Portfolio audit

Executive summary

USD 1.16M/month in Google, 96.9% of it in the United States. Measured against the targets confirmed on Sep 3, four findings hold almost all the actionable value.

Portfolio structure

The split into accounts (Brand - Search, Brand - Shopping / PMAX, Brand - DGEN) is billing plumbing, not a decision unit. What gets decided is brand × channel. The in-house naming convention already makes that cut machine-readable: ONB = own brand · COMP = generic and competitor · PMAX_BRAND vs PMAX_GENERIC · MR_* = campaigns from the metabolicreviews.com comparison site.

$779,928 a month — 70% of the spend that has an ACQ reading — runs below 85% of its own target. $164,029 sits close (85–99%) and $173,657 is on or above. The two cells furthest above target, brand PMax in Ledisa and Cleantra, together carry $25,983.

Shopping Brand and Shopping Non-Brand are absent for every brand, and that is not neglect. Selmore and BlueForce are built and waiting on shopping images. More fundamentally, Shopping does not accept subscription offers under Google policy — it requires a genuine one-time purchase — which is the real constraint on every subscription-first brand in the portfolio, and the root of the BlueForce Merchant Center notice in §06.

Geographic focus

Measured on each campaign's actual location criteria (campaign_criterion.location), not on the name.

Outside the United States the portfolio is marginal (1.0% of spend) and exclusively Non-Brand Search: no international market has a brand campaign, PMax, Shopping or Demand Gen. The bottleneck is execution speed — getting Canada and Australia URLs for Ledisa took roughly a week when the only change needed was currency.

12-week trend — weekly purchase-only ROAS

Bars run Jun 8 → Aug 24 (the partial week of Aug 31 is omitted). Red = week below ROAS 1.0. Google Ads series, purchase conversions only.

The decline is broad — Search and PMax, two brands — which fits seasonality plus creative fatigue plus competition rather than one discrete mistake. Creative fatigue has a specific cause here: ads have been paused on policy grounds since July, so iteration stopped (§07).

Brand Search (ONB)

All 22 enabled brand campaigns lose impression share, and the loss is entirely to rank — 0% to budget (the one exception is 0.5% on LD_GLP_ONB_US_mCPC). That is by design: brand CPCs were reduced deliberately in pursuit of a higher brand ROAS. This section is about whether that target is reachable.

The brand ROAS ceiling

Ledisa's brand traffic, mined from 30 days of actual search terms and sorted from the cheapest query to the most expensive. The last column is the ROAS that would remain if everything below that row were removed.

The structural answer: five exact-match campaigns per brand

Today Core (1.79 at $2.87) and Reviews (1.21 at $4.62) share a campaign and a bid target, so they cannot be priced differently. Split, the core term is defended cheaply and the expensive tier gets its own ceiling. This is the only route to a higher brand ROAS that does not involve giving away the brand term.

Cleantra's brand mirrors the point from the other side: Core returns 3.96 on 958 clicks while Reviews has only 15 clicks — not because the demand is absent, but because the campaign is not pursuing it on 0.7% of the brand's budget.

Mining the brand queries PMax is already buying

Brand PMax campaigns are capturing brand queries today, with no per-term bid control, no negatives and no way to separate intent. That is exactly the inventory that should migrate into the new structure, on exact match.

The GENERIC campaigns are clean at 0.8% brand leakage, so the brand negatives are correctly placed. The work is in the BRAND campaigns: 4,788 brand clicks in Ledisa and 4,522 in Cleantra bought blind. The categories returned by campaign_search_term_insight fall straight into the five buckets — ledisa patches (2,916), cleantra lymphatic drainage (1,324), ledisa glp 1 patches reviews (267), cleantra reviews (325).

Method note: campaign_search_term_insight returns aggregated categories, not raw queries, and accepts only one campaign_id per request. It is good for sizing and for seeding the exact-match set; fine negatives get tuned afterwards from Search terms. The reverse leak is already controlled: Cleantra's and Ledisa's COMP campaigns spent $16 in 30 days on brand queries.

Where the expansion is

1. Settle the brand target first. Everything below depends on whether 5–6× is a first-order or an LTV number. Bring the ceiling table to that conversation.
2. Build the five-campaign exact structure in Ledisa and Cleantra, which are 88% of brand spend — seeded from the mined Search terms plus the brand categories PMax captures today.
3. Fund Cleantra's brand. It exists, it returns 2.36 against 0.84 for non-brand, and it gets 0.7% of the money at 25–62% impression share. Nothing needs building.
4. Create brand campaigns for Wellanova, RejuvaPaw and GL-ME. The internal benchmark holds: brand PMax returns 2.69–4.22 and brand Search 2.31–4.68 on purchase ROAS.
5. Give Reviews its own landing page, not just its own bid. It is 53% of brand spend at a $4.62 CPC against a $2.87 core — and review intent is the one brand query an editorial comparison page is actually built to answer.

PMax ↔ Merchant Center: the bridge nobody crosses

The asset groups are "Feed Only" and hand-pick product item IDs. If an ID is disapproved for the DisplayAds destination — the one that feeds PMax — the group is bidding on inventory it cannot serve. Google reports this as ordinary auction loss; Merchant Center reports it as a feed statistic; nobody joins the two.

In Ledisa the cause is copy, not product: the same physical SKU appears in around 30 title variants, and Google blocks only the ones using body-outcome language (personal_hardships_policy_violation). The same five to nine variants repeat across US/UK/CAN/AUS — fixing nine titles unblocks all 24 IDs.

Words present only in the blocked set: glp-1 (hyphenated), fat, burn, cravings, berberine. The eleven safe variants are already written and approved — nothing needs drafting, only withdrawing. Cleantra's six blocks follow no title pattern and need manual review.

