The algorithm is not the lever. The feed is.
One obsession: making sure every product in your catalogue is bid on for what it is actually worth to the business, not for what one channel happened to attribute to it last week.
Feed'em reads your whole digital ecosystem, scores every product, sorts the catalogue into four tiers, and keeps budget moving to the tier that is earning.
You cannot bid your way out of a bad feed.
Smart Shopping and Performance Max took the bidding controls away and left you the feed. That is not a downgrade, it is a relocation: the feed became the place where the decisions are made, and almost nobody treats it that way.
The platform bids on what already converts
Advertising platforms favour products with the highest conversion rate, sometimes a small fraction of the whole catalogue, over the rest of the feed. A product with a slow or partial signal gets under-bid, converts less because it is under-bid, and stays there. The feed does not correct itself.
Two levers is not optimisation
Inside the platform you can raise or lower budget, and you can exclude a product. That is the whole toolbox, and both are manual, both act on one variable at a time, and both need someone to remember to undo them.
The conversion signal arrives late
Per product, conversion depends on each platform's attribution window. By the time the signal is trustworthy enough to act on, the demand that created it has moved, and the price that would have won the sale has moved with it.
The data lake is not a data model
Most retailers already hold the answer, spread across platforms, businesses, markets and countries, in a shape nothing can act on. Aggregating it is not a reporting exercise, it is the input the bidding machine never gets.
Where that consolidation has to be built or moved, it is built properly, with Cloud Services, rather than improvised per client.
Feed'em is the Google side of the pair, feed and budget on Shopping and Performance Max. Wandrome is the Meta side, creative and audience. It sits under Programmatic & Agentic, the same discipline applied to media buying, and the work it produces shows up in Clients & Cases. Both systems are proprietary to Wise Pirates.
The catalogue you are not bidding on is the growth you are not booking.
Three ways an unmanaged feed costs money quietly, without ever producing a bad-looking report.
The long tail never gets a turn
Products that could sell are never given enough pressure to prove it. The account looks efficient because it only ever bids on the winners it already found.
Spend keeps flowing into cooling products
A product that was a winner last season keeps absorbing budget until someone notices. Cost climbs before performance falls, so the signal to reduce pressure arrives after the money is gone.
The machine learns from whatever you feed it
Automated bidding gets better with data, and it charges you budget to acquire it. Feed it a cleaner signal and the same investment buys a better learning curve.
None of the three is a bidding problem, which is why more budget does not fix any of them. They are classification problems, and classification is what the feed can carry.
A slice of the catalogue takes most of the pressure.
When conversion rate is the only variable the platform optimises against, spend concentrates on the products that already convert. The rest of the feed is technically live and practically invisible.
Of the catalogue can absorb most of the pressure, a concentration we see repeatedly in retail accounts.
Levers the platform gives you: move the budget, or exclude the product.
Tiers Feed'em replaces them with, recalculated on the window you choose.
Score per product, aggregated across every channel in the ecosystem.
Provenance. The concentration above is a pattern we observe across retail accounts and it is stated in our own Feed'em documentation. It is not a figure published by Google, and it varies with catalogue size, margin and seasonality. Every number on this page is sourced in the FAQ.
Every product gets a verdict.
Feed'em scores each product on its aggregated transactional value and places it against a cutpoint the algorithm sets, not a threshold somebody typed into a spreadsheet. Four tiers, four campaigns, one product in exactly one of them at any moment.
Top Sellers
Maximum transactional performance, above the cutpoint the algorithm sets. These carry the growth, so they get the pressure and the headroom.
Performers
Above the average of the catalogue but not yet at Top Seller level. The tier where most upside lives, because a small push moves a product up a level.
InLine
Inside normal parameters, under permanent observation. Enough pressure to keep generating signal, never enough to bleed budget while it decides what it is.
Cooldown
Cost above the cutpoint with performance below it. The correct move is less pressure, not more budget and not deletion.
