Everything above converges on a problem of scale: a catalogue of hundreds or thousands of products, each with its spend, its real sales, its margin, its search terms and its title. The unit of decision is the product, and you cannot decide about a thousand products by hand. This lesson describes the workflow with automated management —built around the Suite's Shopping Ninja— and which part stays yours.
The per-product analysis
Each run builds a table by product with:
- Spend, clicks, impressions, CTR for the period.
- Real sales (confirmed orders, not just the tag's value) and net revenue.
- Margin (by product, category or label, from the spreadsheet) → ROAS and POAS.
- Search terms it triggered and which of them were bad.
- Current title and its diagnosis (length, missing attributes).
- Status: no impressions (disapproved?), no clicks, spends without selling, profitable, hero product.
The traffic-light Dashboard
| Colour | Criterion | Reading |
|---|---|---|
| 🟢 | POAS > 1.3 and volume | Profitable: feed it (more budget to its group, protected titles) |
| 🟡 | POAS 0.8–1.3 or little data | Watch it; improve the title or the listing before touching the bid |
| 🔴 | POAS < 0.8 with meaningful spend | Loss-making: exclude it or raise its group's target ROAS |
| ⚫ | No impressions or no clicks | Disapproved, bad title, out of stock: check Merchant |
The traffic light is not calculated on revenue: it is calculated on profit. It changes the colour of many "hero products" (high on ROAS, low on margin).
The three checkbox actions
- Exclude the product from the campaign (or group): for the 🔴 ones with spend and the ones that never convert. Reversible.
- Negativise the product's bad terms (lesson 3), with the twin question already answered: the script points to the term and the title that triggered it.
- Improve the title: a proposed new title with the vocabulary that converts (the good terms) and the attributes that are missing (brand, model, key attribute), ready to copy into the feed or apply via a supplemental feed (module 6).
Nothing is applied without the checkbox; every action is recorded in a log with its date and status, and is verified on the next run (did the exclusion get applied? did the title change in the feed?).
TEST mode and preview
Before switching the actions on, TEST mode shows what the script would do with the configured thresholds (how many products it would exclude, how much spend they represent, which terms it would negativise) without touching anything. It is the simulation from the scenarios lesson (module 1) applied to Shopping: tune the thresholds until the actions are few and clear.
The split of roles
| The script | You |
|---|---|
| Cross-references spend, real sales and margin per product | Decide the margins (the truth Google does not have) |
| Calculates ROAS/POAS and the traffic light | Set the traffic-light and TEST mode thresholds |
| Proposes exclusions, negatives and titles | Approve with the checkbox; edit titles if you want |
| Verifies that each action was applied | Decide strategy: what gets excluded on margin, what stays as a hook |
| Emails you what matters | Read Monday's email |
What the workflow uncovers in the first few weeks
- The ten best-selling products are not the ten most profitable ones.
- A handful of 🔴 products concentrate 20-30 % of the spend.
- Titles with no brand or key attribute across half the catalogue.
- Repeated irrelevant terms pointing to 3-4 ambiguous words in the titles.
- Hero products that vanished from the report without anyone noticing (disapprovals).
💡 Ninja trick: the piece that makes all of this possible is the margin in the spreadsheet: a product → margin table (or category → margin, or the feed's
custom_label) that only you can fill in. With it, the Shopping Ninja turns Shopping into a profit and loss account per product; without it, it is a revenue report like any other.
What you should remember
- The unit is the product; at scale, only with automation.
- Traffic light on POAS (profit), not on ROAS.
- Three checkbox actions: exclude, negativise, improve the title; log and verification.
- TEST mode to calibrate; the margin is yours to supply.
End of Module 5. Module 6 is the data feed in depth: attributes, titles, rules and supplemental feeds.