// Tracking and data

How Digital Ad Algorithms Work from Retrieval to the Auction

Ad systems pick a winner in 3 stages: retrieval, ranking, and auction. Read the 3 factors in Meta total value and where creative work fits in.

How Digital Ad Algorithms Work from Retrieval to the Auction

An ad system picks a winner in three stages: retrieval, ranking, then auction. Retrieval filters millions of candidates down to thousands. Ranking scores the candidates that remain. The auction picks the one ad that appears.

This page uses Meta as the example, because Meta publishes an official explanation for all three stages.

Three stages of ad selection

01RetrievalThe engine filters ad candidatesinto a short list.02RankingA model scores every candidatethat passed.03AuctionThe highest total value wins theplacement.

Scroll the diagram sideways

Every chance to show an ad runs all three stages. Meta states that billions of auctions happen each day.

Meta calls its retrieval engine Andromeda. Meta reports a 6 percent recall gain in the retrieval system and an 8 percent ad quality gain on selected segments. The ranking layer uses a foundation model named GEM.

Three factors that decide the auction

Meta states that the auction winner is the ad with the highest total value, subject to a price floor. Total value comes from three factors.

FactorMeaningWhat you control
BidWhat you are willing to pay for the outcomeBid strategy and budget.
Estimated action rateThe chance a person takes your desired actionConversionConversionThe action you count as a result, for example a paid order, a sign-up, or a chat that becomes a qualified buyer.Open the glossary signal quality and offer fit.
Ad qualityA quality measure built from people feedbackCreative, honest claims, and the landing pageLanding pageThe page an ad points to. It carries 1 offer and 1 action, with no menu pulling the visitor elsewhere.What is a landing page?.

Meta states that clickbait and engagement bait do not improve ad performance. Both behaviours lower the quality measure.

Why creative became the main lever

Retrieval and ranking run automatically at large scale. Manual targeting lost part of its role, because the engine selects the audience from far more signals. What stays in your hands is the creative, the offer, and the quality of your conversion signal.

Old leversNarrow interest listsMany small ad setsManual schedulingCurrent leversCreative varietyClean conversion signalsA clear offer

Scroll the diagram sideways

Clean conversion signals raise the estimated action rate. Read the Conversions API guide.

What to work on

  1. Send correct conversion events, with value and currency.
  2. Prepare creatives that truly differ, not five copies of one idea.
  3. Match the ad promise to the landing page content.
  4. Give the campaign learning time before you change the budget.
  5. Change one thing at a time, then record the change date.

Questions and answers

Does a higher bid win the auction? The bid is only one of three factors. An ad with a low estimated action rate and low quality can lose to an ad with a smaller bid.

Why do results move after a budget change? A budget change alters how many placements the system pursues. The system needs to learn again, so measure results after that learning period ends.

Does a small audience help? An overly narrow audience shrinks the candidate pool at the retrieval stage. Costs then rise, because competition for that small group becomes tighter.

// references

Official sources

We checked every claim on this page against the documents below on 11 September 2026. Open the source before you change a live system.

// next step

Need help to apply this?

Send your target. We check the context and write the first step before implementation.