Across 26 months, 30 campaigns and 8.6 lakh unique search terms, this account turned ₹62,89,776 of spend into 5,028 purchases worth ₹2.82 crore — a blended 4.48× return. But Performance Max, on 8.1% of the budget, produced 43.8% of that revenue at 24.3×, while everything else returned 2.74×. This review reconciles eight exports and shows where the next rupee should go.
All-time totals reconciled across the overview cards, the 15.8-lakh-row search terms report, the landing page report and the conversions summary. Reach is not the constraint here — 6.7 crore impressions is a large audience for a single-showroom Bengaluru jeweller. Revenue is not the constraint either. Attribution is.
The overview cards cover 9 Jul 2024 onward (₹57,77,946). The search terms report is All Time and picks up six earlier campaigns, taking the true account total to ₹62,89,776. Of that, ₹39,36,329 (62.6%) is visible at search-term level, ₹16,63,213 (26.4%) sits in Google's hidden "other search terms" bucket, and ₹6,30,049 (10.0%) is Demand Gen, which reports no terms at all. Every figure in this report is stated against the layer it came from.
Four campaign types. Shopping takes four rupees in five and returns the most expensive conversions in the account. Search takes one rupee in a hundred and returns the cheapest — by a factor of 15.6×.
Same axis, same units — percentage of the account total. Where the gold bar overshoots the emerald bar, the channel is consuming more budget than it earns.
One measure, one colour. Search is highlighted because it is the outlier that matters.
Shopping's CPA is 15.6× Search's on 80× the budget.
Moving 10% of Shopping spend (₹5.09 L) into Search at Search's historic ₹73 CPA would, at face value, buy roughly 6,900 additional conversions — more than the account's entire all-time total. That arithmetic will not hold at scale (brand search inventory is finite, and much of Search's cheap volume is people already typing "mr jewels"), but the direction is unambiguous: the account is under-invested in captured intent and over-invested in interruption.
Six months of 2024–25 had zero delivery. From June 2025 the account scaled hard. Then August 2026 produced 3.1× July's conversions on 11% less spend — the single most important trend line in this data.
Bars, one series. Hover any month for the full set of figures.
Same x-axis as above. Plotted separately, never on a second y-scale.
Lower is better. The line ends at its lowest point on record.
1,351 conversions at ₹303 CPA against a trailing-12-month average near ₹1,200. Spend fell from ₹3,69,010 to ₹4,08,874 while output tripled. This coincides with the new Search | Brand | Bangalore campaign structure and the store PMax going live. It is a partial month (through 17 Aug) and should be re-measured at month close — but it is the cleanest before/after this account has ever produced, and it is the spine of the case study.
The account records ₹2.82 crore of purchase value all time. Performance Max produced ₹1.24 crore of it — 43.8% of all revenue on 8.1% of the budget. Every other channel combined returned 2.74×.
Revenue ÷ spend. The account blend of 4.48× is the weighted average of these two very different businesses.
Conversion value ÷ cost. Only groups with recorded revenue are plotted; the remaining seven returned ₹0 and are listed in the table.
Google's own asset-quality rating, plotted against the revenue each tier actually produced. The relationship is inverted.
"Excellent" groups returned 4.1×. "Average" groups returned 27.4×. Ad Strength measures asset completeness, not commercial fit — worth stating plainly before anyone optimises toward the score.
Feed (Sales_PMax_Hashmukh), Asset Group 1 (Pmax_Roas_Feed_5Aug), Product Based (OutsideIndia) and Performance Max-2 carry no headlines, no descriptions and no images — they run on the product feed alone. Together they consumed ₹25,212 and produced ₹39,023 of value. Every one is rated Poor, and the rating is correct.
Purchase tracking works and carries real revenue. The stages before it do not — and four goals cannot be used for bidding at all. This is the layer everything else in the account depends on.
Google's own conversion-goal grouping. A working e-commerce funnel narrows from left to right. This one inverts.
Purchases outnumber Begin checkout by 15.7× and Add to cart by 152×. In a correctly tagged store the relationship runs the other way — the cart and checkout tags are firing on a small fraction of real sessions.
Coverage across the account's 30 campaigns, and whether each goal can drive Smart Bidding.
Swipe table →Google's own banner on this screen: "4 goals cannot be used for optimization because they do not have any primary conversion actions."
1. Lead form is a primary conversion on 13 of 30 campaigns. On an e-commerce account with ₹5,609 average order value, those campaigns' Smart Bidding is optimising toward form fills rather than purchases. 1,992 of the account's 7,020 conversions — 28% — are non-purchase actions.
2. Purchase is missing from 3 of 30 campaigns and is flagged Needs attention. Any campaign without it is bidding blind on the only action that produces revenue.
3. Add to cart and Begin checkout are both Misconfigured with zero primary actions, and Begin checkout is attached to just 1 of 30 campaigns. There is no usable mid-funnel signal, which removes the obvious remarketing and value-rule levers.
The first version of this review stated that only Performance Max reported revenue and that 81% of spend had no revenue signal. That was wrong. The conversions summary shows Purchase is an account-default action carrying ₹2.82 crore across 27 of 30 campaigns — the revenue signal exists and is substantial. The real problem is narrower and more fixable: the mid-funnel goals are misconfigured, and a lead-form action is competing with purchase for bidding priority on 13 campaigns.
