Programmatic Buying: Advanced · SPO, Identity & Deal Architecture

Supply Path Optimisation (SPO) is the practice of identifying and preferring the most direct, cost-efficient routes from DSP to publisher inventory — reducing the number of intermediary hops (and their fees) between advertiser spend and publisher revenue. The ISBA Supply Chain Transparency Study's documented finding that only 51p of every advertiser pound reached publishers in 2020 made SPO a boardroom-level issue, not merely a trading desk optimisation.

The mechanics: the same publisher impression is available through multiple supply paths — the publisher's primary SSP, secondary SSPs, exchange resellers, and potentially direct DSP-SSP integrations. Each path has different fees, different win rates (due to latency), and different data quality (first-party publisher data may only be available through the direct SSP relationship). SPO involves: auditing which supply paths are being used; measuring the effective CPM and win rate on each path; and systematically concentrating buying through the paths with the best cost-quality ratio.

Ads.txt is the technical foundation of SPO — it allows buyers to verify which SSPs are authorised to sell a publisher's inventory. An advertiser buying through an unauthorised reseller (not listed in the publisher's ads.txt) risks buying counterfeit inventory. Systematic ads.txt checking is the minimum supply chain hygiene standard; sellers.json validation (verifying the exchange-side declaration) is the more complete verification approach.

The industry's 2019 shift from second-price to first-price auctions fundamentally changed programmatic bid strategy. In a second-price auction, bidding your true maximum value was the dominant strategy — you would pay only marginally above the second-highest bid. In a first-price auction, you pay exactly what you bid — bidding your true maximum systematically overpays.

Bid shading is the algorithmic practice of submitting bids below the buyer's true maximum value in first-price auctions, based on historical clearing price data, to reduce overpayment while maintaining competitive win rates. All major DSPs now incorporate bid shading by default — but the quality of bid shading algorithms varies significantly, and understanding the mechanics enables better evaluation of DSP bidding performance.

A well-calibrated bid shading algorithm submits bids just above the historical clearing price for that impression type — winning the auction without significant overpayment. A poorly calibrated algorithm either: bids too low (reducing win rates on high-value inventory) or bids too high (reducing the savings the first-price transition was supposed to produce). Monitoring clearing price vs. bid price ratios in your DSP reporting is the diagnostic for bid shading quality.

Floor price management from the seller side creates additional complexity: SSPs use dynamic floor prices that adjust based on historical bid data. A DSP that consistently bids a specific amount for a specific publisher will find floors creeping toward its bid level — a documented phenomenon where SSP floor algorithms "learn" buyer price points. This is one reason sophisticated buyers maintain some randomness in bid patterns rather than using perfectly consistent bidding.

The deprecation of third-party cookies in Chrome (the final phase of which is proceeding through 2025) eliminates the identity layer that has underpinned programmatic audience targeting since the channel's inception. Understanding which replacement approaches are production-ready, which are experimental, and which are inadequate for your specific use cases is the critical planning decision for programmatic buyers in 2026.

Unified ID 2.0 (UID2): An email-based identity solution operated by The Trade Desk and IAB Tech Lab. When a user authenticates on a publisher's site (login or email submission), the publisher can hash the email address into a UID2 token. DSPs with UID2 integration can match this token to their own first-party data for audience targeting. UID2 is in production at major publishers and DSPs. Its limitation: coverage is limited to authenticated users — typically 20–40% of publisher audiences — leaving the unauthenticated majority unreachable through UID2 alone.

Google Privacy Sandbox (Protected Audience API): Browser-based interest group membership stored in Chrome, used for retargeting without cross-site user tracking. The advertiser defines interest groups via JavaScript on their website; Chrome stores the membership locally; the Protected Audience API conducts an on-device auction when the user visits a publisher page. Privacy Sandbox is in limited production. Its current documented limitations include: no cross-device coverage; limited transparency into auction mechanics; and targeting precision substantially below cookie-based retargeting.

Contextual targeting: The most cookieless-resilient targeting approach — no user identity required. Advanced contextual solutions (Oracle Contextual, IAS Contextual, Peer39) use NLP to classify pages at scale. For brand awareness and upper-funnel campaigns, well-executed contextual targeting performs comparably to cookie-based demographic targeting in documented A/B tests — because contextual relevance is often a better proxy for commercial intent than demographic segments.

First-party data activation: CRM email lists matched to programmatic IDs through identity partners (LiveRamp, The Trade Desk UID2). Match rates of 40–70% are typical. First-party data is the highest-quality targeting input available in a cookieless environment — but only for advertisers with substantial CRM databases.

Private Marketplace (PMP) deals provide advertiser access to publisher inventory at publisher-set floor prices before it flows to the open auction. The strategic case for PMPs goes beyond brand safety: premium publisher first-party data (available only through direct publisher relationships), viewability guarantees, and inventory exclusivity in competitive categories justify PMP CPM premiums that may appear unfavourable in pure CPM comparison but are justified by quality-adjusted outcomes.

