TL;DR

What is Meta’s Andromeda, and why does it change how Meta campaigns work?

Andromeda is Meta’s AI delivery system that learns who to show your ads to based on creative signals, not the audiences you manually define. That shift means more creatives per campaign is now an advantage, but it also means the algorithm acts faster and with more precision than ever before.

If Meta’s Andromeda is so smart, why did my ROAS tank while purchases tripled?

Because Andromeda only controls who sees your ad, not what it’s optimizing for. That’s determined by your bid strategy, and when a “highest volume” objective met a product mix with wildly different price points, the algorithm did exactly what it was told: it chased the cheapest, easiest sale and ignored everything else.

What should ecommerce businesses actually do differently with Meta’s Andromeda?

Before touching campaign structure, get clear on your objective. If your products vary significantly in price, “highest value” tells the algorithm to chase revenue, not just volume. In extreme cases, splitting campaigns by ticket range is about giving each objective the right instruction set.

Andromeda Knows Who to Show Your Ad To. But Who Decides What to Optimize For?

A lot has already been said about Andromeda, the AI system that redesigned how Meta delivers ads. The conversation usually revolves around campaign structure: less manual targeting, more creatives per campaign, let the algorithm learn.

That advice is correct, but incomplete.

What almost no one is talking about is the other side of the equation: bid strategy. And it was exactly that gap that cost a client a 28x return on a spend (ROAS), which dropped to 11x in less than three weeks, while the number of purchases nearly tripled.

This is the story of an accidental experiment that became one of the most instructive lessons about how Meta works today.

What Meta’s Andromeda Is

Andromeda is not just an improved targeting algorithm. It is an architecture change. Before it, Meta required the advertiser to do most of the segmentation work: you defined audiences, exclusions, interests, and the system delivered to that slice.

With Andromeda, the system learns at the creative level. It observes who interacts with which ad, in which context, at which moment, and redistributes delivery in real time based on those signals. Targeting became a consequence of the creative, not the other way around.

What this changes in practice:

Closed and hyper segmented audiences become learning bottlenecks
More creatives per campaign = more data = better optimization
The tightly granular structure that once protected manual control now fragments the signal and weakens the algorithm

The logic is clear: trust manual targeting less, invest more in diverse creatives, and let the AI figure out who buys.

But there is one piece Andromeda does not control, and that piece defines everything about what the system is trying to maximize.

The piece Meta’s Andromeda does not solve: the bid objective

Andromeda is exceptional at answering one question: Who on Meta, is most likely to take the action you want?

But who defines which action you want, and how you want to maximize it, is the bid strategy.

These are two different systems operating in layers:

System	Function
Andromeda	Who sees the ad and when
Bid strategy	What to optimize within that delivery

When both layers are aligned, the campaign performs extraordinarily well. When they are not, Andromeda will be very efficient at doing the exactly wrong thing.

The Case: When the Algorithm Performs Perfectly and the Result is Underwhelming

The account in question is a Canadian ecommerce store specializing in artisanal cookware. The campaign (an Advantage+ Shopping campaign: Meta’s AI-driven format that consolidates budget and creative into a single learning environment) ran multiple video creatives, all pointing to high value products that cost anywhere from $250 to $3,000+. High average order value, good margins, solid ROAS.

The bid strategy: maximize conversion volume.

On February 19, a new creative was added: a video about Pocket Knives. A different product: significantly lower average ticket, but strong visual appeal and high impulse potential. The kind of product people buy without much deliberation.

In the weeks that followed, the numbers moved.

Before vs. after (Jan 29 to Feb 19 vs. Feb 20 to Mar 13)

Metric	Before	After	Change
Purchases	95	270	▲ 184%
Cost per purchase	$13.43	$7.12	▼ 47%
Total spend	$1,275	$1,924	▲ 51%
ROAS	28.02x	11.48x	▼ 59%

On the surface: a win. Meta flagged it with a performance badge: “your cost per purchase is 78% lower than similar advertisers.” More purchases, lower CPA, the dashboard green.

But ROAS dropped nearly 60%.

