ramping to ₹1,950 by week 5. Wave 1: men's brooch, polki & jadau, earrings, kadas. Wave 2 adds rings — 2,549 units a year and no ad group anywhere in the account.
Forty-two issues on the register at the August review. Sixteen were never marketing problems. Here is what the next owner inherits, ranked by money at stake.
₹2.41 Cr a year, concentrated in one checkout and one product family. Get the monthly returns report and an RTO-versus-return split. Test prepaid incentives on the 40%+ pendant lines. Ask why fastrrv3 reverses at 3.6× the native checkout.
Purchase CAPI out of Conversions. Add-to-cart set to no value. Local actions to Secondary, scoped to PMax Store. Enhanced Conversions on. One live test order, then a 7-day Google-vs-Shopify reconciliation. Only then set value to net of reversals.
Raise Sales_PMax_CatchAll in ≤50% steps — 96% of cap at 3.53 ROAS. Put Stand_14Jan (1.13 ROAS) on a 7-day review. Pause Brand Broad 01. Fund Bridal & Festive before 6 Nov.
Exclude 2,153 zero-stock variants. One rule defaults age_group, gender, color and releases 5,656 products. Custom labels 0–3 — mover, category, return band, price band — gate Stage 5 and Volume Core. Then the ≈₹50,000 negative list.
Gross margin by category — five numbers — and monthly Meta spend. Without margin, no campaign in this study can be called profitable. Without Meta, no blended MER after July exists. And move billing to auto-pay with a three-day balance alert: the account went dark twice in August.
Blended review self-score at the last audit: 68/100 — data quality 72, insight 82, financial clarity 40, decision usefulness 78, actionability 68. One number, COGS, caps it.