Marketplace advertising changed when Amazon turned ads from a small seller tool into a retail media machine. The useful lesson is not “run more ad types”. It is sharper than that: once spend reaches roughly €5K per month across Amazon, bol and MediaMarkt, the account should stop acting like a bid-management queue and start acting like a learning agenda.
That sounds a little academic. It is not. A learning agenda is simply the list of questions your ad budget is allowed to answer this month. Can this SKU defend branded demand profitably? Can we buy category discovery without draining the hero product? Does MediaMarkt traffic convert into profitable orders or just cheaper clicks? Should bol.com BE get its own budget, or is NL carrying the numbers? If the team cannot name the question, the campaign is probably just buying noise.
The named mistake here is the “busy account”. It has new keywords, new bids, new placements, a weekly report and plenty of screenshots. Everyone looks active. But after four weeks, nobody can say what was learned. The €6,000 budget bought impressions, not decisions. That is expensive theatre.
For managed marketplace advertising, especially when an external operator owns Amazon, bol and MediaMarkt spend, the better rule is simple: every euro must buy either profit or evidence. If it buys neither, it should not survive the next weekly review.
The gap competitors usually miss: retail media history is an operating problem
Most Amazon advertising history pieces cover the same milestones: Sponsored Products grew up, Headline Search Ads became Sponsored Brands, Amazon DSP matured, video arrived, AMC opened more measurement options and retail media became a serious competitor to Google and Meta. That story is useful, but it leaves operators with a weak conclusion: “the channel is bigger now, so brands need more sophistication.” True, but incomplete.
The practical consequence is that marketplace ad accounts now mix three different jobs inside one dashboard:
- Harvesting demand: protecting searches where shoppers already want your brand or exact product.
- Creating demand: using generic category, competitor, video or audience placements to reach shoppers earlier.
- Proving permission: testing whether a SKU has enough margin, stock, offer quality and conversion power to deserve more spend.
When those jobs are managed with the same ROAS target, the account lies. Branded defense looks brilliant because it captures demand that was already warm. Discovery looks weak because it carries the cost of learning. A retargeting placement can report a lovely return while mostly claiming credit for shoppers who were already on their way back. A low-ROAS generic campaign can still be valuable if it teaches you which search term deserves content, pricing or stock investment.
This is where FiveX’s Advertentie Service needs a different lens from a generic PPC agency. We are not just asking, “What bid should this keyword have?” We are asking, “What decision will this spend unlock for the marketplace P&L?” That question changes the entire account structure.
Start with a decision ledger, not a campaign list
Before creating another campaign, write the decision ledger. It is a small table with five fields: SKU, marketplace, question, budget limit and decision date. That table prevents the operator from turning every interesting idea into permanent spend.
Here is the difference.
A normal campaign plan says: “Launch Sponsored Products for three priority SKUs on Amazon and bol, add competitor targeting, test higher bids, report ROAS weekly.” Fine. But vague.
A decision ledger says: “For the LumaHome Air Fryer 4.2L, spend up to €1,200 over 14 days to test whether generic ‘air fryer compact’ demand can produce at least €4.80 contribution margin per paid order after ad cost. Decision on day 15: scale, isolate, rewrite listing or stop.” Much better. The operator knows what success means before the first click arrives.
FiveX’s marketplace analytics and profitability dashboards are useful here because the ledger needs real SKU economics, not just ad-platform revenue. A campaign can show 480% ROAS and still fail if commission, fulfilment, coupon cost and expected returns leave €0.40 margin per order. The dashboard is not decoration; it is the permission layer for the test.
Example 1: LumaHome learns that category clicks are not the problem
LumaHome sells a compact air fryer on Amazon.nl and bol.com. The team has €7,500 monthly ad spend and wants to “push category visibility”. The old approach would raise bids on generic keywords until ROAS gets uncomfortable. The learning-agenda approach is stricter.
The test question: can generic category demand produce profitable first orders without depending on a coupon?
- SKU margin before ads: €13.20 per unit.
- Loaded break-even ACOS after fulfilment, commission and expected returns: 28%.
