Real-Time Bidding, Auction Theory and Pricing

Introduction

Real-time bidding (RTB) is the transaction mechanism through which an advertising impression is evaluated, priced, allocated and delivered in milliseconds. It connects publishers and other supply owners with advertisers through software platforms rather than through a pre-negotiated manual sale. Although RTB is often described as an auction, its performance depends on a wider system: identity and consent signals, inventory quality, fraud controls, latency, contractual rules, bid strategy, clearing-price logic and the architecture of the supply path.

The central managerial challenge is not simply to bid more. A bidder must estimate the value of an impression, translate that value into a monetary bid, participate in an auction with incomplete information, and do so quickly enough for the ad server to return a creative before the impression expires. At the same time, publishers must balance revenue, user experience, direct-sales commitments, inventory quality and buyer demand.

This chapter analyses the RTB process from bid request to impression delivery and develops the economic logic of first-price and second-price auctions. It also examines bid shading, floor prices, pacing, win rates, supply-path optimisation, latency and market clearing.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain the RTB workflow and identify the role of each principal participant.
2. Distinguish first-price, second-price and hybrid auction outcomes.
3. Calculate clearing prices, expected cost and basic win rates under stated assumptions.
4. Explain why bid shading, floors, pacing and latency affect campaign delivery.
5. Evaluate supply-path decisions using economic, technical and governance criteria.
6. Critically assess auction transparency, market power, data use and measurement limitations.

1. The RTB Market and Its Participants

An impression is an opportunity to display an advertisement to a user in a particular context. The principal actors are:

• Publisher or supply owner: makes inventory available and sets commercial and technical rules.
• Supply-side platform (SSP): aggregates inventory, conducts or facilitates auctions, and sends bid requests.
• Ad exchange: provides a marketplace and auction infrastructure; in practice, exchange and SSP functions may overlap.
• Demand-side platform (DSP): evaluates bid requests for advertisers and submits bids.
• Advertiser or agency: defines objectives, audiences, budgets, creatives and measurement standards.
• Ad verification and measurement providers: assess viewability, invalid traffic, brand safety and outcomes.
• Ad server: selects and serves the winning creative, subject to publisher and campaign rules.

The market is two-sided. Buyers seek audiences and outcomes; sellers seek monetisation of attention. The value of a single impression is heterogeneous. It varies with device, placement, context, geography, audience signals, time, predicted conversion probability, viewability and privacy permissions.

2. The RTB Workflow

2.1 From page request to auction

A simplified sequence is as follows:

1. A user opens an application, website or connected-TV service.
2. The publisher identifies an available ad slot and invokes its ad server or header-bidding wrapper.
3. One or more SSPs create bid requests containing inventory, device, contextual, geographic, consent and auction information.
4. DSPs receive requests, apply eligibility rules and retrieve or infer relevant user and contextual signals.
5. Each DSP estimates the impression’s value, applies budget and campaign constraints, and submits a bid with a creative reference.
6. The SSP or publisher compares eligible bids, applies auction rules and determines the winner and clearing price.
7. The winning creative is returned through the ad-serving chain.
8. The impression is rendered. Tracking events may record ad request, response, render, viewability, click or conversion.

The auction is therefore not an isolated event. It is embedded in a sequence of decisions in which an apparently valid bid can fail because of timeout, creative policy, malware screening, budget controls, invalid traffic concerns or a mismatch between the bid and the final auction rules.

2.2 Anatomy of a bid request

A bid request commonly includes:

The exact fields vary by protocol and environment. A request is not a guarantee that all fields are accurate, current or legally usable. Signal quality must therefore be treated as probabilistic rather than as fact.

2.3 Bid evaluation

A bidder normally performs four tasks: eligibility, valuation, price selection and pacing. Eligibility excludes impressions that violate campaign rules. Valuation predicts the probability and economic value of an outcome. Price selection converts value into a bid under the expected auction mechanism. Pacing determines whether the bid is affordable at that moment given the remaining budget and time.

