Retailmedia onsite sponsored products

Retail media onsite sponsored products

Build and tune the onsite search and product-page ads that sit next to the buy button on a retailer's own site: Amazon Sponsored Products and Sponsored Brands, Walmart Connect Sponsored Search, Instacart sponsored product, and Roundel Target Product Ads. These are CPC auctions where the retailer owns the purchase data, so the whole game is a keyword and bid structure that harvests converting search terms and starves the ones that only spend.

This skill runs the onsite search book. It assumes you know CPC, ACOS, ROAS, and CTR; for that math and closed-loop definitions see programmatic-foundations. To choose which retail network and surface to buy in the first place, start in retailmedia-foundations-and-networks. For audience-based display, video, and offsite on Amazon, hand off to the amazon-dsp-* skills.

When to use this skill

• "Set up Sponsored Products / Sponsored Brands for this catalog."
• "Walmart Connect Sponsored Search structure and bids." / "Instacart sponsored product setup."
• "Roundel Target Product Ads / how do onsite retail search ads work?"
• "Auto vs manual targeting?" / "Broad, phrase, exact, or product targeting?"
• "My ACOS is too high / ROAS is too low, how do I fix it?"
• "Harvest search terms." / "Bid down or negate wasted spend." / "What is TACOS?"
• "Dayparting, placement multipliers, or top-of-search bid boosts on a retail console."
• "Why is my sponsored campaign not spending / not showing in search?"

Boundaries with sibling skills:

• Which network and surface to buy (onsite vs offsite vs in-store): retailmedia-foundations-and-networks.
• Amazon DSP display, video, streaming TV, audiences, and offsite: the amazon-dsp-* skills.
• Amazon clean-room analysis and custom audiences: amazon-marketing-cloud.
• KPI math (ACOS, ROAS, TACOS, incrementality definitions): programmatic-foundations.
• Packaging retail audiences onto open-web supply as a curated deal: curation-and-curated-marketplaces.

Quick reference

Goal	Ad type to lead with	Targeting to lead with	Why
Defend branded search, own your own shelf	Sponsored Products, then Sponsored Brands	Exact keyword on your brand terms	Cheapest conversions; block competitors on your name
Win non-branded category demand	Sponsored Products	Auto first to discover, then manual exact	Auto finds converting terms; exact locks them in cheap
Launch a new ASIN or item with no history	Sponsored Products auto	Auto (all match types)	No search-term data yet; let the engine find intent
Cross-sell and defend against competitor pages	Sponsored Products product targeting	ASIN / category product targeting	Place on competitor and complementary detail pages
Build brand awareness at top of search	Sponsored Brands	Category and branded keywords	Banner and logo real estate above the fold
Efficiency is failing (ACOS too high)	Existing Sponsored Products	Tighten to exact, negate waste, bid down	Concentrate spend on proven converters

Naming ladder across consoles (same job, different labels):

Job	Amazon Ads	Walmart Connect	Instacart	Roundel (Target)
Onsite search product ad	Sponsored Products	Sponsored Search (Sponsored Products)	Sponsored product	Target Product Ads
Onsite banner / brand ad	Sponsored Brands	Sponsored Brands	Display / brand pages	Display
Product-page display	Sponsored Display	Onsite Display	Display	Display

Core process

1. Confirm retail readiness before spending a dollar. An onsite ad only converts if the item is in stock, buy-box eligible, priced competitively, and has a complete detail page (title, images, bullets, A+ content where available). Ads amplify the page; they cannot fix a page that will not convert. Full checklist in references/retail-readiness-prerequisites.md.
2. Set the account and campaign taxonomy. One structure that separates branded, non-branded category, competitor-conquesting, and auto-discovery so each has its own budget and target, because you cannot manage ACOS on a bucket you cannot see. Layout in references/campaign-taxonomy-and-targeting.md.
3. Pick the ad type per the Quick reference. Lead with Sponsored Products (the search product ad) on every network; it carries the most volume and the cleanest attribution. Add Sponsored Brands once Sponsored Products is defending your branded terms.
4. Choose targeting. Start non-branded categories on auto to discover search terms, run branded terms on exact, and use product (ASIN/item) targeting for conquesting and cross-sell. Match types and when to use each are in references/campaign-taxonomy-and-targeting.md.
5. Set the starting bid off the console's suggested-bid range, not a guess. Suggested bids reflect the current clearing price for that keyword; start near the low end of the range on discovery campaigns and near the top on must-win branded exact.
6. Set an ACOS or ROAS guardrail per campaign bucket, not one number for the account. Branded defense runs a low ACOS, category prospecting runs a higher one, launch runs the highest. Targets and the bid math in references/roas-and-acos-guardrails.md.
7. Human-gate the launch. Recommend the full structure, bids, budgets, and per-bucket targets, then stop and let a human approve the spend and go live. Never push budgets or a live bid change automatically.
8. Run the weekly optimization loop: harvest converting search terms into exact, negate wasted terms, adjust bids toward the target, and rebalance budget to the buckets hitting goal. Thresholds for each move are in references/roas-and-acos-guardrails.md.