Both Incremental conversion value for customer acquisition and for customer retention read Not set, with 0 audience segments selected. For a jeweller with a physical showroom and a repeat-purchase category, telling Google what a new customer is worth versus a returning one is one of the cheapest available uplifts — and it costs nothing but configuration.
The full all-time search terms report runs to 15,77,944 rows. Collapsed to unique terms, the distribution is brutally concentrated at the top and almost entirely inert below it.
Cumulative share of spend as terms are added, richest first. The curve is near-vertical at the origin: a handful of queries carries the account.
The x-axis steps through named breakpoints, not linear term counts — each tick is labelled with the exact number of terms it represents.
Terms bucketed by what the searcher wanted. Buckets overlap; the base is ₹39,36,329 of term-level spend.
One brand term sits in this list. It cost ₹1.09 a click and returned 574 conversions; the other eleven averaged ₹1,540 per conversion.
Swipe table →₹3,62,482 — 9.2% of term-level spend — bought traffic on rivals' names. Two of the largest returned nothing at all.
Swipe table →1,042 terms have ≥₹300 of spend and zero conversions, totalling ₹6,33,931. Inside them sit obvious structural leaks: "1 gram gold jewellery online shopping cash on delivery" (₹3,427), "rectangular stone earrings" at a ₹1,116 CPC on two clicks, "man brooch" at ₹355 a click. These are Shopping feed matches, not keyword decisions — which means the fix is a feed and negative-list job, not a bidding one.
21,295 landing-page rows collapse to 4,188 unique pages. The efficient page type holds 4.2% of the budget.
Product pages dominate by construction — Shopping sends every click to a PDP. What matters is the comparison below.
Two measures, two panels, one axis each — never stacked on a shared scale.
Two collections do almost all of the work. Everything below /earrings/ is effectively unvisited.
Swipe table →LP report totals ₹75,49,271 against an account total of ₹62,90,669 — Shopping and PMax log both the advertiser-specified and auto-selected URL for the same click. Use these figures for relative comparison, not absolute spend.
1. The home page runs at 0.62% CTR — the worst of any destination — and is split across two URL variants (mrjewels.in and mrjewels.in/), fragmenting its data.
2. One live final URL contains an unresolved {ignore} ValueTrack parameter and has absorbed ₹32,289 across 18,561 clicks.
3. 189 /en-us/ locale variants are serving impressions at a ₹58.55 CPC, and an Angola (country=AO) parameter has taken ₹11,772 at the highest CPC of any geography in the account.
The demographic and daypart signals are consistent and actionable — and they do not match the creative or the schedule the account is currently running.
168 cells, one sequential gold ramp, light to dark by volume. The evening block from 9pm to midnight is the densest inventory in the week.
Impressions where demographics are known (25,31,371 — about 4% of total, so directional).
Mobile takes 91.7% of spend. Desktop converts at twice the rate on 7.4%. Two measures, two panels — never one shared scale.
One location: 100 Feet Road, Indiranagar, Bengaluru 560038. It receives 0.22% of the account's budget.
Driving directions cost ₹32.57 each and account for 80% of all local actions — the cheapest high-intent signal anywhere in this account.
Reconciliation gap: the overview credits Local store visits_30june with 314 conversions while the per-store report records 543 total local actions. Confirm which local actions are set as primary before either number is quoted externally.
Ranked by expected rupee impact against effort. Items 1–3 are recoverable spend already sitting in the account; 4–6 are structural; item 7 is the foundation every other number here rests on — and it is now a P0.
Search holds 1.0% of spend at a ₹73 CPA and 3.21% conversion rate, and Search impression share sits below Google's 10% reporting floor — meaning there is essentially unclaimed inventory on every non-brand query the Shopping feed is already matching. Build exact/phrase ad groups from the 2,692 converting terms, starting with polki, kundan, jadau, hasli and men's brooch.
Every term with ≥₹300 spend and zero conversions, ₹6,33,931 in total. Layer it as a shared negative list across Shopping and PMax, with a second pass for the competitor names that returned nothing — mia by tanishq, anumerton, house of parati, candere, apala.
Resolve the {ignore} ValueTrack parameter leaking into a live final URL, canonicalise the two home-page variants, and exclude the /en-us/ and non-IN country= feed variants. This is a feed and tracking-template job, not a bidding one.
Catch All and Asset Group - Open inside Sales_PMax_CatchAll produced ₹1.19 Cr of the account's ₹1.24 Cr recorded revenue on ₹3.78 L. Retire the four zero-creative feed-only groups, port the winning asset set into them, and increase budget against the structure that is already working.
/collections/studs and /collections/necklaces run 2.0–2.24% CTR at ₹3.57–3.71 CPC against 1.30% and ₹6.53 for product pages. Build Search and PMax destinations against category pages rather than single SKUs, and expand the collection set for polki, kundan, hasli and men's brooch — all of which have proven query volume and no dedicated landing page.
The Indiranagar store takes 0.22% of the budget and returns driving directions at ₹32.57 in a category where the purchase closes in person. Scale the local campaigns against a 45–64 female audience, weighted to the 9pm–midnight and 2–6pm blocks, and fix the conversion-action mapping so store actions reconcile.
Demote Submit lead form from primary on all 13 e-commerce campaigns so Smart Bidding chases purchases, not form fills. Add Purchase to the 3 campaigns missing it. Re-tag Add to cart and Begin checkout — both Misconfigured, both with zero primary actions — so the mid-funnel signal exists for remarketing and value rules. Then set the customer acquisition and retention values, which currently read Not set with 0 segments.