For a simple conversion campaign, an impression’s expected value can be represented as:

Here, V_i is the expected net value of impression i in currency per impression, p_i is the predicted probability of the desired outcome, R_i is the advertiser’s monetary value of that outcome, and C_i represents variable costs other than media. The expression assumes that the probability estimate is calibrated and that the outcome value is additive. Real systems may include long-term value, marginal profit, attribution uncertainty and constraints.

If the campaign is buying on a CPM basis, the corresponding value per thousand impressions is:

The factor 1000 is a unit conversion from currency per impression to currency per thousand impressions. A bid should not exceed an impression’s expected value unless the advertiser is deliberately accepting negative direct value for a strategic reason, such as learning, reach or lifetime customer value.

2.4 The response and impression

A DSP response usually contains a bid amount, creative markup or creative identifier, landing-page information and tracking instructions. The response must arrive before the exchange’s deadline. After selection, the creative may pass additional checks. The ad request, bid response, win notification and rendered impression are distinct events; confusing them produces incorrect delivery and cost reporting.

3. Auction Theory Applied to RTB

Auction theory studies the allocation and pricing of scarce goods when participants possess private or asymmetric information. In RTB, the scarce good is usually a single impression available for a very short period. Participants do not know competitors’ values, and the auction may be repeated millions of times under changing conditions.

3.1 Second-price auctions

In an idealised second-price auction, the highest eligible bidder wins but pays the second-highest bid, possibly subject to a reserve or floor. With bids ordered as b₁ ≥ b₂ ≥ ... ≥ b_n, the winner’s price without a floor is b₂. With a reserve price r, the winner pays:

This formulation assumes that the winner’s bid is at least the reserve and that the auction has a transparent, genuine second price. Historically, second-price logic encouraged truthful bidding when bidders had independent private values and the object was awarded to the highest-valuing participant. RTB environments often violate these assumptions because of floors, fee structures, multiple auctions, correlated values, identity uncertainty and strategic platform behaviour.

3.2 First-price auctions

In a first-price auction, the highest eligible bidder wins and pays its own bid:

The price equals the winning bid, so a bidder has a direct incentive to shade its bid below its estimated value while retaining a reasonable probability of winning. The optimal bid depends on the distribution of competitors’ bids, the bidder’s value, risk tolerance and campaign objectives.

For a simple independent-private-value model with n risk-neutral bidders whose values are independently and uniformly distributed on [0,1]], the symmetric equilibrium bid is:

Here, v is a bidder’s private value and n is the number of symmetric bidders. This is a theoretical benchmark, not a production rule. Its assumptions—independent values, risk neutrality, known distribution, symmetric bidders and a single indivisible good—rarely hold exactly in digital advertising.

3.3 Worked comparison

Suppose four eligible bids are $8.00, $6.50, $5.25 and $3.00 CPM, and the floor is $4.00 CPM.

• Under first price, the winner pays $8.00 CPM.
• Under second price with a reserve, the winner pays max($4.00, $6.50) = $6.50 CPM.

If the second-highest bid were $3.20 CPM, the second-price outcome would be max($4.00, $3.20) = $4.00 CPM. The floor therefore acts as a lower bound on the clearing price.

A $2.00 CPM difference is economically meaningful. At one million impressions, the gross media-cost difference is:

The calculation assumes the impressions are actually won and that both prices are reported on the same CPM basis.

4. Bid Shading and Value-Based Bidding

Bid shading is the practice of submitting a bid below a bidder’s estimated maximum value in a first-price auction. If an advertiser values an impression at $10 CPM and submits $7 CPM, the advertiser retains an expected surplus when it wins, but it may lose impressions that would have been profitable at prices between $7 and $10.

A conceptual objective is to maximise expected surplus:

S(b) is expected surplus per opportunity, v is value, b is the bid and Pr(win | b) is the probability of winning at that bid. This is not a complete campaign objective because it excludes volume, pacing, learning, reach and contractual constraints. It is useful for explaining the trade-off: increasing the bid may reduce surplus per win but increase the probability of winning.

