Demand-Side Platforms · How to Buy, Configure & Optimise

A Demand-Side Platform is software that enables advertisers to buy digital advertising inventory programmatically — connecting to multiple ad exchanges, SSPs, and publisher networks through a single interface. The DSP handles every part of the programmatic buying workflow: configuring targeting and bidding parameters; evaluating incoming bid requests in real time; submitting bids to auctions; delivering creatives when bids are won; and aggregating impression, click, and conversion data for reporting.

The distinction between a DSP and an ad network is important. An ad network is a reseller — it buys inventory from publishers, bundles it, and sells it to advertisers at a markup without revealing the underlying inventory sources. A DSP is a transparent buying tool — the advertiser is bidding directly in auctions against other buyers, with visibility into which sites their ads appeared on and what CPMs they paid. This transparency is the foundation of supply path optimisation and brand safety management.

The distinction between a DSP and social advertising platforms (Meta Ads Manager, LinkedIn Campaign Manager) is also important. Social platforms are walled gardens — the audience data, inventory, and measurement all sit within one company's ecosystem. A DSP provides access to the open web, connected TV, and audio across thousands of publishers, using a mix of third-party data, contextual signals, and first-party data that the advertiser brings in. Each has different strengths: social provides more precise first-party audience data within its platform; DSPs provide broader reach across the open web with more control over supply path and data sourcing.

Most sophisticated advertisers use multiple DSPs rather than one — running DV360 for YouTube integration, The Trade Desk for open web and CTV, and Amazon DSP for commerce-adjacent targeting. The choice of DSP affects not just campaign performance but supply chain economics: each DSP has different fee structures, different supply partnerships, and different data relationships that affect the effective CPM the publisher receives and the effective CPM the advertiser pays.

Every DSP has three functional layers: the bidder, the UI, and the data layer. Understanding each helps explain what DSPs can and cannot do.

The bidder is the technical core of the DSP — the system that receives bid requests from ad exchanges, evaluates them against active campaign criteria, determines bid prices, and submits bids. The bidder must operate at extremely high speed: it receives millions of bid requests per second and must respond to each within approximately 100 milliseconds. Bidder latency — the speed at which the DSP can process and respond to bid requests — directly affects win rates, particularly for high-competition inventory. DSPs with faster, more globally distributed bidding infrastructure win more auctions for the same budget.

The UI (User Interface) is the campaign management layer — where advertisers define targeting, upload creatives, set bids and budgets, and review reporting. UI quality varies significantly between DSPs. The Trade Desk's Kokai interface and DV360's integration with the Google Marketing Platform are generally considered the most feature-rich; smaller DSPs often trade UI sophistication for cost or specialisation.

The data layer is the DSP's audience data infrastructure — its connections to data marketplaces (where third-party audience segments can be purchased), its identity graph (how it matches users across devices and browsers), and its lookalike modelling capabilities. The data layer is where DSPs are most differentiated: Amazon's purchase data, The Trade Desk's Unified ID 2.0 partnerships, and Google's signed-in user graph each provide different targeting capabilities.

DSP campaign structures have a three-level hierarchy, broadly equivalent to how Google Ads and Meta Ads structure their campaigns:

• Campaign level: Sets the overall budget, campaign dates, advertiser, and high-level objective. Some DSPs handle frequency capping at the campaign level to limit how often the same user sees ads across all line items.
• Line item / Insertion Order level: Defines specific targeting criteria, inventory sources, bidding strategy, and budget allocation. A single campaign typically has multiple line items — for example, one line item targeting a prospecting audience, one targeting website visitors (retargeting), and one targeting a custom first-party CRM list.
• Creative / Ad level: The specific ad creatives assigned to each line item. Multiple creatives per line item enables rotation, A/B testing, and dynamic creative optimisation.

Best practice campaign structure separates audiences into distinct line items rather than combining them — because different audiences require different bid levels, different frequency caps, and different creative approaches. Mixing a retargeting audience (high intent, high bid justified) with a broad prospecting audience (low intent, lower bid appropriate) in the same line item means you are likely underbidding for retargeting impressions and overpaying for prospecting impressions simultaneously.

DSP targeting layers can be used independently or layered together. Understanding each layer's data source, accuracy, and appropriate use case prevents targeting combinations that reduce scale without improving relevance.

Audience targeting targets users based on who they are rather than where they are. Sources include: third-party segments from data marketplaces (demographic, interest, in-market — based on third-party cookie data, mobile ad ID data, or modelled signals); first-party data uploads (CRM email lists matched to programmatic identifiers); and retargeting (previous visitors to the advertiser's website or app, matched through pixel data).

Contextual targeting targets pages based on their content rather than their audience. A contextual segment targeting "personal finance" would reach users reading financial news, regardless of who they are. Contextual has regained prominence as third-party cookies phase out — because it requires no user tracking, it is inherently privacy-safe and works in any browser environment. Contextual accuracy depends on the quality of the page classification model; leading contextual vendors (Grapeshot, now Oracle; Peer39; IAS) use natural language processing to classify pages at scale.

Environmental targeting filters by the context of the impression itself: geography (country, region, city, postcode), device type (desktop, mobile, tablet, CTV), operating system, browser, day and time, and content category. Environmental targeting is coarse but reliable — it does not depend on user data or page classification accuracy.