- Test budget: €1,400 across Amazon generic Sponsored Products and bol Sponsored Products.
- Success rule: at least 60 paid orders, blended ACOS below 24%, and total SKU contribution margin above €300 after ad cost.
After two weeks, the dashboard shows mediocre news: Amazon generic ACOS is 31%, bol NL is 22%, bol BE is 39%. A bid-only operator would lower Amazon and BE bids. The learning ledger finds the real issue: Amazon clicks are converting poorly only on the black variant, where the main image still shows old packaging and the price is €3 above the category median. bol NL works because the white variant has better price position and 22 days of stock cover.
The decision is not “pause generic”. It is: keep bol NL scaling, freeze bol BE, rewrite Amazon creative for the black variant, and retest with a €450 cap. That is a better outcome than a pretty ROAS table because the budget bought an operational decision.
This is also where FiveX inventory insights matter. Scaling the winning bol NL lane would be silly if the white variant has six days of cover. The ad decision needs stock permission, otherwise the campaign buys a stockout and then calls it growth. We try not to be that dramatic before coffee.
Separate harvest, learn and scale budgets
A €5K+ marketplace ad account should not have one blended budget. It should have at least three budget lanes.
- Harvest budget: protects branded, exact-product and high-intent terms where the SKU already has proof.
- Learn budget: funds experiments with strict caps, short windows and written questions.
- Scale budget: increases spend only after profit and evidence thresholds are met.
The ratio depends on maturity, but a practical starting point is 55/30/15 for stable brands: 55% harvest, 30% scale and 15% learn. For launches, flip it closer to 30/50/20, but only if stock and margin can absorb the volatility. The mistake is letting learning budget sneak into harvest campaigns because the ROAS looks nicer there. That creates a calm dashboard and a weak business.
FiveX advertising automation and AI recommendations can help with bid movement inside these lanes, but the lane label should come first. Automation is brilliant at enforcing rules. It is dangerous when nobody has defined what the campaign is allowed to prove.
Example 2: NordTrail stops paying MediaMarkt to answer an Amazon question
NordTrail sells laptop backpacks across Amazon.de, bol.com and MediaMarkt. The operator sees MediaMarkt CPCs at €0.38, much cheaper than Amazon’s €0.91, and wants to move budget there. Sensible instinct. Still, cheap clicks are not a strategy.
The decision ledger asks: does MediaMarkt create incremental profitable demand for the 15.6-inch backpack, or does it mainly serve bargain comparison traffic?
- Monthly spend available: €9,000.
- MediaMarkt test cap: €900 for 21 days.
- Required evidence: 120 clicks minimum, conversion rate above 3.2%, return rate below 8%, and contribution margin per paid order above €6.
- Stop-loss: pause if spend reaches €450 with fewer than 6 orders.
By day 10, the campaign has 1,070 clicks, 24 orders and a reported ROAS that looks acceptable. But FiveX’s profit view shows the average order includes a promotional shipping subsidy and a higher return reserve. Contribution margin after ad cost is only €2.10 per order. Meanwhile Amazon competitor targeting is more expensive but produces €8.40 contribution margin per paid order on the same SKU.
The decision is not that MediaMarkt is bad. The decision is that MediaMarkt should not answer the scale question yet. It gets moved into a cheaper learning lane with tighter product-page targeting, while Amazon receives the scale budget. Without the learning agenda, the team would have celebrated low CPC and moved money away from the more profitable channel.
Use incrementality language without pretending every account has a lab
Everyone loves talking about incrementality now. Good. The industry needed it. But not every marketplace account has the data volume, clean control groups or AMC setup for perfect causal measurement. Smaller NL/BE accounts still need practical decisions next Monday.
So use an incrementality ladder:
- Level 1: blended movement — did total SKU revenue and contribution margin move when spend changed?
- Level 2: paid versus organic mix — did ads grow total demand or just shift organic sales into paid sales?
- Level 3: market or placement holdout — can NL, BE, branded, generic or PDP placements be paused safely for a short window?
- Level 4: advanced measurement — use AMC, DSP reporting or retailer measurement when volume and access justify it.