A shading model may use auction type, exchange, placement, historical clearing prices, competition density, floor behaviour and win-rate feedback. The model must distinguish a low price caused by weak competition from a low price caused by a strict floor or a supply-path peculiarity. Over-aggressive shading can produce underdelivery and biased learning because the algorithm observes outcomes mainly on impressions it wins.

Practical controls for shading

• Maintain separate models for auction environments rather than assuming one universal discount.
• Monitor win rate and effective CPM together.
• Use exploration bids to prevent feedback loops.
• Exclude suspicious paths where the reported auction mechanics are inconsistent.
• Evaluate outcomes, not only cheap impressions; low CPM can coexist with poor viewability or conversion quality.

5. Floors, Pacing and Delivery Control

5.1 Floor prices

A floor price is the minimum acceptable clearing price for an impression. It may be static, dynamic, hard or soft. A hard floor rejects bids below the threshold. A soft floor may permit a bid below the advertised threshold under a different pricing rule, depending on platform design.

Floors can protect publisher yield, but excessive floors reduce fill and may exclude legitimate demand. A rational floor balances marginal price improvement against the probability of selling the impression. It should also account for direct deals, user experience, format quality and buyer diversity.

5.2 Pacing

Pacing distributes a campaign budget over time and eligible opportunities. If remaining budget is B_r and remaining time is T_r, a simple target spending rate is:

where s_t is the target currency-per-unit-time rate. This is a control heuristic, not a universal optimum. It assumes relatively stable opportunity volume and value. Real systems adjust for hour-of-day demand, expected conversion performance, frequency, geography, budget commitments and auction prices.

A campaign that bids aggressively early may exhaust budget before high-value periods. A campaign that paces too conservatively may finish with unused budget. Smooth pacing is not always desirable: a time-sensitive promotion may rationally concentrate spend, while a brand campaign may value consistent reach.

5.3 Win rate and effective price

Win rate is commonly defined as:

WR is win rate, W is the number of won auctions and E is the number of eligible bid opportunities or submitted bids, depending on the organisation’s definition. The denominator must be stated because win rate calculated over submitted bids differs from win rate calculated over all eligible opportunities.

Effective CPM is:

Spend and impression counts must refer to the same reporting period, currency and event definition. Win rate alone is not a performance measure. A higher win rate may result from bidding more, accepting lower-quality inventory or benefiting from weaker competition.

6. Supply-Path Dynamics and Market Clearing

The supply path is the sequence through which an impression travels from publisher to buyer. Multiple SSPs, resellers, wrappers and exchanges can create duplicated auctions, additional fees and inconsistent information. Supply-path optimisation (SPO) seeks paths that provide efficient access to quality inventory with acceptable cost, transparency and technical performance.

A buyer evaluating a path should consider:

Market clearing describes the price and allocation at which available demand and supply are matched under the rules of the marketplace. In a simple auction, the clearing price is determined by the highest eligible demand and the reserve. In a more complex publisher decision, direct campaigns, preferred deals, guaranteed commitments and multiple SSP responses may be ranked together. Therefore, the exchange clearing price may not equal the publisher’s final monetisation outcome.

Duplicate supply can create the appearance of competition without increasing unique demand. For example, the same impression may be offered through three SSPs. If a buyer bids through all three, it may compete against itself or pay different fees for identical inventory. SPO aims to reduce such inefficiency, but excessive concentration in a small number of paths can reduce resilience and bargaining power.

7. Latency as an Economic Variable

Latency is the time between an ad opportunity becoming available and the deadline for returning a usable response. It affects both technical delivery and economic value. A bid that arrives after the timeout has zero practical value, even if its predicted value is high.

The end-to-end process includes publisher processing, wrapper or exchange forwarding, DSP decisioning, identity or data calls, fraud and creative checks, response transmission and rendering. Parallel requests can reduce elapsed time, but they may increase competition, infrastructure cost and duplicated supply.

Latency should be analysed jointly with win rate and post-win render rate. A path with a high bid response rate but poor rendering may be inferior to a slightly slower path with reliable delivery. Engineering interventions include regional infrastructure, caching, fewer sequential calls, strict timeout budgets, lightweight creative assets and pre-approved creatives.

