Programmatic Advertising and Ad-Tech

Introduction and Context

Programmatic advertising is the automated buying and selling of digital advertising inventory. It uses software, data, standardised communication protocols and, frequently, auctions to match an advertiser’s demand with a publisher’s available impression. Programmatic does not mean that advertising strategy is fully automatic. Human decisions remain essential in defining objectives, audiences, budgets, creative policy, brand safety, measurement and governance.

For media planners and buyers, programmatic advertising matters because digital inventory is fragmented across websites, mobile applications, connected television, digital audio, games and other environments. Manual negotiations cannot efficiently evaluate every available impression. Ad-tech systems provide the infrastructure for selecting, pricing, delivering and measuring advertising at large scale.

The central challenge is to obtain useful reach and outcomes while controlling cost, waste, privacy risk, fraud, inappropriate adjacency and excessive frequency. A media professional therefore needs both operational knowledge of the technology and critical understanding of its commercial and social consequences.

Learning Objectives

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

1. Explain the functions of DSPs, SSPs, ad exchanges, ad servers, DMPs and CDPs.
2. Describe real-time bidding, private marketplaces and programmatic guaranteed buying.
3. Distinguish audience targeting, identity resolution and frequency management.
4. Analyse supply-path optimisation and the economics of the ad-tech supply chain.
5. Explain how privacy regulation and the reduction of third-party cookies are changing programmatic practice.
6. Evaluate a programmatic campaign using appropriate operational, media and business metrics.
7. Recommend responsible practices for privacy, transparency, brand safety and measurement.

1. The Programmatic Advertising Ecosystem

A programmatic transaction involves several participants:

• Advertiser or agency: Defines the campaign objective, audience, budget, creative and buying rules.
• Demand-side platform (DSP): Software used by buyers to evaluate impressions and submit bids across multiple sources of inventory.
• Publisher or media owner: Makes advertising space available on a website, application, streaming service or other medium.
• Supply-side platform (SSP): Software used by publishers to manage, package and sell inventory to potential buyers.
• Ad exchange: A technology marketplace that facilitates transactions between buyers and sellers, often through auctions.
• Ad server: Technology that stores, selects, delivers and records advertisements. Publisher ad servers manage inventory and priority; advertiser ad servers often manage creative rotation, tracking and campaign reporting.
• Data provider: Supplies audience, contextual, geographic or other signals, subject to applicable permissions and contracts.
• Verification provider: Measures viewability, invalid traffic, brand suitability and related quality indicators.
• Identity and consent services: Help recognise permitted users or households and record privacy choices.

These components are not always separate companies. A large platform may provide several functions, while an agency may use multiple vendors. The apparent simplicity of an advertisement appearing on a screen conceals a sequence of requests, decisions and validations.

The basic transaction flow

1. A user opens a page, application or streaming programme.
2. The publisher’s ad server determines that an impression is available.
3. The SSP or exchange sends an auction request containing permitted information, such as placement, device type, geography, content category and an identifier or contextual signal.
4. Eligible DSPs compare the impression with campaign rules and estimate its value.
5. DSPs submit bids and creative references within a very short time period.
6. The exchange applies auction and deal rules, selects the winning eligible bid and returns the creative or ad-server instructions.
7. The advertisement is rendered.
8. Measurement systems record delivery, viewability, clicks, conversions and quality signals, subject to technical and privacy constraints.

The process is often called real-time bidding (RTB), although not every programmatic transaction uses an open auction.

2. Core Platforms and Their Distinct Roles

Demand-side platforms

A DSP is a buyer-side decision system. It receives bid requests, filters impressions according to campaign rules, predicts the probability or value of an outcome and submits a bid when the impression is suitable. Typical controls include geography, device, operating system, time, contextual category, audience membership, frequency, budget pacing, inventory source and brand-safety requirements.

A DSP is not itself an audience. It is not necessarily an ad exchange, and it does not automatically guarantee that every impression is high quality. Its performance depends on data, algorithms, integrations, auction access, measurement and the quality of campaign configuration.

