What is bid shading, and what should buyers ask about it?
Part of Supply path transparency
The short answer
Bid shading is a bidding technique that lowers a bid from the full value a buyer places on an impression toward the lowest price the bidder predicts will still win. It appeared when exchanges moved to first price auctions, where winners pay their own bid. A model estimates the winning price from past auction feedback.
Key takeaways
- Bid shading trades win rate against price: a deeper shade pays less per won impression and wins fewer of them.
- Shading models predict the minimum price to win from feedback that exchanges may share in full, in part or not at all.
- Shading is a bidding choice, separate from seller discounts, floors and bid caps, and buyers should ask who applies it.
- Because the shade varies impression by impression, it can shift delivery between sites on the same inclusion list.
What is bid shading?
Definition
- Bid shading
- The practice of lowering a bidder's bid from its estimate of an impression's value toward the predicted minimum price needed to win, so that a first price auction does not charge more than necessary.
Researchers at a demand side platform described the idea in a 2021 paper, An Efficient Deep Distribution Network for Bid Shading in First-Price Auctions (opens in a new tab). If competing bids were known, the best bid would sit just above the highest one. In practice the platform has to estimate the minimum winning price and lower the bid it would have sent under second price rules. The paper notes the technique is new to online advertising but has been used in auctions in other industries.
The core of bid shading is a trade off. The more a bid is shaded, the paper says, the better the return if it wins, but the lower its chance of winning. Researchers call the buyer's gain surplus: the gap between the unshaded bid and the price paid, zero when the bid loses. A shading model chooses, impression by impression, how much win probability to give up for a lower price.
Why did bid shading appear?
Bid shading appeared because the auction rule changed under bidders that were built for the old one. In a second price auction, the 2021 paper notes, auction theory makes bidding the true value the dominant strategy, since the winner pays the runner up's price. When exchanges and sell side platforms shifted to first price auctions, a change the paper dates to 2019, the same bids would have paid full value on every win. Platforms had to update their strategies to avoid bidding unnecessarily high. The mechanics of that switch are covered in the first price auction guide.
Shading is not only a buyer side tool. A 2020 paper on bid shading by win rate estimation (opens in a new tab) compares its own buyer side method with a shading service run by a sell side platform. So a single impression can, in principle, have its price adjusted by the buyer's platform, by the seller's, or by both, and a buyer reviewing prices needs to know which.
| Term | Who sets it | What it does | Where a buyer sees it |
|---|---|---|---|
| Bid shading | A bidding model in the buying platform, or on the sell side | Lowers the bid toward the predicted price needed to win | Platform settings and reports, if disclosed |
| Seller discount | The seller | Reduces the price after the auction; OpenRTB reports it in AUCTION_DISCOUNT_PCT and AUCTION_DISCOUNT_CPM | Win and billing notice macros |
| Market bid ratio | Calculated by the exchange | OpenRTB's AUCTION_MBR macro: clearance price divided by bid price | Win and billing notice macros |
| Floor | The seller or its ad server | Sets the lowest price that can win | bidfloor in the impression or deal |
| Bid cap | The buyer | Sets the most the campaign will bid | Line item or campaign settings |
How does bid shading estimate a winning price?
Every shading model needs some idea of the competition, and the best clue is the minimum winning price: the highest competing bid, including the floor when one is set. The 2021 paper explains that it is up to the sell side platform whether to share that price after the auction. In what researchers call open, or non censored, auctions it reaches all participants whatever the outcome; in closed, or censored, auctions the bidder learns only whether it won or lost.
OpenRTB 2.6 gives exchanges a standard way to share the number, the AUCTION_MIN_TO_WIN macro (opens in a new tab), and allows it to be left empty, for example when the publisher or the winning bidder chooses not to share price data. Google's Authorized Buyers OpenRTB guide (opens in a new tab) documents a minimum_bid_to_win field that is filled only for bids that took part in a first price auction and not for bids filtered before it.
| Feedback after the auction | What the bidder learns | Approach described in the research |
|---|---|---|
| Minimum price to win, for wins and losses | How far its bid sat above or below the winning line | Learn the best shade from shared winning prices, as a 2020 study of non censored auctions (opens in a new tab) does |
| Winning bid only | The price it lost to, but not how close a win came | One sided feedback, studied for budgeted bidding in a 2023 ICML paper (opens in a new tab) |
| Win or loss only | Whether the bid won, nothing about prices | Estimate the payoff of each possible shaded bid without minimum to win data, as the 2020 win rate study does |
| A prediction from a separate model | An estimate of rivals' highest bid, which may be wrong | Treat the prediction as a hint, a point or an interval, as a 2022 paper (opens in a new tab) does |
What should a buyer ask about bid shading?
A buyer may see a shading model only through its effects in price and delivery reports. The useful questions are about who applies it, what it learns from and how its benefit is measured. The Media Rating Council's January 2026 auction transparency standards (opens in a new tab) set a useful frame: for closed loop platforms they require disclosure of how pacing systems adjust bid levels or sit out auctions, and they say complex AI and machine learning auction systems still need high level disclosure of the rules and variables used to set prices.