Reducing pressure is not switching off
The instinct with an expensive, low converting product is to exclude it, and the platform is happy to help. But exclusion is a decision somebody has to remember to undo, and nobody does. Cooldown is a state the algorithm revisits every cycle: the product keeps a small share of pressure, keeps producing signal, and climbs back on its own the moment its score moves, because the price changed, the season turned, or the market did. If a provider proposes pruning your catalogue, that is the question to ask them: what brings a product back, and who remembers to do it?
Score the product. Allocate the product. Fund the campaign.
Three moves, in that order, on a loop. Each one is a decision the platform will not make for you, and each one is only as good as the aggregation underneath it.
Score
- Performance pulled from Analytics, Google Ads and Merchant Market Insights
- Normalised by GTIN, so market demand and benchmark price line up with your data
- Aggregated into transactional value over a window you set, month or week to date
Allocate
- The tier is written back into the feed as a product attribute
- Merchant Center carries the label into Google Ads automatically
- Four campaigns per product category, one per tier, no manual moving of SKUs
Fund
- Dynamic budget allocation by product and performance ratio
- Pressure directed to the campaigns with the best ratio, calculated not argued
- Recalculated every cycle, so yesterday's winner does not keep last month's budget
Nothing here asks the platform to behave differently. It changes what the platform is told.
From unsupervised to semi-supervised.
Automated bidding improves with data, and today the way most advertisers buy data is by spending budget until the model works it out. That is a learning curve you are paying for twice.
The model learns alone
Left unsupervised, the platform infers value from the signal it can see in its own channel, which is late, partial and biased to whatever already converts.
The model learns with a hint
The classification is a supervised input: you are telling the model what a product is worth across the whole business before it has to guess.
Return on the learning curve
Same investment, better return, because the budget stops paying for discovery you could have handed over on day one.
Nothing gets restructured before it gets measured.
Five steps, in order. The first two produce an answer whether or not you go ahead with the rest.
Read the feed you already have
Coverage, structure, attribute quality and how spend is currently distributed across the catalogue. This is where you find out how concentrated your pressure really is, in your own numbers rather than in a benchmark.
Bring the three sources together
Analytics, Google Ads and Merchant Market Insights, joined on normalised GTINs so ecosystem performance, channel performance and market demand describe the same product.
Set the cutpoint and score the catalogue
The algorithm establishes the cutpoints, scores every product on aggregated transactional value, and writes the tier back into the feed as an attribute. You see the distribution before anything is published.
Four campaigns per product category
One campaign per tier, populated automatically from the labels, so a product moves between campaigns without anyone editing a campaign. Rolled out category by category, not in one weekend.
Recalculate, reallocate, report
Every cycle the catalogue is rescored, products move tier, and budget shifts to the best product and performance ratio. What moved and why is reported, not inferred.
Better ratio, more of the catalogue, faster growth.
Three effects, and they compound in that order: the ratio improves because pressure follows value, volume grows because more of the catalogue becomes biddable, and growth accelerates because the learning curve stops being paid for in budget.
Better ROAS
The range our own deployments report, driven by pressure moving to the products that carry the transactional value.
More volume
What opens up when the tail of the catalogue becomes biddable instead of technically live and practically invisible.
Faster growth
The compounding effect of a cleaner learning signal: the same investment buys a shorter path to the same certainty.
Provenance. These three figures are internal benchmarks from our own Feed'em documentation. They are not independently audited, they are not a guarantee, and they move with catalogue size, margin and seasonality. What we will commit to before an engagement is the audit in step 01, which gives you your own baseline rather than ours.
Want the number for your catalogue instead of ours? Start with the free feed audit →
Five parts. Skip to the one you need.
Everything below is written for the people who will connect the accounts and own the structure.
Three sources in. One score out.
The Product Feed Enhanced Management Algorithm is the part of Feed'em that is ours. It consolidates three Google sources into a single transactional value per product, over a window you control, and publishes the verdict back into the feed where the platform can act on it.