Supply-side platforms

An SSP helps a publisher sell inventory. It may connect the publisher to exchanges, DSPs and direct deals; enforce floor prices; package audiences; manage formats; and provide yield reporting. The publisher may use an ad server to establish priority among direct campaigns, sponsorships, private deals and open-auction demand.

The publisher’s objective is not simply to obtain the highest nominal bid. It also considers revenue quality, latency, user experience, contractual obligations, demand reliability and the risk of unsuitable advertisements.

Ad exchanges

An ad exchange coordinates marketplace activity. It standardises bid requests and responses, applies auction logic and returns a winning result. An exchange may support open auction inventory, private auctions and other deal types. In practice, boundaries between exchanges and SSPs can be commercially blurred, so planners should inspect the actual service and fee structure rather than rely only on labels.

Ad servers

An ad server is a delivery and decision layer. It can select an advertisement based on campaign priority, targeting, rotation, frequency rules and delivery goals. It also issues impression and click-tracking events. Ad servers are especially important for deduplicated reporting and creative governance.

An ad server should not be confused with a DSP. The DSP decides which external impressions to buy; the ad server primarily decides which eligible creative or campaign should be delivered in a particular ad slot and records delivery events. Some platforms combine these functions.

Data management platforms

A data management platform (DMP) traditionally stores and organises large volumes of pseudonymous audience data, often using cookie or mobile-advertising identifiers. DMPs have been used to create segments such as “recent travel researchers” or “automotive intenders” for activation in advertising systems.

DMPs are less suitable for storing rich, persistent customer records. Their value has declined where third-party cookies and other cross-site identifiers are restricted. They remain useful in some contexts for short-lived segmentation, contextual signals and campaign data, but their legality and effectiveness depend on consent, provenance and platform access.

Customer data platforms

A customer data platform (CDP) generally unifies first-party data from a known organisation’s interactions, such as authenticated web activity, transactions, subscriptions, customer service and loyalty records. It creates customer profiles for permitted marketing, service and analytics uses.

A CDP is not automatically privacy-compliant or automatically interoperable with every DSP. Identity matching, consent status, data minimisation, retention and access controls must be designed deliberately. The distinction is useful: a DMP historically emphasises pseudonymous audience activation, whereas a CDP emphasises an organisation’s first-party customer relationships.

3. Buying Methods and Auction Logic

Open auction and real-time bidding

In an open auction, eligible buyers compete for an impression available to a broad set of market participants. The bid request may include an auction identifier, placement details, content information, technical characteristics, permitted location signals, deal identifiers and privacy signals.

A simplified buyer valuation can be represented as:

Here, V_i is the buyer’s estimated value of impression i; P(O_i | x) is the predicted probability of the desired outcome given available signals x; W_O is the monetary value assigned to that outcome; C_i represents the expected media cost; and R_i represents expected risk or additional cost, such as fraud, unsuitable placement or measurement uncertainty. This is a decision model, not a universal auction formula. It is applicable only when the terms are expressed in comparable monetary units and the probability and value estimates are reasonably calibrated.

A DSP may bid only when V_i exceeds a required threshold or when the impression satisfies strategic constraints. The actual bid can be lower or higher than estimated value depending on budget, competition, pacing and optimisation policy.

Historically, many exchanges used second-price-style auctions, where the winner paid approximately the second-highest eligible bid plus a minimum increment. Modern auctions frequently use first-price mechanics, in which the winner pays its submitted bid, although floors, fees, deal rules and auction adjustments complicate the simple description. Buyers must therefore know the auction mechanics and distinguish the bid submitted from the amount ultimately paid.

For a first-price auction with a clearing price equal to the winning bid, a campaign’s media cost for N impressions is:

where B is the cost per thousand impressions (CPM), expressed in currency units per 1,000 impressions. The formula assumes that B is the applicable media CPM and excludes separately charged technology, verification, agency and data fees. For example, 250,000 impressions at a media CPM of £4.00 cost:

250,000 × £4.00 ÷ 1,000 = £1,000.

Private marketplaces

A private marketplace (PMP) is a controlled programmatic auction in which selected buyers receive access to specified inventory. A publisher may offer a preferred deal, a private auction or a curated package of sites and formats. PMPs can provide greater transparency, negotiated conditions and stronger publisher relationships than an open auction, but they are not automatically superior. A PMP may still contain invalid traffic, limited scale, unclear fees or weak audience definitions.