8. Measurement, Attribution and Decision Quality

RTB decisions are made under uncertainty. A bidder may optimise toward clicks, conversions, revenue, reach or attention. Each objective generates different bids. A conversion model that ignores selection bias may learn from impressions that were already favoured by earlier bidding rules. Post-auction data may be missing, delayed or affected by privacy restrictions.

Advertisers should separate:

• Auction metrics: bid rate, response rate, win rate, clearing price and spend.
• Delivery metrics: render rate, viewability, reach and frequency.
• Outcome metrics: clicks, conversions, revenue and incremental lift.
• Quality metrics: invalid traffic, fraud, brand-safety incidents and complaint rates.

Observed conversions do not automatically establish causal effectiveness. Where feasible, geo experiments, holdouts or other incrementality designs should complement platform-attributed results. The analytical unit, attribution window, currency, timezone and event definitions must be documented.

9. Governance, Privacy and Ethical Issues

RTB can involve extensive data processing and rapid distribution of user-related signals. Governance must address lawful basis or consent requirements, purpose limitation, data minimisation, retention, access controls and vendor accountability. A signal’s presence in a bid request does not prove that its use is lawful or ethically justified.

Important controls include:

1. Transmit only necessary signals.
2. Honour consent and regional restrictions.
3. Maintain approved-vendor and data-source registers.
4. Use ads.txt, app-ads.txt, sellers.json and supply-chain information to verify sellers.
5. Audit fees, auction mechanics and hidden reselling.
6. Screen creatives and landing pages for malware, discrimination and misleading claims.
7. Test models for unequal delivery or exclusionary effects across protected or sensitive groups.

Auction design can also create conflicts of interest when a platform operates exchange, supply and demand functions simultaneously. Independent measurement, clear documentation and auditable logs improve accountability.

10. Illustrative Case: A Retailer’s First-Price Transition

Background

A national retailer runs a prospecting campaign through a DSP. It has a daily budget of $50,000 and values each impression using predicted incremental profit. Its principal exchange has moved from predominantly second-price outcomes to first-price auctions. The retailer’s historical bids are therefore systematically high relative to the new pricing environment.

Problem

Spend remains on target, but the campaign’s eCPM rises by 28%, while incremental conversions increase only 4%. The team initially considers reducing all bids by a fixed percentage.

Analysis

A fixed reduction ignores differences among mobile web, in-app and connected-TV inventory, as well as differences in floors and competition. The team separates auction logs by exchange, format, placement quality and reported floor. It finds that:

• high-quality in-app inventory has dense competition and requires modest shading;
• long-tail web inventory has lower competition but inconsistent floor reporting;
• some supply paths duplicate the same publisher impression;
• delayed responses are associated with lost auctions and lower render rates.

The team calculates eCPM using spend divided by rendered impressions, rather than merely won-auction notifications. It also compares conversion value after controlling for placement and audience.

Approach

The retailer introduces a constrained shading model. Bids are capped at predicted impression value, adjusted by campaign-level pacing. It conducts controlled bid experiments across comparable inventory groups. Low-quality resellers are removed unless they provide unique supply. Timeout budgets are tightened, and creatives are pre-approved to reduce response failures.

Outcome

The campaign’s eCPM falls, but the key success is not the lowest price. Rendered impressions increase because fewer responses expire, and incremental conversion efficiency improves. A small amount of high-value inventory is retained at relatively high prices because its measured marginal value justifies the cost.

Learning points

1. A first-price transition requires auction-aware pricing, not a universal bid cut.
2. Cost must be measured at a clearly defined event: bid, win, render or billable impression.
3. Supply-path duplication can inflate apparent competition and fees.
4. Latency is part of effective bid value.
5. Bid shading must be evaluated against outcomes and delivery, not eCPM alone.