Measurement: the report is inflated, the bidding is not

In two accounts, Advertorial Click (fixed $1 value) counts inside conversions and conversions_value. The biddable account-level goal is purchase only, so the algorithm optimises correctly. The damage is human: tROAS targets look met when they are not. Both ROAS columns in this table are Google Ads, not Triple Whale — see §09.

With editorial funnels carrying roughly 90% of spend, an advertorial-click signal may well be deliberate — but if so it belongs outside conversion value. Either way the targets need recalibrating: to demand a real 1.40 in Nervalley the tROAS would have to read about 1.55. Also worth stating plainly: zero bleeding — no search term above $300 with 0 purchases in 30 days, verified counting purchases only — and one primary conversion per account.

Merchant Center & Search Console

Summary of the Aug 31 phase, re-read against what the Sep 3 kickoff clarified.

Coverage gaps confirmed with the team on Sep 2: there is no Search Console property for Nervalley, BlueForce, RejuvaPaw or GL-ME, and no Merchant Center account for RejuvaPaw, Serabel or Nervalley. These are not pending permissions — the properties do not exist. That reframes one finding: Nervalley runs PMax on $4,902/30d with no Merchant Center account of its own, which needs understanding before it is scaled.

Account management

Two overlapping readings: the 28 days the API returns (Aug 4 → Sep 1) and the 24 months of the full export (21,144 changes, Sep 2024 → Sep 2026). The second corrects the first on important points, and the Sep 3 kickoff corrects both.

Fair reading

Defensible, and more of it than the short window suggested: thirteen months at 24–30 active days, no weak month; zero bleeding in search terms; one correctly configured primary conversion per account; the COMP/ONB/PMAX-brand separation and the brand negatives in generic campaigns are deliberate, clean architecture; the MCC-per-brand proposal with shared conversion learning was his and it was right; and in August he moved to the lever that actually operates under tROAS. He also inherited these accounts from another agency fourteen months ago.

What the audit cannot see, and should be asked rather than inferred: whether proposals were made and not approved, and whether the creative stop was a block imposed elsewhere. Both would change the reading, and neither is derivable from the data.

How we improve it

1. Effort proportional to spend: a fixed weekly review per account weighted by spend. Cleantra cannot receive 6% of the attention on 39% of the money.
2. The right lever per bid strategy: under tROAS you move the target and the terms; keyword bids only under Manual CPC.
3. Automate the mechanical work: a brand impression-share watchdog, a bleeding rule, a feed ↔ PMax monitor, and a daily Slack performance report with day-over-day, week-over-week, target comparison and RAG status — built on top of the existing daily data pull rather than beside it.
4. Differentiated tROAS per campaign based on real purchase ROAS, recalibrated without advertorial-click value, and aligned to the stated 1.3 / 1.6 targets.
5. Inefficient spend paused within 48 hours: GL-ME at 0.26 and MR_CL_LYMPH at 0.23 should not wait for a monthly cycle.

Prioritised action plan

1. decision needed Settle the brand ROAS target. Is 5–6× measured on first order or on subscription LTV? The ceiling table in §03 shows the bid lever cannot deliver it on first order. Everything in brand depends on this answer.
2. before Sep 6 BlueForce Merchant Center: define the real offer, remove the display:none on the disclosure, and expose a genuine one-time purchase path. Shopping does not accept subscription offers, so this is the structural fix, not a workaround. Check whether Ledisa and Cleantra share the funnel pattern — that is where USD 140k/month of PMax is exposed.
3. highest return Ledisa: withdraw or rewrite the nine blocked titles → unblocks 37% of the inventory in a $68,786/month asset group at the brand's best channel ROAS.
4. Reallocate toward brand PMax. It returns 216–296% of target in three brands on 2.4% of budget while $779,928/month runs below 85% of target. This is the largest single lever in the portfolio and it needs no new build.
5. Brand structure: build the five exact-match campaigns in Ledisa and Cleantra, fund Cleantra's existing brand campaigns, and create brand Search for Wellanova, RejuvaPaw and GL-ME.
6. Landing page velocity. Roughly 90% of spend flows through top-five editorial comparison funnels that have not been iterated in over a year, and requests to the internal team take weeks. Producing publish-ready HTML directly removes the bottleneck — starting with a dedicated page for review-intent brand traffic.
7. Measurement: remove Advertorial Click from conversion value in Nervalley and RejuvaPaw and recalibrate targets (1.40 → ~1.55; review RejuvaPaw's 1.20 against a real 0.99).
8. Cut dead spend: GL-ME Search (0.26 with a suspended feed) and MR_CL_LYMPH (0.23) — pause or justify.
9. Unblock idle capacity: shopping images for Selmore and BlueForce, and the Merchant Center ↔ Ads link for Wellanova's 199 healthy products.
10. Demand Gen: before reactivating Ledisa's seven campaigns, set their location criteria — they are named _US_ and served worldwide. The geo fix travels with the funnel rebuild, since the pause is a policy pause.

Metric glossary and traceability

Three systems measure the same thing and return three numbers. None is wrong; they answer different questions. This table fixes the name of each so there is no ambiguity — and where a figure comes from Triple Whale, it is named with Triple Whale's literal column.

Rules needed to reproduce the TW figures: spend repeats identically across the ~35 model × window combinations, so a query that does not pin both multiplies it. Derived metrics (roas, new_customer_roas, ncpa, mer) exist in the UI but not in the API's column space — compute them from the measures. Google's channel in TW is google-ads, not google. And the join between TW and Google Ads is on campaign_id, never on name.

What is still open