Programmatic guaranteed and preferred deals

Programmatic guaranteed resembles a direct insertion order: buyer and seller agree on inventory, price, schedule and often audience or placement conditions. The transaction is executed through technology rather than manual trafficking. A preferred deal gives a buyer first opportunity at an agreed price but may not guarantee delivery. These approaches can be useful for premium video, connected television, sponsorships and inventory requiring predictable placement.

4. Targeting, Identity and Frequency

Targeting approaches

Contextual targeting selects inventory based on the content or environment, such as an article about home renovation. It does not require identifying an individual and has regained importance in privacy-sensitive environments.

First-party targeting uses information collected directly by an organisation, such as consented customer status, subscription, purchase history or logged-in engagement.

Second-party data is another organisation’s first-party data shared through a defined partnership.

Third-party data is collected or aggregated by an external provider and may involve multiple sources. Its quality, permission basis, recency and representativeness must be examined carefully.

Geographic and device targeting use permitted location, region, language, device or connectivity signals. Precision claims should be treated cautiously because signals may be approximate or deliberately withheld.

Behavioural targeting infers interests or likely actions from observed activity. It can improve relevance but creates privacy, discrimination and inference risks.

Lookalike or modeled targeting finds users or households statistically similar to a seed audience. Similarity is a prediction, not proof of identity or intent.

Identity resolution

Identity resolution links records believed to refer to the same person, household, device or organisation. Deterministic resolution uses a known common key, such as a permitted login identifier. Probabilistic resolution infers a relationship from signals such as device, timing, location and browsing patterns. Probabilistic matching is uncertain and should not be treated as fact.

A responsible identity system should specify the entity being resolved, the confidence level, the permission basis, retention period, permitted purposes and deletion mechanism. Identity graphs can improve reach and measurement, but they can also amplify errors, create hidden surveillance and produce inaccurate frequency counts.

Frequency management

Frequency is the number of times an individual, household or device is exposed to a campaign. A simple average frequency is:

If a campaign delivers 1,200,000 impressions to 300,000 counted users, average frequency is four exposures per counted user. This calculation assumes that reach and impressions use the same population definition, time period, deduplication method and measurement source. It does not show the distribution: some users may see one advertisement while others see many more.

Frequency caps attempt to limit exposure, such as no more than three impressions per user in seven days. Caps are imperfect when identifiers are unavailable, browsers restrict storage, several buying platforms operate independently or identity resolution is incomplete. Cross-device and household-level caps are especially difficult. Planners should distinguish platform-reported frequency from independent, deduplicated frequency.

5. Supply-Path Optimisation and Measurement

Supply-path optimisation (SPO) is the practice of evaluating and selecting efficient, transparent and reliable routes from buyer to publisher. The same impression may be reachable through multiple SSPs, exchanges, resellers and intermediaries. Unnecessary duplication can increase fees, auction competition and latency without increasing publisher value.

An SPO review examines:

• publisher authorisation files and seller declarations;
• directness of the route and number of intermediaries;
• working media cost versus total cost;
• win rate, bid rate and timeout rate;
• viewability, invalid traffic and brand-suitability results;
• unique reach and incremental supply;
• inventory quality and format availability;
• auction dynamics and fee transparency;
• sustainability and operational reliability.

A route with the lowest technical fee is not necessarily best. A more direct route may have less scale, while a curated intermediary may add verification or valuable access. SPO is therefore a constrained optimisation problem involving cost, quality, reach, performance and risk rather than a simple search for the cheapest path.

Metrics

Common operational metrics include:

• Impressions: Recorded ad deliveries or opportunities, according to the measurement method.
• Reach: Number of distinct measured users, devices or households exposed.
• CPM: Media cost per 1,000 impressions: CPM = Cost ÷ Impressions × 1,000.
• Click-through rate: Clicks divided by measured impressions, usually expressed as a percentage.
• Conversion rate: Conversions divided by a stated denominator, such as clicks or sessions.
• Viewability: The proportion of measurable impressions meeting a defined visibility standard.
• Completion rate: Completed video plays divided by initiated or measurable video plays, with the denominator clearly specified.
• Return on advertising spend: Attributed revenue divided by advertising spend. It is not the same as profit and depends on attribution rules.