Decision rules and thresholds

Auto vs manual targeting

• New item or new category with no search-term history: start auto. It surfaces the queries and ASINs that actually convert so you are not guessing keywords.
• Once auto has spend and clicks on a term with orders: harvest that term into a manual exact campaign and add it as a negative in the auto campaign, so the two do not bid against each other and the proven term gets a controlled bid.
• Branded terms: always manual exact from day one. You know the query and you want tight control and a low bid, since your own listings should win cheaply.

Match type ladder (search-term targeting)

• Broad: widest reach, most waste; use only in discovery, watch the search-term report daily.
• Phrase: middle ground; the query contains your term in order.
• Exact: tightest, cheapest per converting click; the destination for every harvested winner.
• Product / ASIN / item targeting: not a match type; places your ad on a specific product page or category. Use for conquesting competitors and cross-selling complements.

Harvest and negate thresholds

Use clicks-without-orders as the negation trigger and orders-at-target as the harvest trigger. A practical default on a mid-volume catalog:

• Harvest a search term to exact when it has driven multiple orders at or below your target ACOS in the auto or broad campaign, because it is now a proven converter that deserves a dedicated bid.
• Negate a search term when it has spent roughly 1.5 to 2x your target cost-per-order with zero orders (for example, 15 to 20 clicks and no sale on an item that normally converts around 1 in 8 clicks), because continued spend is very unlikely to pay back.
• Tune both thresholds to the item's real conversion rate: a 1-in-4 converter can negate faster, a 1-in-20 converter needs more clicks before you call it dead.

ACOS and ROAS guardrails by bucket

ACOS (ad spend / ad-attributed sales) and ROAS are inverses: ROAS = 1 / ACOS. Set the guardrail by intent, not one account number:

• Branded defense: lowest ACOS (own terms should convert cheaply). If branded ACOS is high, the listing or price is the problem, not the bid.
• Non-branded category / prospecting: higher ACOS is acceptable because you are buying new demand and, on many catalogs, new-to-brand customers.
• Launch / harvest: highest ACOS tolerated for a defined window to gather velocity and reviews, then tightened once organic rank builds.
• Watch total ACOS (TACOS = total ad spend / total brand sales, ad plus organic). Falling TACOS while sales grow means ads are lifting organic rank, which is the real win. Rising TACOS with flat sales means you are buying sales you would have gotten anyway.

Full bid-adjustment math and the exact move for each ACOS-vs-target case are in references/roas-and-acos-guardrails.md.

Bid modifiers and placements

• Top-of-search (or top-of-page) placement usually converts best and costs most. Use the console's placement multiplier to bid up for it only on campaigns already hitting target, so you scale winners, not losers.
• Dayparting: bid down or pause in hours that spend without converting once you have enough hour-of-day data to trust the pattern. Do not daypart a campaign still in its first weeks.

Reference material

• references/campaign-taxonomy-and-targeting.md: the full branded / non-branded / competitor / auto-discovery campaign structure, the match-type and product-targeting decision table, and the naming convention mapped across Amazon, Walmart Connect, Instacart, and Roundel. Read this when building the account or deciding how to target a specific keyword or ASIN.
• references/retail-readiness-prerequisites.md: the pre-flight checklist every item must pass before it is worth advertising (stock, buy box, price competitiveness, detail-page completeness, reviews, content), and what to fix first when a campaign spends but does not convert. Read this before launch and when ACOS is bad despite reasonable bids.
• references/roas-and-acos-guardrails.md: target ACOS/ROAS by bucket, the weekly optimization loop, the exact bid move for each ACOS-vs-target case, harvest and negation thresholds, and the TACOS read. Read this when setting targets or running the weekly tune.