Seven questions for the trading desk or buying platform
- Is shading on for this campaign? Ask whether it applies by default, to which line items, and whether it can be limited.
- Who shades? Ask whether the seller side also adjusts prices on the plan's exchanges, so one impression is not adjusted twice unknowingly.
- What feedback does the model get on each exchange? Minimum to win data, winning prices or only win and loss make different models.
- How is the saving reported? A saving measured against the unshaded bid estimates a price never paid; ask how that baseline is set.
- How does shading interact with the bid cap? Confirm whether the cap applies before or after the shade.
- What happens on fixed price deals? When a deal's at field is 3 the floor is the agreed price, and a bid shaded below it loses with OpenRTB loss code 101, bid was below deal floor.
- Can results be split by site? Ask for win rate and average price per domain, not only per campaign.
Can bid shading affect which sites win?
Yes, it can. The shade is chosen impression by impression from predicted competition, and competition differs by site. Where the model expects few rivals it may shade deeply, and when that prediction is wrong the bid loses or falls under the floor. Where competition is heavy it may shade little and win more. Over a flight, each site's share of impressions can drift from the plan even though the list never changed.
Feedback gaps make the drift harder to see. Where an exchange withholds price feedback, the model learns more slowly on the sites sold through it. OpenRTB's loss reason codes, listed in the OpenRTB 3.0 specification (opens in a new tab), help a desk tell the cases apart: 100 for a bid below the auction floor, 101 for below the deal floor, 102 for lost to a higher bid and 103 for lost to a bid for a deal.
Per site delivery check for a shaded campaign
- Impressions and spend per domain were compared with the share each site was given in the plan.
- Sites well under their planned share were checked for loss codes 100 and 101, which point to floors.
- Sites losing mainly on code 102 were reviewed for bids shaded below the competition.
- Exchanges that withhold minimum to win feedback were noted next to the sites they sell.
- Changes to bid caps or shading settings were dated, so delivery shifts can be traced.
What does a bid shading review look like on a real plan?
Fictional example
A wood stove retailer checks delivery in Newfoundland and Labrador
A fictional St. John's retailer of wood stoves and fireplace inserts runs display across 20 home improvement, energy and local news sites in Newfoundland and Labrador, in English, with a CAD budget. After three weeks, two regional news sites have delivered far below their planned share and a national home improvement site well above.
The planner asks the desk for win rate, average price and loss codes per domain. The two news sites show mostly code 102, lost to a higher bid, and are sold through an exchange that does not share minimum to win prices. Shading is on by default, and the model has little price feedback for those sites.
No site leaves the list. The desk limits shading on the line item for the regional sites, dates the change in the campaign log, and the planner adds a per site delivery check to the weekly report.
How does PlacementGraph keep the site decision separate from bidding?
A shading model decides what to pay for an impression. Which sites deserve impressions is a planning decision.
Frequently asked questions
Does bid shading guarantee a lower price?
No. It lowers the bid, and so the price of each impression it wins, but it also wins fewer impressions. Whether a campaign is better off depends on how well the model predicts competition and on the campaign's goal. The research calls this a trade off between win rate and surplus.
Is bid shading used in second price auctions?
It is not needed in the classic case. In a second price auction the winner already pays the runner up's price, and auction theory makes bidding the true value the dominant strategy. Bid shading was proposed to adjust bids built for second price auctions so they suit first price auctions.
Is bid shading the same as a discount from the seller?
No. Shading is the bidder's choice to send a lower bid before the auction. A seller discount is applied by the seller, and OpenRTB lets the exchange report it after the auction through the AUCTION_DISCOUNT_PCT and AUCTION_DISCOUNT_CPM macros in win and billing notices.
Can bid shading make a buyer lose a deal it wanted?
It can. If a shaded bid falls under the deal floor, the exchange rejects it, and OpenRTB loss code 101 says the bid was below the deal floor. On fixed price deals, where the floor is the agreed price, ask the desk to confirm shading cannot push bids below it.
Sources
- An Efficient Deep Distribution Network for Bid Shading in First-Price Auctions (opens in a new tab)Zhou et al., KDD 2021 (arXiv)
- Bid Shading in The Brave New World of First-Price Auctions (opens in a new tab)Gligorijevic et al., arXiv, September 2020
- Bid Shading by Win-Rate Estimation and Surplus Maximization (opens in a new tab)Pan et al., arXiv, September 2020
- Leveraging the Hints: Adaptive Bidding in Repeated First-Price Auctions (opens in a new tab)Zhang et al., arXiv, November 2022
- Learning to Bid in Repeated First-Price Auctions with Budgets (opens in a new tab)Wang et al., ICML 2023 (PMLR 202)
- OpenRTB 2.6 specification (opens in a new tab)IAB Tech Lab (GitHub)
- OpenRTB 3.0 specification (loss reason codes) (opens in a new tab)IAB Tech Lab (GitHub)
- OpenRTB Integration (Authorized Buyers real-time bidding) (opens in a new tab)Google for Developers
- MRC Digital Advertising Auction Transparency Standards (opens in a new tab)Media Rating Council, January 2026
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.