For example, £6,000 spent on 2,000,000 impressions gives:

CPM = £6,000 ÷ 2,000,000 × 1,000 = £3.00.

If 40,000 clicks were recorded, the click-through rate is 40,000 ÷ 2,000,000 × 100 = 2%. These metrics describe delivery, not necessarily causation. A click or conversion may be associated with an advertisement without being caused by it.

6. The Cookieless Ecosystem

A third-party cookie is a browser-stored value set by a domain different from the site the user is visiting. It historically supported cross-site recognition, retargeting and measurement. Browser restrictions, mobile platform policies, regulation and public concern have reduced its availability and reliability. The industry is therefore moving toward a mixed ecosystem rather than a single replacement.

Important components include:

• First-party identifiers: Accounts, consented email-derived tokens, subscriptions and customer records used within permitted relationships.
• Publisher-provided signals: Authenticated or consented information supplied by a publisher to an authorised buyer.
• Contextual signals: Content, semantic category, programme, time, geography and device information without individual tracking.
• Cohort or browser-mediated approaches: Systems that limit direct sharing of user-level information and provide aggregated or controlled signals.
• Clean rooms: Controlled environments in which parties compare or analyse data without freely exchanging raw identifiers. A clean room is a governance and technology arrangement, not a guarantee of anonymity.
• Conversion APIs and server-side measurement: Server-to-server transmission of permitted events, usually with stronger control than browser-only tags but with continuing privacy responsibilities.
• On-device and aggregated measurement: Processing or reporting designed to reduce exposure of individual-level data.

There is no universally equivalent substitute for third-party cookies. A solution may improve privacy while reducing reach, addressability, frequency accuracy or measurement granularity. Media plans should consequently use several methods and disclose their limitations.

7. Privacy, Ethics and Governance

Programmatic advertising involves personal data, inferred interests and automated decisions. Applicable law differs by jurisdiction, but core governance principles commonly include lawful and transparent processing, purpose limitation, data minimisation, accuracy, security, retention control and rights of access or deletion. Consent signals must be meaningful and respected throughout the supply chain; a consent management platform does not make an unlawful purpose lawful by itself.

Important risks include:

• collecting data beyond what is necessary;
• using sensitive or inferred attributes for exclusion or exploitation;
• discriminatory delivery caused by optimisation models;
• opaque data brokerage and unclear vendor responsibility;
• re-identification through supposedly anonymous datasets;
• surveillance of children or vulnerable audiences;
• ad fraud, malvertising and domain spoofing;
• ads appearing beside harmful or misleading content;
• excessive frequency and manipulative personalisation.

Governance should include a vendor register, data-flow mapping, approved purposes, access controls, retention schedules, audit rights, incident procedures, model testing and documented exclusion rules. Brand safety should be defined positively and negatively: not only what content is excluded, but also what quality, editorial and social environments the brand seeks.

8. Applied Case: A Fictional Sustainable-Transport Campaign

Background

MetroMove, a fictional regional public-transport organisation, wants to increase applications for a discounted annual travel pass among adults in three cities. It has a £120,000 media budget and wishes to reach potential applicants without relying on third-party cookies.

Problem

An initial plan using multiple open-auction vendors produces duplicated reach, inconsistent frequency reports and a high proportion of impressions on low-viewability inventory. The organisation also lacks a documented process for handling customer data.

Analysis

The planning team defines the primary outcome as completed applications, with secondary goals of qualified reach and cost control. It audits the supply chain, removes unauthorised resellers, and compares DSP reports with independent verification. The organisation’s existing pass-holder database is not automatically used for advertising; it is reviewed for permission, purpose compatibility and secure activation.

Approach and solution

The team uses four coordinated methods:

1. Contextual targeting around commuting, local employment, education and sustainable travel content.
2. Publisher-direct and PMP inventory from trusted local news and transport publishers.
3. A consented first-party activation segment for users who have requested information but not completed an application.
4. Broad geographic delivery with creative variation by city, rather than highly granular inferred personal profiles.

