How Meta's AI Bidding Actually Works (2026)

Meta's bidding system gets called a black box, but it's more knowable than most advertisers think. The AI isn't guessing — it's running billions of micro-auctions per day, calculating an expected value for every impression, and deciding whether to bid based on your chosen strategy.

This post walks through each bid strategy, when to use it, and the practical tradeoffs for UK advertisers in 2026.

What happens in a Meta ad auction

Every time a Facebook, Instagram or Audience Network user loads a new surface, an auction runs. Meta's system calculates a "total value" for each eligible ad using three components:

Advertiser bid — how much you're willing to pay
Estimated action rate — how likely this user is to take your desired action
Ad quality — engagement signals, feedback, relevance

The ad with the highest total value wins. Your bid is only one input — often not the decisive one.

The four bid strategies

1. Lowest cost (no bid cap)

The default. Meta tries to get you as many conversions as possible at the lowest average cost, using all of your budget. No ceiling, no target.

Best for: most advertisers most of the time, especially when scaling new campaigns.

2. Cost cap

You set a target average cost per result. Meta will try to stay at or below that number. Some conversions may cost more, but the average should hit your cap.

Best for: accounts with a known profitable CPA that needs to be protected.

3. Bid cap

You set the maximum bid Meta will place in any auction. This is a hard ceiling. The system may spend less than your daily budget if it can't find auctions below your cap.

Best for: experienced advertisers with a precise margin calculation.

4. ROAS goal / Minimum ROAS

You set a minimum return on ad spend target. Meta optimises for transactions with value high enough to hit that target.

Best for: e-commerce brands with value-based tracking and known unit economics.

When each strategy wins

Strategy	Scale	Control	Risk of underspending	Recommended for
Lowest cost	Highest	Low	Never	Scaling campaigns, new accounts
Cost cap	Medium	Medium	Rare	Mature accounts with CPA targets
Bid cap	Low	High	Common	Tight margin products
ROAS goal	Medium	Medium	Moderate	E-commerce with value tracking

Why lowest cost usually wins

In 2026, Meta's machine learning has enough signal that artificial constraints usually hurt more than help. Cost cap and bid cap can choke delivery before the system finds its rhythm. Lowest cost lets the AI explore, learn, and converge on an equilibrium that's usually close to what you'd have capped at anyway — but at higher volume.

The exception is when you're protecting a margin. If a product only makes money below £12 CPA and your campaign drifts to £18, cost cap at £13 keeps you disciplined.

How value optimisation changes the game

If your pixel sends purchase value (not just purchase events), Meta can optimise for "highest value" instead of "most conversions." This matters for brands where AOV varies wildly. A £400 order and a £30 order shouldn't count equally.

To enable:

Ensure your pixel sends "value" and "currency" with Purchase events
Switch optimisation to "Value"
Use ROAS goal bid strategy if you need a floor

Value optimisation needs at least 50 purchase events per week to stabilise.

The learning phase in bidding

When you launch a new campaign, Meta marks it "Learning." During this window (usually 7-14 days, or until 50 optimisation events), the bidding is less precise. CPA can swing wildly. This is normal.

Every edit — budget change, creative change, targeting change — resets the learning phase. Resist the urge to tweak.

FAQ

What's the difference between cost cap and bid cap?

Cost cap targets an average cost. Bid cap sets an absolute maximum on any single auction bid. Cost cap is softer and more forgiving; bid cap is strict and can underspend.

Can I change bid strategy mid-campaign?

Yes, but it resets learning. Only change if you have strong data showing the current strategy isn't working.

Why is my CPA higher than my cost cap?

Cost cap targets an average, not a ceiling. Individual conversions can cost more. Also, during the learning phase, cost cap is a soft target, not a hard constraint.

Is Meta bidding against me in its own auction?

No. The auction is between advertisers. Meta's role is to calculate total value and pick a winner. The system does apply a "minimum bid" floor in some cases to maintain inventory quality.

What's a good CPA for UK e-commerce?

It depends entirely on your AOV and margin. Most UK DTC brands target a 3:1 to 5:1 ROAS, which translates to very different CPAs depending on AOV.

Does bid strategy affect which placements I get?

Indirectly. Lowest cost lets the algorithm find the cheapest conversions anywhere. Bid cap may force more expensive placements if cheaper ones are outside your cap.

How often should I review my bid strategy?

Monthly at most. Weekly reviews cause over-editing and learning resets.

The practical playbook

New campaign? Start with lowest cost.
Stable campaign with known CPA? Switch to cost cap to protect margin.
Brand-new category test? Stay on lowest cost until you have 14 days of data.
Scaling aggressively? Lowest cost with a big budget lift.

Pix-Vu and bidding

Pix-Vu doesn't change how bidding works — but giving the algorithm more creative variety improves the Estimated Action Rate part of the auction calculation. Better creative means higher predicted value, which means you win more auctions at lower cost. That's the fastest path to lower CPA without changing a single bid setting. Try it at pix-vu.com.