Programmatic Targeting · Audience, Contextual & Cookieless

Programmatic targeting operates across three fundamental dimensions, which can be used independently or layered:

Audience targeting answers: who is this person? It uses data signals — browsing history, purchase history, demographic data, declared interests — to identify users who match defined characteristics, regardless of what page they are currently on.

Contextual targeting answers: what is this person reading right now? It uses the content of the page being viewed to infer relevance and intent, without using any user data.

Environmental targeting answers: when and where is this impression? It uses the technical context — device, geography, time of day, browser, screen size — to filter impressions without content or user data.

Third-party audience segments are pre-built audience definitions — "in-market for SUVs," "household income $100k+," "interested in travel" — built by data providers from browsing data, purchase data, and demographic data collected across websites. These segments are available for purchase in DSP data marketplaces, typically priced at $0.50–$3.00 CPM layered on top of the media cost.

The documented accuracy challenge with third-party segments is significant. A 2019 study by the Ponemon Institute (commissioned by Lotame) found that third-party demographic segments had accuracy rates of 44–62% — meaning that a significant percentage of impressions served to a "female 25–34" segment were actually reaching people outside that demographic. For interest and in-market segments, accuracy is generally better because they are based on behavioural signals rather than inferred demographics, but still imperfect.

Third-party segments are most useful for upper-funnel prospecting where the targeting provides directional relevance at scale — reducing wasted impressions on clearly irrelevant audiences — rather than precision targeting. For high-precision targeting, first-party data substantially outperforms third-party segments.

First-party data — data collected directly from customers and prospects through owned channels — is the highest-quality targeting input in programmatic. A CRM email list of lapsed customers, a segment of high-value buyers, or a list of users who completed an onboarding flow but did not convert are all actionable first-party audiences that can be activated in programmatic.

Activation process: first-party data (typically email addresses or phone numbers) is uploaded to a DSP or identity partner that matches the identifiers to programmatic device IDs and cookie IDs. The match rate — the percentage of uploaded records that can be matched to reachable programmatic users — varies by data quality and identity solution, typically 40–70% for email list uploads.

The privacy imperative: first-party data activation requires that the data was collected with appropriate consent and that its use for advertising matches the consent provided. GDPR (in the EU and UK), CCPA (in California), and platform terms of service all set requirements for how first-party data can be used in advertising. See the compliance guide for the legal framework.

Contextual targeting shows ads based on the content of the page the user is viewing, rather than on data about the user. An ad for running shoes appears on a page about marathon training; an ad for business software appears on a page about productivity; an ad for a financial product appears in a financial news article. No user data is required — the match is between the ad's category and the page's content category.

Contextual targeting has experienced a significant resurgence as third-party cookies phase out, because it is inherently privacy-safe — it requires no cross-site tracking and works in any browser or device environment. The leading contextual vendors use natural language processing to classify pages at scale: Grapeshot (acquired by Oracle), Peer39, and IAS Contextual all use NLP models to classify web pages across the IAB Content Taxonomy categories in real time.

Advanced contextual targeting goes beyond category matching to semantic understanding — identifying pages that are relevant to a specific topic or message, not just a broad category. A campaign for a new electric vehicle model can target pages specifically about EV technology, range anxiety, charging infrastructure, and the decision to switch from petrol — a much more precise contextual signal than the broad "Automotive" IAB category.

Environmental targeting uses the technical metadata of each impression — who it was served to in what device, location, and time context — rather than content or audience data. It is the most reliable targeting layer because the signals are verifiable and do not depend on third-party data accuracy or page classification models.

Environmental targeting dimensions used in programmatic:

• Geography: Country, region, city, and postcode targeting — essential for local advertisers and campaigns with geographic relevance. IP-based geolocation accuracy is typically high at country/region level and somewhat lower at city/postcode level.
• Device type: Desktop, tablet, smartphone, connected TV, gaming console. Different device types have different engagement patterns, CTRs, and conversion rates — mobile typically has lower conversion rates for desktop-optimised checkout flows but higher engagement rates for visual content.
• Operating system and browser: Targeting specific OS versions enables relevant creative experiences (iOS App Store vs Google Play download ads); browser targeting is relevant for campaigns that use features not supported universally.
• Time of day / day of week: Dayparting — restricting campaign delivery to specific hours — reduces waste for time-sensitive campaigns (a lunch promotion running only 11am–2pm; a B2B campaign running only weekday business hours).