What the algorithm did

When the creative breakdown was opened for the period after Feb 19, the story became explicit:

Creative	Spend	Purchases
Pocket Knives	$1,915.94	265
Creative product 1	$3.27	4
Creative product 2	$0.46	1
Creative product 3	$0.61	0
Creative product 4	$1.02	0
Creative product 5	$2.38	0

Andromeda concentrated 99.6% of the budget on the product with the lowest ticket.

And it was right. According to the objective it had been given.

“Highest volume” means: find the path of least resistance to generate the highest number of purchase events. The cheaper product converts more easily, more frequently, to a larger audience. The algorithm did exactly what it was instructed to do.

The mistake was not Andromeda’s. It was brief it had been given.

The misalignment: the problem nobody is naming

Here is the central tension of the Andromeda era for ecommerce:

The system has become so good at optimizing for the declared objective that a poorly chosen objective now has much faster and more severe consequences than before.

When targeting was manual and delivery was slower, a slightly wrong objective was buffered by the other constraints of the campaign. Today, with Andromeda learning in real time and redistributing budget dynamically, misalignment gets amplified, fast.

This creates a specific trap for ecommerce stores with varied product mixes:

You add creatives for products with different ticket prices to the same campaign, which Andromeda encourages, since more creatives mean better learning.
You use “highest volume” because you want volume
The algorithm finds the product that converts most easily and concentrates everything there
Your volume metrics look great; your value metrics collapse

It is a trap built with good intentions.

How to align bid strategy with Andromeda

The right question before setting up any campaign today is no longer “how do I segment my audience?” It is: what am I asking this algorithm to optimize, and does that reflect what I actually want?

Here are the main scenarios and how to think through each one:

When to use “Highest Volume”

This is the right objective when:

Your product mix has similar ticket prices
You are building data volume in the early learning phase, the period when Meta’s algorithm needs enough purchase events to understand who buys from you
The product with the lowest ticket still has healthy margins

The risk: if you have products with very different tickets in the same campaign, the algorithm will always prioritize the cheapest one.

When to use “Highest value”

This is the right objective when:

You have products with varying ticket prices and want the algorithm to prioritize revenue over volume
Your tracking pixel (the small piece of code on your site that reports purchases back to Meta) is configured to pass the purchase sale amount correctly (mandatory prerequisite)
You want fewer purchases at higher value over many purchases at low value

In this case, “highest value” would have been the correct choice from the start. Andromeda would have learned to prioritize the premium knives, even with the pocket knives creative in the campaign.

When separating campaigns by ticket range still makes sense

Sometimes separation still makes sense, not for audience segmentation, but for objective alignment. If the ticket difference between products is very large (for example, a $30 product and a $400 product), running different objectives for each group is more strategic than trying to solve it through value strategy within a single campaign.

This runs counter to the “consolidate everything” approach that Andromeda has popularized, but there are situations where separation makes sense. The distinction is important: you are not segmenting audiences here, you are separating by business objective.

What changed and what stayed the same

Andromeda changed a lot about how to structure campaigns on Meta. But the business logic behind bid strategy did not change, it became more consequential.

What changed:

Manual targeting became a bottleneck, not a control lever
More creatives per campaign produce better results, not worse ones
Granular structure by funnel or audience fragments the signal

What stayed the same:

You still need to define what you want to optimize
The algorithm still optimizes literally for the declared objective
A product mix with very different ticket prices still needs strategic attention at the bid level

The Andromeda era did not eliminate the need to think about bid strategy. It made that decision more consequential, because now the system executes with much greater speed and precision.

Conclusion

Andromeda is genuinely powerful. But power aimed in the wrong direction generates efficiency at the wrong target.

The system worked exactly as designed. The problem was the objective it was given. “Highest volume” with a product mix of very different ticket prices is an instruction the algorithm will follow, and you will watch volume metrics climb while ROAS collapses.

The lesson is not to distrust Andromeda. The lesson is to understand that the system operates in two layers that need to be aligned: the delivery layer, which Andromeda handles well, and the objective layer, which is still your responsibility to define correctly.

When those two layers are talking to each other, the result is what Meta promised: more performance with less manual work. When they are not, you have a high-performance algorithm optimizing in the wrong direction, and doing so very efficiently.

Book a clarity call to find out if your Meta bid strategy is costing you revenue while your volume numbers look fine.