Ranking, demand and yield: one system: Demand Engine — Connect demand sources. → External Demand Governance — Govern outside demand with a relevance bar and ad-load caps. → Unified Ranking — One ranking pass, with gradient boosting along the relevance curve. → Yield Optimizer — Measure what each placement earns.

Governance sits between the demand you connect and the single ranking pass, and its results show up in placement yield.

External demand governance

External demand governance: keep the outside demand, keep control of your shelf.

The retailer sets the rules. PA gates and ranks. The outside stack plugs in behind. No forced slotting.

Outside ad stack (Demand · unchanged): Advertiser demand, auction, campaigns, budgets and billing stay with the licensed or plugged-in partner.

Particular Audience (Relevance gate & ranking): A relevance bar sponsored tiles must clear, placement inside an agreed ad-load envelope, and one ranking across organic and sponsored results.

Retailer (Storefront & rules): Search hierarchy, ad-load caps, own-brand priority, labelling and approval stay with the retailer.

Auction over relevance: On a short shelf, bid order can push the item the shopper came for below the fold. A somewhat-relevant ad can win slot 1 on bid alone.

Lost decisioning: The signals the retailer built for its own baskets may not carry into a licensed ad stack.

Crowded shelves: More advertiser demand on the same few slots means more sponsored tiles and fewer answers.

Ad load creep: Ad load becomes the only dial. Sponsored density rises and conversion falls.

Split measurement: Direct and outside campaigns reported separately, with no single incrementality view.

Conquesting by block list: A manual list of top keywords misses misspellings and near-variants, and brand-vs-brand conquesting can't be controlled inside the outside stack.

Name confusion: "Crown" could mean a whisky, a cola or a soap. A dog food search should never show cat food. (Made-up examples.)

Money left on the table: No ads on product pages, and duplicate "buy again" tiles when an ad wins the organic item's slot. A gift search serving a canned staple in the top ad slot (made-up example).

Organic and sponsored results need one relevance policy across the page.

Top-slot gate: slot 1 has a higher relevance bar than slot 2; ads below the bar still compete lower in the grid, inside the existing auction.

Ad-load envelope: with governance, sponsored tiles stay within a per-surface cap (illustrative: 2 of 6 top slots).

Nothing here changes who clears advertiser demand. No partner scores or models are requested — only room to place.

The ask: Room to place ads on the retailer's own pages; A relevance bar for sponsored tiles; An ad-load envelope by surface; One incrementality read-out across direct and outside campaigns, against a control.

Signals to track: Sponsored share of top slots; Search exits and no-result rate; Conversion on sponsored vs organic tiles; Incremental sales vs a holdout.

Compounding flywheel (illustrative): Relevance in every placement → Higher click-through and conversion → Stronger advertiser returns → Advertiser re-investment → More niches and places to spend; turn 1, Head terms: Relevant ads on the busiest searches earn more clicks, and advertisers see better returns. turn 2, Long-tail niches: Re-invested budget reaches thousands of niche searches that keyword lists never covered. turn 3, New surfaces: Product pages, cart, buy-again and out-of-stock alternatives open more places to spend, and the loop widens again.

Gradient-boosted search: sponsored items are nudged along the organic relevance curve in proportion to relevance, not pasted into fixed slots; low-relevance items stay buried and an item at rank 100 can't reach slot 1. Decisioning is where gradient boosting lives. Proof: Bosch & TCL gradient boosting — Share of voice +36%, Share of clicks +121%, Share of sales +11.8%, Avg click position 11.4 → 9.6 (One reported deployment on an electronics retailer; results vary.).

Long tail: Chris Anderson's The Long Tail argued that the many low-demand niches, added together, can match or beat the few hits. In retail search, keyword-bought ads cluster on head terms like "dog food". Relevance matching lets the right ad serve on "grain-free dog food for senior labradors" too. Each niche search is small; together they are most of the searches. Source: Chris Anderson, "The Long Tail", Wired (October 2004); book, 2006 (https://www.wired.com/2004/10/tail/). dog food → senior labrador grain-free → Raised slow-feeder bowl; coffee → decaf oat-milk pods → Milk frother; tyres → 17-inch all-season EV tyres → Tyre pressure monitor; paint → bathroom anti-mould white → Extractor fan; baby → newborn reflux bottle → Bottle steriliser. More niche searches become sellable without building new keyword lists. Smaller suppliers can win niches they would never outbid on head terms. Shoppers in niches see ads that answer them, so the relevance bar protects trust.

Now — Gate and grow: Relevance gate on today's fixed top slots; More long-tail searches made monetisable; New surfaces: buy again, product pages, cross-sell, out-of-stock alternatives, cart.

Next — The full grid: Sponsored items nudged up from their organic rank in proportion to relevance; An item at slot 100 can't buy its way to slot 1; Boosts by attribute, category or term.

Later — Agentic: Sponsored placements inside AI chat, governed by the same relevance rules.

Daily eligibility: The outside stack asks PA once a day. PA returns eligible items per search term; the outside stack caches them. Simplest and lowest latency. Search only.

PA calls the stack: PA makes the ad call with relevance-filtered candidates. The outside stack still ranks by auction. Broader surfaces: buy again, cross-sell, upsell and out-of-stock alternatives as distinct ad products.

PA runs the auction: The outside stack shares its demand; PA runs the auction. Ad sets and bids arrive in batches. PA runs a hybrid relevance-and-bid decision across all onsite inventory. The most complete option.

Merchant budgets join as a new bidder. Own-brand and merchant budgets join the same decision through Promotion Hub "ghost bids": an own-brand product carries a bid equal to its value to the retailer, and a supplier ad wins only if it beats it. Commercial value competes only after relevance and policy qualify. Relevance is never sold to the highest bidder.

PA notifies AdOps and advertisers on pacing; floors, caps and makegoods stay the retailer's decisions. Start with a joint pilot on two or three categories.

External Demand Governance

Keep demand from outside ad platforms while the retailer keeps control: relevance gate, ad-load envelope, one ranking and one incrementality read-out.

Works with

• Demand Engine — Connect demand sources through a governed auction.
• Unified Ranking — Relevance and yield scored in the same pass.

Add next

• Yield Optimizer — Forecast inventory and model CTR uplift.

Outcomes it moves

• Profit Rate Optimisation
• AdOps & Trading Efficiency