Templates and examples

Example structure for a mid-size Amazon catalog (one brand, one hero product line):

• SP | Brand Defense | Exact: exact-match on your brand plus product name (for example, "acme cold brew concentrate"). Target ACOS around 8 to 12 percent, low bids, high budget priority. Job: never let a competitor buy your name cheaply.
• SP | Category | Auto: auto targeting across the catalog to discover converting queries and ASINs. Target ACOS around 30 to 40 percent while it learns. Job: find winners to harvest.
• SP | Category | Manual Exact: harvested winners from the auto campaign (for example, "cold brew concentrate", "unsweetened cold brew"). Target ACOS around 20 to 30 percent. Job: own proven non-branded demand at a controlled bid.
• SP | Conquest | Product Targeting: ASIN targeting on competitor detail pages and complements. Target ACOS higher, watch it closely. Job: steal consideration at the competitor's page.
• SB | Brand | Category + Branded: Sponsored Brands banner above search for branded and top category terms once Sponsored Products is defending the shelf. Job: brand real estate and new-to-brand reach.

Weekly loop (filled-in example move):

• Search term "nitro cold brew can" in the auto campaign spent your target cost-per-order twice over with 3 orders at 22 percent ACOS. Move: harvest to SP | Category | Manual Exact at a bid near the suggested midpoint, add it as an exact negative in the auto campaign.
• Search term "cold brew maker machine" spent 18 clicks, 0 orders, on a concentrate that has no machine. Move: add as a negative exact in auto and any broad campaign. It is the wrong intent.

Walmart Connect and Roundel mapping: the same five-bucket structure applies. On Walmart the search product ad is Sponsored Search / Sponsored Products with keyword bidding and an auction model; on Roundel the onsite search product ad is Target Product Ads on Target.com and the Target app. Instacart runs sponsored product against SKU and category context in the grocery cart flow. Keep the branded/non-branded/conquest/auto split on every network; only the labels and available match types change.

Common pitfalls

• Advertising an item that is not retail-ready. If the page will not convert (out of stock, lost buy box, uncompetitive price, thin content, no reviews), ads burn budget. Fix the page first. See references/retail-readiness-prerequisites.md.
• One ACOS target for the whole account. Branded and prospecting have different economics; a single number either starves prospecting or overpays on defense. Set targets per bucket.
• Never harvesting. Leaving winners in a broad or auto campaign means proven terms keep bidding at broad-match prices and against your own manual campaigns. Harvest weekly.
• Never negating. Auto and broad campaigns leak spend on wrong-intent queries. Negate on the clicks-without-orders threshold every week.
• Bidding up placements or dayparting on a losing campaign. Multipliers and boosts scale whatever the campaign already does; apply them only to campaigns hitting target.
• Reading ACOS without TACOS. Low ad ACOS while total brand sales are flat can mean you are just harvesting organic sales as paid. Track TACOS to see if ads are actually growing the business.
• Judging a launch campaign in week one. New campaigns and new items need time and clicks before the numbers mean anything; do not gut a launch bucket before it has data.
• Jumping to the DSP before the sponsored-search shelf is defended. Audience display and video are the next step, not the first. Own your branded and category search first, then hand off to the amazon-dsp-* skills for offsite and audiences.

Sources

• Unboxed 2025 recap: Sponsored Products video, Sponsored Brands reserve share of voice and collections, Full-funnel Campaigns (as of July 2026)
• Unboxed 2025: campaign analysis with unified reporting (as of July 2026)
• Brand+ and Performance+ overview (ROAS-driven tools) (as of July 2026)
• Amazon DSP launches Performance+ tactics (ROAS/CPA optimization) (as of July 2026)
• Walmart Connect resources and advertising products (as of July 2026)
• Instacart Ads for brands (as of July 2026)
• Roundel (Target Product Ads and retail media products) (as of July 2026)
• The Trade Desk guide to supply path optimization (as of July 2026)