Quick answer
Real-time bidding, or RTB, is a programmatic transaction in which an ad opportunity is described in a bid request and eligible buyers respond with bids under a short deadline. A publisher page or app initiates an ad request through an ad server, SSP or exchange. A DSP evaluates available fields, campaign eligibility, predicted outcome value, budget, frequency, creative and supply controls, then returns a price and creative reference or declines. The seller applies auction and publisher rules and the winning eligible ad is served. Auction type, floors, fees, bid shading, identity, privacy signals and supply paths affect the clearing outcome, so the highest submitted price is not always a complete explanation of cost or delivery. RTB is not synonymous with all programmatic advertising. Manage it with transparent inventory, authorized sellers, log-level diagnostics, value-based outcomes, pacing and frequency, and causal measurement. A higher win rate or lower CPM can reflect weak inventory rather than better performance.
What is real-time bidding?
Real-time bidding is an automated auction process for individual advertising opportunities. When eligible inventory becomes available, seller systems send a structured request to potential buyers, who decide whether and how much to bid before a short deadline.
RTB enables impression-level decisions across many publishers and campaigns. It is one part of programmatic advertising. Guaranteed and preferred programmatic deals can automate transactions without putting each impression through an open real-time auction.
Real time describes the transaction relative to the ad request, not an absence of planning. Campaigns, creative, data permissions, models, budgets and supply rules are configured before each opportunity arrives.
Follow the RTB transaction step by step
A page, app or media player creates an ad call. A publisher ad server, SSP, header-bidding system or exchange packages an opportunity and distributes a bid request to connected demand systems. The exact route can include several hops.
A DSP reads available fields, checks campaign eligibility and predicts value. It returns a bid response with price, creative and required metadata or sends no bid. The seller applies auction, floor, policy and publisher rules to select an eligible outcome.
The winning creative is served or rendered, followed by impression, viewability, click and conversion measurement where allowed. Each stage can time out, disagree or lose data, so auction success and ad delivery should be reconciled.
Opportunity
Validate publisher, placement, format, context, seller path and permitted request signals.
- What inventory is actually offered?
- Is the seller authorized and the context suitable?
Eligibility
Apply campaign, geography, audience, context, frequency, creative and budget constraints.
- Is this opportunity allowed for the campaign?
- Can the approved creative render?
Value and bid
Estimate incremental economic value and translate it into a bid under uncertainty and auction rules.
- What is the expected valuable outcome?
- How much of that value can this impression justify?
Auction and serve
Return a valid response in time and verify the eligible winner renders and is measurable.
- Did the response meet protocol and deadline?
- Was the impression viewable and valid?
Learn and govern
Use win, quality and outcome logs to improve bidding without relaxing non-negotiable controls.
- Why did bids win or lose?
- Did won impressions create incremental value?
OpenRTB standardizes the conversation
IAB Tech Lab's OpenRTB specification defines common objects and fields that let selling and buying platforms exchange bid requests and responses. It covers inventory, media formats, deal information, user or device signals, regulations, supply chain and creative metadata.
A standard enables interoperability but does not require every party to send every field or guarantee accuracy. Extensions, media-specific conventions and platform differences remain. Buyers must validate field coverage and not treat a populated value as verified truth.
Data minimization matters. A technically supported field should enter the bidstream only when necessary, permitted and appropriately signaled. Sensitive or precise combinations can create risk even if each isolated field seems ordinary.
Translate expected outcome value into a bid
A simple conceptual bid begins with the probability of a valuable outcome multiplied by its economic value, then accounts for incrementality, uncertainty, platform costs and required return. Production bidders also manage budgets, goals, competition and learning.
The outcome label determines behaviour. Click optimization rewards click probability, conversion optimization rewards the tracked event and value bidding rewards the values supplied. If low-quality leads or gross revenue are mislabeled as value, the bidder can scale them efficiently.
Attribution is not incrementality. A user predicted to purchase may have high attributed value but low causal response to another impression. Calibrate bidding and targets with lift evidence when feasible and avoid claiming a precise causal value per bid request without support.
Floors and auction rules affect price
Publishers or intermediaries may set price floors, deal priorities and eligibility rules. Auction mechanics can vary, and the submitted bid does not always equal the amount ultimately paid. Fees and currency handling also affect advertiser cost and publisher revenue.
In a first-price setting, a buyer may pay its submitted price if it wins, which creates an incentive for bid shading: estimating a lower bid that can still clear while protecting value. The correct strategy depends on information and auction implementation.
Do not infer inventory value solely from clearing price. Competition, floor changes, duplicated auctions and supply routing can move cost. Inspect bid landscape and path-level outcomes rather than celebrating a CPM decline without context.
Latency and reliability shape the reachable auction
Bid requests operate under deadlines because the page or stream must render. Slow models, network hops or creative checks can cause timeouts, reducing access even when a campaign would bid. Server-side routing can shift but not eliminate latency tradeoffs.