11. Best-Practice Framework

A mature RTB operating model should:

• document auction type, floor interpretation and billing events for each supply partner;
• maintain calibrated value models and explicitly record assumptions;
• evaluate bid shading by segment and monitor exploration;
• define win-rate denominators consistently;
• pace against both budget and expected opportunity quality;
• optimise supply paths using net cost, quality, uniqueness and technical reliability;
• monitor latency, timeout loss and render failure;
• reconcile platform logs with independent measurement;
• implement privacy, consent, fraud and brand-safety controls;
• review model performance for bias, drift and unintended exclusion;
• preserve audit trails for bids, auction responses, creative versions and payments.

Challenges and Limitations

RTB models are limited by incomplete information, changing auction rules, non-stationary competition, privacy restrictions and measurement error. A reported floor may not reveal the true reserve. A win notification may not equal a rendered impression. A conversion may be attributed to several media exposures. Historical data may encode prior delivery bias. Furthermore, platform-level optimisation can make causal interpretation difficult because the platform selectively chooses which opportunities to bid on.

Auction theory remains useful as a conceptual framework, but its textbook equilibria should not be treated as empirical laws. Production markets contain repeated interactions, strategic intermediaries, correlated values, mixed deal types, identity loss and operational constraints. The appropriate response is disciplined experimentation, transparent definitions and continuous reconciliation of technical and commercial data.

Chapter Summary

RTB is a rapid, distributed process that begins with an ad opportunity and ends with a creative rendered to a user. The bid request contains inventory, contextual, device, privacy and auction information, but each signal is uncertain and subject to governance constraints. DSPs evaluate eligibility, estimate value, select a bid and apply pacing before responding within a strict latency budget.

In a second-price auction, the winner generally pays the higher of the reserve and the second-highest eligible bid; in a first-price auction, the winner pays its own bid. First-price markets make bid shading economically important, but shading must be balanced against win probability, campaign volume and outcome value. Floors affect both publisher yield and buyer access. Pacing controls budget distribution, while win rate and eCPM require precise denominator and event definitions.

Supply-path optimisation considers fees, duplication, quality, uniqueness, latency and governance rather than price alone. Market clearing is shaped by auction rules and by publisher-level decisions involving direct and programmatic demand. Finally, privacy, fraud, brand safety, transparency and causal measurement are integral to responsible RTB management.

Key Terms

Real-time bidding: Automated auction-based buying and selling of individual advertising impressions.

Bid request: Data message describing an available advertising opportunity and its auction conditions.

Demand-side platform: Technology used by buyers to evaluate opportunities and submit bids.

Supply-side platform: Technology used by sellers to manage and monetise advertising inventory.

First-price auction: Auction in which the highest eligible bidder pays its submitted bid.

Second-price auction: Auction in which the highest bidder generally pays the second-highest bid, subject to applicable reserve rules.

Floor price: Minimum acceptable price for an impression.

Clearing price: Price at which the winning transaction is settled under the applicable auction rules.

Bid shading: Deliberate submission of a bid below estimated maximum value in a first-price auction.

Pacing: Distribution of campaign spending and bidding over time.

Win rate: Proportion of defined opportunities or submitted bids that result in wins.

Effective CPM: Spend divided by impressions, multiplied by 1,000.

Supply path: Sequence of intermediaries through which inventory travels from seller to buyer.

Supply-path optimisation: Evaluation and selection of efficient, transparent and quality-assured routes to inventory.

Latency: Elapsed time available for processing and returning a usable bid response.

Viewability: Degree to which an advertisement meets a defined on-screen visibility standard.

Invalid traffic: Non-human or otherwise illegitimate activity that makes an impression unreliable.

Market clearing: Matching of eligible demand and supply under marketplace rules.

Incrementality: Additional outcome caused by advertising compared with a suitable counterfactual.

Review Questions

Short Answer Questions

1. What are the principal stages in an RTB transaction from bid request to impression delivery?
2. How does a first-price auction differ from a second-price auction?
3. Define bid shading and explain why it is relevant in first-price markets.
4. Why must the denominator for win rate be specified?
5. Identify two ways in which latency can reduce the value of a bid.