The most effective targeting strategies for each objective:

DSP bidding strategies determine how the DSP prices each bid request and optimises spend over time. The right bidding strategy depends on the campaign objective and the maturity of the campaign's conversion data.

Fixed CPM: The advertiser sets a specific maximum CPM. The DSP bids at or below this level for every impression that matches targeting criteria. Appropriate for campaigns with specific CPM targets or when buying premium inventory at known market rates.

Viewable CPM (vCPM): Bids are optimised to win impressions that are likely to be viewable (meeting the IAB standard of 50% of pixels visible for at least 1 second for display; 50% visible for 2 seconds for video). The DSP uses historical viewability data by site and placement to predict which impressions are likely to be viewable and bids more aggressively for them. Appropriate for brand awareness campaigns where seen impressions are the currency.

CPC bidding: The DSP optimises to minimise cost per click, bidding more aggressively for impressions it predicts will generate clicks based on historical data. Appropriate for traffic-driving campaigns. Note that CTR in programmatic display is extremely low (0.05–0.10% is typical) — CPC programmatic is generally far less efficient than CPC paid search for driving high-intent traffic.

CPA (Target Cost Per Acquisition): The DSP optimises toward a target cost per conversion. It uses conversion signal data from the ad server or conversion API to train models that predict which impressions are most likely to lead to conversions. Requires significant conversion volume (typically 50+ conversions per line item per week) to train the model effectively. Most powerful when conversion data is reliable and volume is sufficient.

ROAS bidding: The DSP optimises toward a target return on ad spend, using conversion value data to allocate more budget to impressions more likely to drive high-value conversions. Appropriate for e-commerce campaigns with variable order values where a £20 CPA for a £200 order is worth more than a £20 CPA for a £30 order.

Programmatic campaigns require creatives to be trafficked into the DSP's ad server or linked from an external ad server (typically Google Campaign Manager 360). Creative quality directly affects Quality Score equivalents in some DSP systems and viewability rates across placements.

Standard IAB display formats (as defined in the IAB New Ad Portfolio) include: medium rectangle (300×250), leaderboard (728×90), wide skyscraper (160×600), half-page (300×600), large rectangle (336×280), and mobile banner (320×50). For programmatic campaigns targeting multiple screen sizes, Responsive Display Ads — where the DSP assembles creatives from headline, image, logo, and description assets — reduce the trafficking burden and enable dynamic assembly for different placements.

Creative rotation: multiple creatives per line item enables both A/B testing and frequency management. Users who have seen a creative multiple times will have diminishing response; rotating to a different creative at defined frequency thresholds maintains engagement without requiring a full pause-and-restart workflow.

DSP reporting provides delivery metrics (impressions won, impressions served, win rate), engagement metrics (clicks, CTR, video completion rate), and — when conversion tracking is integrated — outcome metrics (conversions, CPA, ROAS). The most important reporting dimensions for optimisation are: performance by audience segment (which targeting segment drives the most conversions?), performance by placement domain (which sites produce the best results?), performance by creative (which ad variants drive the most engagement?), and performance by time of day and day of week.

Post-view attribution is a specific measurement challenge in programmatic: a user may see a display ad and convert later without clicking on it. Post-view attribution credits the display ad for conversions within a defined window (typically 1–7 days) after an impression was served. This can overstate programmatic's contribution to conversions — any user who saw the ad and later converted naturally is credited as a programmatic conversion, even if the display ad played no role. Holdout testing (withholding ads from a matched control group) is the most rigorous way to measure incremental lift from programmatic, as documented in academic studies of advertising effectiveness.

Systematic programmatic optimisation follows a hierarchy from macro decisions (campaign structure and budget allocation) to micro decisions (individual placement and audience adjustments):

Week 1–2: Establish baseline. Run all targeting segments and placements without heavy filtering. Collect impression, viewability, click, and conversion data across the full targeting matrix. Resist the urge to optimise based on insufficient data — statistical significance requires sufficient volume.

Week 3–4: Remove obvious waste. Pause or exclude placements with zero conversions and high spend (minimum £50–100 spend per placement before excluding). Exclude times of day with systematically poor performance. Review placement brand safety reports and exclude inappropriate content categories.

Month 2+: Invest in what works. Increase bids or budgets for audience segments, placements, and creatives that are performing above target KPIs. Build site inclusion lists from the highest-performing placement domains. Develop lookalike audiences from converters to expand reach with maintained performance.

Supply Path Optimisation (SPO) is the practice of preferring shorter, cheaper, more transparent paths to publisher inventory over longer, more expensive, less transparent ones. In open programmatic, the same publisher impression may be available through multiple SSPs and exchanges — each adding a fee at each step. SPO identifies which supply paths are most direct and cost-efficient and concentrates buying through them.

Ads.txt (Authorised Digital Sellers) is the IAB-standard declaration that publishers use to specify which SSPs and exchanges are authorised to sell their inventory. DSPs that check ads.txt files before bidding significantly reduce the risk of buying counterfeit inventory — impressions that claim to be from premium publishers but are actually from fraud operations. The IAB's documented ads.txt specification is the foundation of supply chain verification.

The practical benefit of SPO: buying a publisher impression through a direct SSP relationship rather than through three intermediary exchanges reduces fees from 35–45% of advertiser spend to 10–15%, improving the effective CPM that reaches the publisher and often improving the clearing price for the advertiser simultaneously.

Sources & Further Reading