Track bid rate, response time, timeout, invalid response and render failure by partner and format. A low win rate can reflect price, eligibility or response reliability; one aggregate metric cannot distinguish them.
Optimize infrastructure without weakening privacy or quality checks. A bidder that responds faster by skipping seller validation or consent logic creates risk, not efficiency.
Real-time bidding example
The cycling-safety brand evaluates the complete opportunity before price. Context, seller, placement and expected contribution shape eligibility and value, while non-negotiable quality rules can produce a no-bid.
Win rate is interpreted with supply quality and incremental customer economics. This prevents the bidder from learning that unknown cheap inventory is desirable simply because it clears easily.
A hypothetical cycling-safety brand sells durable lights and reflective gear. It wants prospecting reach around practical commuting content without paying for opaque low-quality placements.
Bid requests must resolve to approved web or app inventory, suitable content, supported formats and transparent seller paths. Current customers and over-frequency users are excluded where the permitted signals support it.
The DSP estimates a verified first-order probability and expected contribution after returns. Context, placement quality and creative match influence the estimate; a cheap predicted click is not the outcome.
The bidder applies budget pacing, auction rules, fees and an uncertainty haircut. It declines opportunities below quality requirements even when the floor is low.
Win rate is reviewed with bid landscape, clearing cost, supply path, viewability, invalid traffic and reach. A sudden win-rate increase on unknown apps is treated as a quality alert.
Verified customer contribution informs learning, while geo or audience holdouts estimate lift. Bid values are adjusted from causal evidence rather than assuming every attributed purchase was created by the impression.
This example is hypothetical. Actual auction mechanics, available bidstream fields and privacy obligations vary by seller, DSP, media type and jurisdiction.
Read bid, win and outcome metrics together
Bid rate shows how often available requests pass eligibility and receive a response. Win rate shows wins among bids or opportunities under the chosen definition. Both depend on targeting, floors, price, timeouts, deals and supply mix.
CPM describes cost per thousand impressions but not unique reach, viewability, attention or business value. Add reach, frequency distribution, placement, invalid traffic, supply path and creative render diagnostics.
Reconcile auction and billing logs with impression and conversion systems. Measure verified downstream outcomes and contribution. Use controlled holdouts or geo tests to estimate lift, because auction logs observe won opportunities rather than the no-ad counterfactual.
The bidstream requires careful data governance
A bid request can be distributed to multiple potential buyers and intermediaries, making minimization and accountability critical. Map which fields leave the publisher, which parties receive them, retention, purpose and whether a bid is placed.
Consent strings and regulatory flags help systems communicate state but do not prove that every processing purpose is lawful or expected. Requirements differ by jurisdiction. Current UK guidance covers cookies, pixels, scripts and related storage or access technologies used around online advertising.
Contextual and coarse signals can support bidding with less identity dependence. Avoid sensitive segments, precise location or fingerprinting unless a lawful, necessary and ethically defensible use is established. A no-bid is an appropriate privacy outcome.
Limitations and common mistakes
RTB is distributed and time constrained. Bid requests can be incomplete or inaccurate, supply paths duplicated, auctions opaque and measurement partial. Optimization can exploit flaws in outcome data faster than manual buying would.
Common mistakes include equating RTB with all programmatic, maximizing win rate, bidding to clicks without quality, trusting domain fields without seller validation, ignoring timeouts and fees, and treating attributed conversions as incremental lift.
The impression is not the customer. Even a well-priced opportunity can fail because of weak creative, product, landing experience or distribution. RTB improves transaction decisions within the system it receives; it does not create a marketing strategy.
A rational RTB bid prices an eligible, trustworthy opportunity against expected business value. It does not reward winning for its own sake.
Frequently asked questions
What happens in an RTB auction?
A seller sends a bid request, eligible DSPs respond with a price and creative, the seller applies auction and policy rules, and the winning eligible ad is served.
Is RTB the same as programmatic advertising?
No. RTB is one programmatic buying mechanism. Automated guaranteed or preferred transactions may not auction each impression in real time.
What is a bid request?
It is a structured description of an advertising opportunity, potentially including inventory, format, context, deal, regulatory and permitted user or device fields.
Is a higher win rate better?
Not by itself. It can reflect competitive bidding or weak cheap supply. Read it with reach, quality, clearing cost and incremental outcomes.
How should an advertiser set bid value?
Base it on a validated valuable outcome and contribution, then account for probability, uncertainty, incrementality, costs, constraints and auction mechanics.
Sources and further reading
- IAB Tech Lab: OpenRTB 2.6 Specification ↗Primary technical specification for real-time bid requests, responses, objects and transaction flow
- IAB UK: Back to Basics Guide to Programmatic ↗Industry explanation of auctions, DSPs, SSPs, exchanges and ad serving
- IAB Tech Lab: sellers.json and SupplyChain ↗Official standards for identifying sellers and transaction paths in programmatic supply
- ICO: Guidance on Storage and Access Technologies ↗Current UK guidance relevant to cookies, pixels, scripts and related advertising technologies