The team applies a campaign-level frequency policy, while clearly stating that platform caps are not a complete cross-device guarantee. It uses one primary DSP for open-market buying, a limited number of direct publisher integrations, independent viewability and invalid-traffic measurement, and server-side conversion events governed by a data-protection review.

Outcome

In this illustrative case, the campaign delivers 1,500,000 measured impressions at a media cost of £45,000, giving a media CPM of £30. It records 18,000 completed applications attributed under the organisation’s declared attribution model. The team does not claim that all applications were caused by advertising. It conducts a geographic holdout test in comparable areas to estimate incremental effect, then reallocates budget toward publisher-direct and contextual inventory that provides stronger incremental results and clearer reporting.

Learning points

The case illustrates that programmatic success is not simply a matter of purchasing the cheapest impressions. Clear objectives, privacy-compliant first-party use, contextual alternatives, controlled supply paths, independent measurement and causal evaluation are mutually reinforcing. It also demonstrates why reported conversions should be interpreted alongside incrementality, reach quality and public accountability.

9. Professional Best Practices

A media planner should:

1. Begin with the communication objective, not the available technology.
2. Document the buyer, seller, data, identity, measurement and fee relationships.
3. Use contextual and publisher-direct options where they meet the objective.
4. Separate media cost from technology, data, verification and agency fees.
5. Establish frequency rules and state their measurement limitations.
6. Verify inventory quality, authorised sellers, viewability and invalid traffic.
7. Use independent measurement where platform self-reporting is insufficient.
8. Test incrementality rather than relying only on last-touch attribution.
9. Require transparent contracts, audit rights and deletion procedures.
10. Review algorithmic outcomes for unfair exclusion or harmful optimisation.
11. Maintain contingency plans for identity loss, consent changes and platform outages.
12. Explain uncertainty to clients instead of presenting modeled estimates as facts.

Chapter Summary

Programmatic advertising is a software-mediated system for buying and selling digital inventory. DSPs represent buyers, SSPs represent publishers, exchanges coordinate transactions, ad servers manage delivery and measurement, DMPs organise largely pseudonymous audience data, and CDPs unify an organisation’s first-party customer information. RTB is one form of programmatic buying; PMPs, preferred deals and programmatic guaranteed provide more controlled alternatives.

Targeting may be contextual, geographic, first-party, behavioural or modeled. Identity resolution can improve activation and deduplication but introduces uncertainty and privacy risk. Frequency is meaningful only when the identity unit, measurement source and time period are clear. SPO evaluates routes through the supply chain using cost, transparency, quality, reach and reliability rather than cost alone.

The cookieless ecosystem combines first-party data, publisher signals, contextual methods, clean rooms, server-side events and aggregated or on-device measurement. No replacement fully reproduces the former capabilities of third-party cookies. Effective contemporary planning therefore combines technical competence with privacy governance, independent measurement, brand safety and critical interpretation of algorithmic results.

Key Terms

DSP: A buyer-side platform that evaluates impressions and submits bids.

SSP: A publisher-side platform that manages and sells advertising inventory.

Ad exchange: A marketplace technology that facilitates buying and selling of impressions.

Ad server: A system that selects, delivers and records advertisements.

DMP: A platform traditionally used to organise pseudonymous audience data for activation.

CDP: A system that unifies an organisation’s first-party customer data.

Real-time bidding: Automated auction-based buying of an individual impression.

Private marketplace: A controlled programmatic marketplace open to selected buyers.

Programmatic guaranteed: Technology-enabled buying with agreed inventory, price and delivery conditions.

Contextual targeting: Selecting inventory according to content or environment rather than an identified individual.

Identity resolution: Linking records believed to represent the same person, household, device or organisation.

Frequency cap: A rule intended to limit exposures to a defined entity over a specified period.

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

Viewability: The extent to which an advertisement meets a defined visibility standard.

Invalid traffic: Impressions or interactions that do not represent legitimate advertising opportunities.

First-party data: Data collected directly by an organisation through its own relationship or service.

Third-party cookie: A browser value set by a domain different from the visited site.

Clean room: A controlled environment for permitted data analysis without unrestricted raw-data exchange.

