Why Waterbucket is different
There is one decision that separates Waterbucket from Marpipe, Hunch, Confect, Socioh, and much of the rest of the field. They remove the additional images Meta needs for context and replace them with treatments the AI reads as noise.
We don’t.
That sounds small. It isn’t. It is the difference between feeding Meta’s AI more good information and starving it of the signal it actually optimizes on. Everything on the home page follows from that single fork in the road, so it is worth the long version: exactly what happens downstream when a tool gets this wrong, and why the “test your catalog creative” pitch you have heard from Marpipe, Hunch, and the rest works against the algorithm you are paying to perform.
Pack shot
Lifestyle
In-use
Diagram
What these tools promise
Plain product-on-white catalog ads are ugly, and they all look the same. So these platforms offer to make your dynamic product ads look like real brand creative: branded frames, lifestyle backgrounds, price badges, promo flags, video. Then they promise testing, different catalog creative at different stages of the funnel, served through different ad sets.
It is a genuinely appealing story. Your catalog ads stop looking like a spreadsheet. You get “creative control.” You get an A/B testing narrative to bring to the next performance review. These are capable, well-built products run by smart people, and we are not here to tell you they don’t work. We are here to tell you what they actually do under the hood, because almost nobody explains the mechanism, and the mechanism is the entire story.
Every one of these tools reaches the same place by the same road
A supplemental feed that overwrites the image fields in your catalog. Your catalog has two image fields that matter. image_link holds the primary photo, and additional_image_link holds the set of extra images Meta can draw from, up to twenty per product. The tool generates “treatments,” your product re-skinned with a frame or background, and points both fields at their images instead of yours.
Look at what that wipes. A typical product carries a pack shot, a few angles, an in-use or lifestyle image, maybe an instructional diagram. Each tells Meta something distinct. The overwrite model replaces all of them at once, usually with variations of a single hero dressed in different backgrounds. To Meta’s AI, those read as substantially the same image, repeated.
And the setup instructions say it out loud: connect the supplemental feed, prioritize it for image and additional image link, and do it inside your existing single catalog, because Meta penalizes you for spinning up new ones. That last detail is the trap. Meta consolidates learnings on one catalog and warns that a new one means lower ROAS. So the move looks logical: don’t make a new catalog, just overwrite the images in the one you have. It preserves your event learnings. The catalog ID stays the same. But it quietly destroys the thing Meta’s modern delivery actually runs on.
You just starved the algorithm
There isn’t one algorithm. There are several, stacked, and they don’t all want the same thing from you. To see the cost of an image overwrite, you have to see all five.
Three things happen at once when a supplemental feed replaces your images, and all three cut against the system you are paying for. You collapse the menu twice. Within the product, twenty shots become twenty near-duplicates. Across placements, Meta no longer gets to pick the angle for the surface where it would have won. You homogenize the signal Andromeda reads, because every product wearing the same frame flattens the creative signal across your whole catalog. And you can’t un-bake it. A treatment burned into the feed is permanent, so Meta can’t peel your frame off to test the clean photo, because the clean photo no longer exists.
It isn’t that the treatments look bad. They often look great. It is that you have stripped Meta’s context and homogenized its targeting signal in exchange for a fixed design you chose in advance.
Who’s holding your performance data?
Some of these platforms enjoy active support from Meta: co-marketing, partner status, lift studies. On the surface that looks like a credential. Ask why the support flows and the picture changes. They earn that partnership because they manage the account. Several don’t just supply a feed. They operate inside your ad account, run your campaigns, and buy your media from their own interface. To do that, they need standing access to your performance data, your audiences, your conversion events.
Strip the friendly label off and what is left is a third party with a permanent seat between you and Meta, holding your first-party data. Three sets of incentives now sit in your account: yours, Meta’s, and a vendor whose partner status depends on keeping certain behavior flowing.
By default, Waterbucket delivers a feed and never touches your customer data. No seat in your ad account, no harvesting of Meta’s performance data, no managed third party between you and the auction. Want us to run your ads? That is an optional managed service you choose on your own terms.
What Waterbucket does instead
The algorithm is the most expensive asset in your account. So don’t blind it. Feed it. The whole thing is one concept: overlay, not overwrite. Our competition replaces your contextual images with dressed-up copies of the hero. We overlay each image and enrich it. The lifestyle shot stays a lifestyle shot and now carries a price. The in-use photo still shows the product in use and now carries proof.
The test is simple. After their pass, is your in-use photo still in the catalog? No. After ours? Yes, and it is working harder.
The home page shows what those enriched angles look like in motion: Price, Promo, Proof, and real BNPL, each served to the right person. This piece is about the part underneath that you never see in the ad: why the catalog those angles ride on stays clean, owned, and readable by Meta.
The right way to test runs in the exact opposite direction
The “A/B test your catalog creative” story is testing done at the worst possible layer, in the worst possible way. There is a better hierarchy, and these tools sell you the bottom of it.
Pull the actual image out and test it raw vs. treated, same audience, same budget. It isolates the one variable you care about, touches nothing in your catalog, and costs you zero optionality. The only method that changes nothing except the thing you are measuring.
Meta prefers one catalog so you don’t contaminate the asset it has spent months learning on. A quarantined test catalog respects exactly that: clean data, a real head-to-head, without poisoning production.
What the supplemental-feed model does by default, and the one method that fails on every axis. No holdout to measure against, Meta’s image context stripped, and the exact asset its single-catalog guidance tells you to protect, contaminated.
The overwrite model vs. Waterbucket
Questions, answered
Those tools generate treatments and use a supplemental feed to overwrite your image fields, so your real catalog shots get replaced with look-alikes of a single hero. Waterbucket leaves your images in place and enriches them. The same outcome on the surface, the opposite effect on the algorithm underneath.
Overlay, not overwrite. We add a layer to an image rather than swapping its subject, so your lifestyle, in-use, and instructional shots stay in the catalog as the context Meta reads for targeting, each now carrying a little more. The overwrite tools do the reverse.
Meta’s retrieval engine reads your creative as the primary signal for who sees the ad. Replacing every product’s distinct shots with one uniform treatment strips Meta’s image context and homogenizes that signal, degrading the exact diversity the algorithm uses to find audiences.
Pull the actual images and run them as static, single-image ads in a controlled ad set, raw vs. treated, same audience and budget. It is the only method that changes nothing except the variable you are measuring, and it leaves your catalog untouched.
Several of these platforms operate inside your ad account and hold your performance data to do it. Waterbucket delivers a feed and never touches your customer data by default. Running your ads is an optional service you can choose, never a requirement.
Yes. Waterbucket’s image and video overlay rendering is covered by two issued U.S. patents (11,899,656 and 12,455,874), with further patents pending. It is owner-operated and built to stay, with no investor pressure toward fire-sale exits, feature sunsets, or forced migrations.
The whole overwrite model is built on one bet: that you, in advance, can out-guess Meta’s algorithm about which creative to show which person, and that the way to act on that guess is to overwrite your catalog with a fixed treatment.
Waterbucket is built on the opposite bet. The algorithm is the asset. Your job isn’t to fight it for control of the creative. It is to give it more good options and a clean catalog to learn on, then get out of its way. Enrich, don’t strip. Add optionality, don’t collapse it. Keep your context, your account, and your data.
Work with the algorithm, not against it.