Long Answer Questions

1. Explain the RTB workflow in detail and evaluate the role of the DSP, SSP, publisher and ad server.
2. Compare first-price and second-price auctions, including their assumptions, pricing incentives and practical limitations in digital advertising.
3. Discuss how floors, pacing and win rates interact in campaign delivery and publisher monetisation.
4. Evaluate supply-path optimisation using economic, technical, quality and governance criteria.
5. Explain why auction logs, rendered-impression data and conversion data may produce different assessments of campaign performance.

Analytical / Application Questions

1. Four bids are $9.00, $7.00, $5.00 and $4.00 CPM, with a $6.00 CPM floor. Calculate the clearing price under first-price and second-price rules, assuming the highest bid wins and the second-price auction uses the floor as a reserve.
2. A campaign has $18,000 remaining and 6 hours remaining. Using the simple pacing heuristic, calculate the target spending rate per hour and explain one reason why the actual rate might rationally differ.
3. A DSP observes a lower eCPM after removing one reseller path, but also observes a lower win rate. Design an analysis to determine whether the change improved campaign efficiency rather than merely reducing price.

Multiple Choice Questions with Answers

Question 1

In a first-price RTB auction, what does the winning bidder normally pay?

A. The lowest eligible bid

B. Its own submitted bid

C. The second-highest bid only

D. The publisher’s average historical CPM

Correct Answer: B. Its own submitted bid

Question 2

Four bids are $8.00, $6.50, $5.25 and $3.00 CPM, with a $4.00 CPM reserve. Under a second-price rule, what is the clearing price?

A. $3.00 CPM

B. $4.00 CPM

C. $6.50 CPM

D. $8.00 CPM

Correct Answer: C. $6.50 CPM

Question 3

Which expression defines effective CPM?

A. Impressions ÷ spend × 1,000

B. Spend ÷ impressions × 1,000

C. Spend × impressions ÷ 1,000

D. Wins ÷ eligible bids

Correct Answer: B. Spend ÷ impressions × 1,000

Question 4

What is the principal economic trade-off in bid shading?

A. Higher bids always reduce win probability

B. Lower bids can increase surplus per win but reduce the probability of winning

C. Shading changes a second-price auction into a guaranteed deal

D. Shading eliminates the effect of floor prices

Correct Answer: B. Lower bids can increase surplus per win but reduce the probability of winning

Question 5

If remaining budget is $18,000 and remaining time is 6 hours, what is the simple target spending rate?

A. $108,000 per hour

B. $24,000 per hour

C. $3,000 per hour

D. $300 per hour

Correct Answer: C. $3,000 per hour

Question 6

Why is a win-rate denominator important?

A. Because win rate is always independent of bid volume

B. Because submitted-bid and eligible-opportunity denominators measure different processes

C. Because it determines the creative file size

D. Because it converts CPM into CPC

Correct Answer: B. Because submitted-bid and eligible-opportunity denominators measure different processes

Question 7

Which factor is most directly associated with supply-path optimisation?

A. Selecting paths based only on the highest gross bid

B. Evaluating uniqueness, fees, quality, latency and transparency

C. Removing all SSPs from the transaction

D. Guaranteeing that every impression converts

Correct Answer: B. Evaluating uniqueness, fees, quality, latency and transparency

Question 8

What happens when a bid response arrives after the auction timeout?

A. It is normally unusable for that auction

B. It automatically becomes the winning bid

C. It is converted into a second-price bid

D. It raises the publisher’s floor retroactively

Correct Answer: A. It is normally unusable for that auction

Question 9

Which statement about a floor price is correct?

A. It is the maximum price a buyer may submit

B. It is a minimum acceptable clearing threshold under the stated auction rules

C. It guarantees that an impression will sell

D. It is identical to the second-highest bid

Correct Answer: B. It is a minimum acceptable clearing threshold under the stated auction rules

Question 10

Why should RTB conversion results not automatically be interpreted as causal effects?

A. Because auctions never record bids

B. Because platforms may selectively bid on opportunities and attribution may not establish a counterfactual

C. Because conversions cannot be measured digitally

D. Because CPM is always lower than CPA

Correct Answer: B. Because platforms may selectively bid on opportunities and attribution may not establish a counterfactual