The operator voice matters here: do not let imperfect measurement become an excuse for lazy measurement. If you cannot run a perfect holdout, you can still compare total SKU contribution margin before and after a spend change. You can still tag branded campaigns as harvest instead of pretending they created all revenue. You can still refuse to scale a retargeting campaign that reports 900% ROAS while total SKU sales stay flat.
Example 3: Mysa proves a “bad” keyword is actually a content brief
Mysa sells espresso-machine descaling tablets. On bol.com, the keyword “ontkalkingstabletten koffiemachine” spends €320 in 10 days with 18% ACOS, which looks good. The broader keyword “koffiemachine schoonmaken” spends €260 with 46% ACOS, which looks bad. A normal optimization pass would reduce the second bid.
The learning agenda asks a better question: is the broad term a profitable ad target, a content gap or the wrong shopper?
- Broad-term test budget: €300.
- Landing evidence: product page conversion rate 2.1% versus 7.8% on the exact term.
- Basket evidence: buyers who convert have €5.60 contribution margin after ads, because they often buy two packs.
- Decision: not a scale keyword yet; turn it into listing content and a bundled offer test.
The keyword failed as a bid target, but succeeded as a merchandising signal. Shoppers searching for “koffiemachine schoonmaken” are earlier in the problem. They need reassurance about compatibility, frequency and machine safety. The next action is not a bid tweak. It is a product-page FAQ, a comparison image and a two-pack bundle. After that, the operator can retest with a €150 cap.
This is the type of work a marketplace advertising service should surface. Ads are often the fastest way to discover what the listing, price, bundle or inventory plan has not solved yet.
The weekly ritual: turn campaign reports into decisions
A useful weekly ad meeting should not start with “ROAS is up”. It should start with the decision ledger.
For each open test, answer four questions:
- Did the test reach enough data to make a decision?
- Did it pass the profit threshold, not just the platform ROAS threshold?
- Did stock, Buy Box, price position or returns distort the result?
- What is the next label: scale, keep learning, fix the SKU, harvest only or stop?
That last label is important. “Monitor” is usually just procrastination wearing a nice shirt. If a campaign has not earned more budget, say so. If a SKU needs a price change before ads continue, say so. If branded defense is profitable but not incremental, keep it in harvest and stop presenting it as growth.
FiveX’s value in this workflow is the connected cockpit: ad spend, marketplace revenue, SKU margin, stock, repricing context and exportable decisions in one place. For an Advertentie Service client, that means the operator can show not only what changed, but why the next euro is moving.
What to put in the first 30-day learning agenda
If you are taking over a marketplace ad account above €5K spend, do not start with 40 micro-tests. Start with five decisions that matter.
- Branded defense: which branded terms are mandatory protection, and where can bids be reduced without losing total SKU revenue?
- Generic category permission: which SKUs can afford discovery traffic after contribution margin and returns?
- Channel role: which marketplace should harvest, learn or scale for each hero SKU?
- Offer readiness: which ad failures are actually price, content, review, Buy Box or delivery failures?
- Budget elasticity: where does an extra €500 produce profit, and where does it only produce more attributed sales?
This gives the client a useful first-month outcome even before the account is “fully optimized”. They get a map of where money is allowed to work, where it needs evidence and where it should stop. That is more valuable than another spreadsheet with green and red percentages.
The bottom line
Amazon’s advertising evolution taught the whole marketplace world a lesson: retail media can become enormous when ads sit close to the transaction. But proximity to the transaction also makes attribution tempting. It becomes very easy to mistake claimed revenue for created profit.
That is why marketplace ad management needs a learning agenda. Harvest campaigns protect what already works. Learning campaigns buy evidence. Scale campaigns receive budget only when SKU economics, stock, offer quality and demand signals agree. The operator’s job is to keep those roles separate.
If your Amazon, bol or MediaMarkt account is spending €5K+ per month, ask one blunt question this week: what did last month’s ad spend teach us that changed a budget decision? If the answer is vague, your next optimization task is not a bid change. It is a better decision ledger.