Programmatic retargeting reaches users who have previously visited the advertiser's website, app, or other owned digital properties. It is the highest-intent programmatic audience because it targets people who have already demonstrated interest in the advertiser — making it typically the highest-converting programmatic line item in most campaigns.

Implementation: a retargeting pixel (a short JavaScript snippet) placed on the advertiser's website adds users who visit to a retargeting audience in the DSP. Different URL patterns create different audience segments — homepage visitors (broad, lower intent), product page visitors (higher intent), and cart abandoners (highest intent, should receive the most aggressive bids and most compelling creative).

Frequency capping is essential in retargeting. Without frequency limits, the same user can be shown the same ad dozens of times per day — a negative brand experience documented to reduce conversion rates among users who find it intrusive. Three to seven impressions per user per week is a commonly used starting point; the optimal frequency varies by category and message.

Lookalike modelling in programmatic takes a first-party seed audience (existing customers, high-value converters) and identifies users who share similar behavioural and demographic characteristics — expanding the reachable audience beyond the seed list to users who have not yet engaged with the brand but are statistically more likely to than the general population.

DSP lookalike models vary in quality and methodology. Amazon DSP's lookalike modelling is based on Amazon's first-party purchase data — users who exhibit similar purchase patterns to the seed audience. The Trade Desk's lookalike modelling uses the Unified ID 2.0 identity graph across partner publisher data. The quality of the lookalike model depends entirely on the quality and size of the seed audience — lookalike models built from 50 converters are substantially less accurate than those built from 5,000 converters.

Third-party cookies — the tracking mechanism that has powered programmatic audience targeting since the channel's inception — are being phased out. Apple's Safari blocked third-party cookies from 2017. Firefox followed with Enhanced Tracking Protection in 2019. Google Chrome's phaseout, after multiple delays, is proceeding through 2024–2025 via Privacy Sandbox.

Chrome's market share (~65% of global browser usage) makes its cookie deprecation the most significant structural change in programmatic's history. The targeting methods that rely on cross-site user tracking — third-party demographic segments, interest segments based on browsing history, cross-site retargeting — will not function as they currently do in a post-cookie Chrome environment.

The industry response has produced several alternative approaches: first-party data activation (using consent-based CRM and logged-in user data); contextual targeting (no user data required); universal IDs based on authenticated email addresses (Unified ID 2.0, LiveRamp RampID); and Google's Privacy Sandbox (Privacy-Preserving cohort-based targeting using Chrome browser data).

Google Privacy Sandbox is a set of APIs built into Chrome that are designed to enable privacy-preserving advertising measurement and targeting without third-party cookies. The key API for programmatic targeting is Protected Audience (formerly FLEDGE) — which enables retargeting without the advertiser seeing user-level data by keeping audience membership in the browser rather than on a server.

Topics API replaces interest-based targeting by having the browser assign the user to broad topic categories (based on browsing history) and sharing a limited set of recent topics with advertisers — without sharing cross-site browsing data directly. The Topics API provides coarser targeting than cookie-based interest segments but does not require user tracking across sites.

Attribution Reporting API replaces third-party cookie-based conversion attribution with aggregated, privacy-preserving measurement that prevents individual-level tracking. This changes how advertisers measure programmatic conversions — individual-level attribution is replaced with statistical modelling of aggregate performance.

One of the most counterintuitive findings in programmatic research is that adding more targeting layers does not always improve campaign performance — and can make it worse. Layering multiple targeting criteria creates smaller audiences, reducing scale and bid competition, which can actually increase effective CPMs while reducing reach. A campaign targeting "in-market for software AND company size 50–200 AND located in London AND job title contains 'Marketing'" may find so few matching impressions that it fails to deliver its budget — while a campaign targeting "contextual finance AND environmental London" reaches a larger, sufficiently relevant audience more efficiently.

The documented research on targeting efficiency (Les Binet and Peter Field's IPA data, Nielsen audience verification studies) suggests that for most brand campaigns, broad contextual targeting with strong creative is more cost-effective than narrow audience targeting with average creative. For direct response campaigns where precise audience matching directly affects conversion probability, tighter targeting is justified — but the accuracy assumptions of the targeting data used must be tested, not assumed.

Sources & Further Reading