Read the diagram left to right: signal arrives, the score is computed against the cutpoint, the tier decides the campaign, and the campaign's share of budget follows the value its products carry.
Google Analytics
Performance of each product across every channel in the ecosystem: cost, views and transaction value. This is the source that makes the score a business number instead of a channel number.
Google Ads
The channel detail the platform will act on: cost, transactions, CPC, CTR and auction data, per product, so the classification knows what pressure the product is currently under.
Merchant Market Insights
Market demand and benchmark price per normalised GTIN. Without it a slow product looks weak, when it may simply be priced above the market for a week.
The output
An enriched Google data feed into Merchant Center, carrying the tier as a product attribute, which Google Ads reads as a label. No parallel system for anyone to maintain by hand.
Five layers between raw signal and a bid you can defend.
Each layer removes a specific way a product gets mispriced. Skip one and the classification inherits its error, which is why the order matters as much as the maths.
Three sources, one product key
Analytics, Ads and Merchant pulled per product and joined on normalised GTINs.
Ecosystem value, not channel value
Transaction value summed across every channel the product converts in, so paid does not get credit alone and does not get blamed alone.
Demand and benchmark price
Market demand and benchmark price per GTIN, so a drop in performance is read against the market instead of in isolation.
The threshold is calculated, not typed
The algorithm sets the cutpoints that separate the tiers, on the window you chose, so the boundary moves with the catalogue instead of with someone's memory of last quarter.
The verdict goes back into the feed
The tier is written as a product attribute, carried by Merchant Center into Google Ads, and used to populate the campaign built for it. The decision and the execution are the same object.
This is also why the audit comes first. If layer 1 cannot join your products cleanly, no amount of clever scoring on top of it means anything, and we would rather tell you that in week one than in month three.
Four campaigns, one budget that keeps moving.
Classification without budget control is a reporting exercise. The second half of the system directs pressure to the campaigns with the best product and performance ratio, calculated from the importance and the need of each campaign rather than argued in a weekly call.
Two advantages, inherent to the structure
Because the catalogue is tiered, each product is allocated to the right campaign dynamically and automatically. Because the campaigns are tiered, budget can be pointed at the best ratio between product and performance, automatically, per category.
Per product category, not per account
The four campaigns exist inside a product category, so a strong category is not funded by starving a weak one, and a weak category is diagnosed instead of quietly subsidised.
Nothing irreversible
Tiers are states and budget shares are targets. Both are recalculated every cycle, both are reported, and both can be overridden by a human who has a reason. The system has opinions, not authority.
Reporting that names the cause
Which products moved tier, which campaigns gained or lost share, and which of the three sources moved the score. A report that says what changed and why, rather than a chart of what happened.
Collect, integrate, store, analyse, distribute.
The classification is the interesting part, but it only exists because of the plumbing underneath it. Analytics platforms, advertising platforms, business intelligence, benchmark and demand, and CRM, through ETL into a warehouse, out through the algorithm, back into the feed.
The same warehouse produces the reporting, which is the reason the report and the bid never disagree: they are reading the same table. Where that infrastructure needs building or moving, it is built with Cloud Services.
Built for Google Shopping. It travels to every other feed.
Feed'em runs on Google Shopping and Performance Max today, end to end. Because the classification is written into the product rather than into a campaign, the same four tiers can drive any destination that reads catalogue fields. Each one is scoped as its own piece of work, never assumed to be free.
Google Shopping & Performance Max
Where the system runs, end to end: enriched feed, Merchant Center labels, four campaigns per category and dynamic budget.
Meta feed catalogs
The same tiers as catalogue segments for advantage catalogue and retargeting, so a Cooldown product is not being pushed on one channel while being cooled on another.
Amazon feeds
Classification applied to marketplace listings, where price benchmark and demand carry even more weight than they do on Shopping.