Incrementality: The additional outcome caused by advertising beyond what would otherwise have occurred.

Attribution: A method for assigning credit for an outcome to advertising contacts or channels.

Review Questions

Short Answer Questions

1. What is the primary function of a DSP?
2. How does a CDP differ from a traditional DMP?
3. Define a private marketplace.
4. Why can frequency caps be inaccurate across several platforms?
5. What is supply-path optimisation intended to achieve?

Long Answer Questions

1. Explain the flow of an impression from a publisher’s ad slot through an RTB auction to final ad delivery.
2. Compare open auctions, private marketplaces and programmatic guaranteed buying.
3. Discuss the advantages and risks of behavioural, contextual and first-party targeting.
4. Explain how the decline of third-party cookies is changing planning and measurement.
5. Evaluate the privacy, ethical and governance responsibilities of a programmatic advertiser.

Analytical / Application Questions

1. A campaign reports 900,000 impressions, 225,000 counted users and £3,600 in media cost. Calculate average frequency and CPM, then identify two limitations of these measures.
2. Design an SPO evaluation framework for a campaign using three DSPs and five SSPs. Specify the data you would collect and the trade-offs you would examine.
3. Develop a cookieless media plan for a local public-service campaign. Justify your choices of contextual, first-party, publisher-direct and measurement methods.

Multiple Choice Questions with Answers

Question 1

Which platform primarily represents the advertiser in programmatic buying?

A. Demand-side platform

B. Supply-side platform

C. Publisher ad server

D. Consent management platform

Correct Answer: A. Demand-side platform

Question 2

What is the defining feature of real-time bidding?

A. Manual negotiation of every placement

B. Auction-based automated evaluation of an individual impression

C. Guaranteed delivery at a fixed annual sponsorship price

D. Distribution of advertisements only through email

Correct Answer: B. Auction-based automated evaluation of an individual impression

Question 3

Which approach selects advertising inventory according to the content or environment rather than an identified user?

A. Deterministic identity matching

B. Behavioural retargeting

C. Contextual targeting

D. Customer-list suppression

Correct Answer: C. Contextual targeting

Question 4

A campaign spends £2,500 on 500,000 impressions. What is its media CPM?

A. £0.50

B. £2.00

C. £5.00

D. £20.00

Correct Answer: C. £5.00

Question 5

Which statement best describes a CDP?

A. It is exclusively an open-auction exchange

B. It generally unifies an organisation’s first-party customer data

C. It only measures viewability for publishers

D. It replaces all advertising creative

Correct Answer: B. It generally unifies an organisation’s first-party customer data

Question 6

Why may a campaign-level frequency cap fail to control total cross-platform exposure?

A. CPM cannot be calculated across platforms

B. Different platforms may not share a common identity or deduplicated measurement system

C. Contextual targeting always increases frequency

D. Ad servers cannot record impressions

Correct Answer: B. Different platforms may not share a common identity or deduplicated measurement system

Question 7

What is the main purpose of supply-path optimisation?

A. To increase the number of intermediaries in every transaction

B. To select routes to inventory using considerations such as cost, quality and transparency

C. To eliminate all publisher-direct buying

D. To guarantee that every impression converts

Correct Answer: B. To select routes to inventory using considerations such as cost, quality and transparency

Question 8

Which is a characteristic of a private marketplace?

A. Access is controlled for selected buyers or negotiated deals

B. No publisher participates in the transaction

C. Every buyer pays an identical annual subscription

D. It cannot use auctions

Correct Answer: A. Access is controlled for selected buyers or negotiated deals

Question 9

Which measurement method is most directly designed to estimate outcomes caused by advertising rather than merely associated with it?

A. Last-touch reporting alone

B. An incrementality or holdout test

C. Counting bid requests

D. Measuring file size

Correct Answer: B. An incrementality or holdout test

Question 10

Which statement about the cookieless ecosystem is most accurate?

A. A single replacement reproduces every former third-party-cookie capability

B. Only probabilistic identity remains legal

C. It combines approaches such as contextual signals, first-party data and aggregated measurement

D. Programmatic advertising is no longer possible

Correct Answer: C. It combines approaches such as contextual signals, first-party data and aggregated measurement