Retail media feeds
Tiering for the retailer networks, which is where a badly prioritised catalogue is most expensive per impression.
Pinterest feed
The same product attribute driving catalogue ads, so seasonal tiers stay consistent across discovery channels.
One classification, many destinations
A product's tier is decided once, from the whole ecosystem, and then honoured everywhere. Consistency between channels is the point, not a side effect.
A focused way to start.
Three pieces of work, in order. Each one produces something usable on its own, and each one is a decision point rather than a commitment to the next.
Feed & catalogue audit
Coverage, structure and attribute quality, how concentrated your spend is across the catalogue, and whether your three sources can actually be joined on a product key.
One category pilot
Feed'em on a single product category: consolidation, cutpoints, the four campaigns and dynamic budget, measured against that category's own before and after.
Rollout and continuous cycle
Category by category rollout, then the recurring loop: rescore the catalogue, reallocate products and budget, report what moved and which source moved it.
Scope is fixed at the audit, because the honest answer to "how long does this take" depends on how many products you have and how cleanly they join. We would rather quote it after step 01 than guess it before.
Catalogues we have handled.
Fashion, sport, beauty, grocery, electronics, books and home. Different catalogue sizes, the same problem underneath: too many products, two levers, and one channel's opinion of what each one is worth.
Provenance. Retail and brand partners of Wise Pirates, as named in our own Feed'em product documentation. Not every account listed runs Feed'em, and we will tell you which comparable case is the honest reference for your catalogue when we know its size and structure.
We built the algorithm, so we know what it cannot see.
Ours, not licensed
The Product Feed Enhanced Management Algorithm is proprietary. When the classification needs to change for your catalogue, we change it, rather than filing a feature request with a vendor.
Data people and media people
The same team builds the warehouse and runs the account. That is why the classification is something the campaigns can actually act on, instead of a model that looks clever in a slide.
Ecosystem first
Every score is aggregated across all channels before it becomes a bid. It is the same principle we apply to measurement, and the reason the report and the bid agree.
Reversible by design
Tiers are states, budget shares are targets, and every cycle is recalculated and reported. Nothing about the system requires you to trust it blindly to use it.
Certified handling
ISO 27001 and ISO 9001, with the accounts and the warehouse inside that scope. The score works on product aggregates, so no customer level personal data is needed.
One purpose
The best of digital, marketing, people and technology, aimed at driving your business results.
Certified to ISO 27001 and ISO 9001, with more than 20 security and cloud certifications concentrated in Google and Cisco, part of an agency trusted by 500+ brands since 2018.
Feed'em connects to the rest of the agency.
A classification is only worth what the media, the data and the creative around it can do with it.
Start with a free feed audit.
Before any engagement, we read the catalogue you already have and tell you where the pressure is going, at no cost. It is the same audit a paid engagement opens with.
What the free audit gives you.
Tell us a little about your catalogue and a senior specialist comes back with a genuine read. No cost, no obligation.
- How concentrated your spend is across the catalogue, in your numbers
- The tier distribution your products would produce today
- Whether your three sources can be joined cleanly, and what it takes if not
- The one category where the upside is largest, and why
A specialist reviews every request. This is not an automated scan, and we only use your details to prepare and discuss your audit.
Tell us a little through our main form and a senior specialist sets up your free feed audit.
Request my free feed audit →Goes to our main contact form, tagged so it reaches the Feed'em team.
The questions we hear most.
What is Feed'em?
How is Feed'em different from a feed management tool?
Does Feed'em work with Performance Max?
How does a product get classified?
What are the four tiers?
Is a product in Cooldown excluded?
What access does Feed'em need to our accounts?
How long before the classification is trustworthy?
Which channels and feeds can it drive?
Where do the numbers on this page come from?
Your catalogue already knows what it is worth.
Tell us what you sell and where you sell it. We will show you where the pressure is going today, and what the same budget looks like once every product is scored on what it actually earns.
Start a conversation →