When PMPs outperform open auction: in categories where contextual adjacency significantly affects campaign performance (finance, health, luxury — where being seen on a trusted, authoritative publisher matters); when publisher first-party audience data is the most valuable targeting input (financial services publishers with verified subscriber demographics); and when brand safety is the primary concern and the advertiser needs pre-vetted, contractually guaranteed placement.

When PMPs underperform: when the floor price is significantly above the open auction clearing price without corresponding inventory quality improvement; when the PMP publisher does not have audience data assets that justify the premium; and when scale is the primary objective — PMPs are inherently limited in volume.

PMP negotiation levers: floor price (always negotiable at meaningful spend levels); viewability guarantee (ask for minimum 70% viewable CPM guarantee); first-party data inclusion (negotiate inclusion of publisher audience segments as targeting signals); and reporting transparency (insist on domain-level reporting, not just aggregate deal metrics). Publishers who refuse all negotiation on these points for sizable budgets are not partners worth prioritising.

Programmatic Guaranteed (PG) is the most direct form of programmatic — a guaranteed volume of impressions at a fixed CPM, with the delivery automation of programmatic but none of the auction uncertainty. PG combines the certainty of a direct insertion order with the targeting, reporting, and creative management capabilities of programmatic infrastructure.

PG is appropriate when: the campaign requires guaranteed delivery against a specific inventory (homepage takeovers, sponsorship integrations, contextually exclusive placements); the brand cannot accept auction variability for high-visibility campaigns; and the premium for guarantee (typically 20–40% above open auction CPM for equivalent inventory) is justified by the campaign objectives. PG should not be used as a default for all programmatic buying — the premium is only justified when the guarantee itself has value.

PG vs direct IO: PG uses the same inventory and often the same commercial terms as a direct insertion order, but all creative trafficking, targeting, and reporting flows through the DSP rather than through manual ad server trafficking. For buyers already using DV360 or The Trade Desk, PG deals offer the operational efficiency of DSP-based management while maintaining the inventory certainty of a direct deal.

DSP selection at scale involves evaluating: supply access (which publishers and SSPs does this DSP have direct integrations with?); data infrastructure (what first-party data activation, identity graph, and audience modelling capabilities does it offer?); bidding technology (bid shading quality, latency, win rates on high-value inventory); measurement integration (compatibility with independent measurement vendors, incrementality testing tools, and MMM data feeds); and fee structure (what percentage of media spend goes to the DSP versus reaching publishers?).

DSP fee negotiation: standard DSP fees are 15–20% of media spend for self-serve accounts. At £1M+ annual spend, fees of 10–12% are achievable. At £5M+ spend, fees of 7–10% may be possible with committed spend agreements. The fee conversation should always include: what is included in the fee (technology, data, support) and what costs extra (premium audience data, viewability measurement, brand safety activation). Hidden fees that are not disclosed upfront — data costs, tech fees, priority access fees — are common in programmatic and should be contractually addressed.

Advanced fraud prevention goes beyond switching on IAS or DoubleVerify pre-bid filtering. Sophisticated fraud operations have evolved to avoid basic detection: domain spoofing is prevented by ads.txt validation; ad stacking is detectable by viewability measurement; but more sophisticated forms — datacenter traffic masquerading as residential IPs through VPN botnets, fraudulent app traffic using device ID farms, and sophisticated invalid traffic (SIVT) designed to mimic human browsing behaviour — require more advanced detection.

The Trustworthy Accountability Group (TAG) Certified Against Fraud programme requires participating platforms to implement documented IVT filtering at the GIVT level as a minimum standard. For SIVT, accredited third-party measurement (IAS, DoubleVerify) with MRC accreditation is the industry standard — but no solution catches 100% of sophisticated fraud.

Brand safety at scale requires more than GARM category blocking. Context is more complex than category — a page about "drug policy reform" is different from a page about "illegal drug use" despite both potentially falling into the same broad content category. Advanced brand suitability configurations use semantic NLP scoring (IAS Proximity, DoubleVerify Authentic Brand Suitability) to assess the specific meaning and tone of page content, not just its broad category.

CTV programmatic is the fastest-growing and most structurally complex segment of programmatic. Advanced CTV strategy requires understanding the fragmented ecosystem, the measurement limitations unique to the channel, and the targeting approaches that actually work in a cookieless, device-ID-dependent environment.

CTV supply hierarchy: premium streaming (Netflix, Disney+, Amazon Prime Video) is primarily sold direct or through managed service, not open programmatic. Mid-tier streaming (ITVX, Pluto TV, Tubi, Peacock free tier) is accessible programmatically through specialised CTV DSPs. AVOD (ad-supported VOD) content on smart TV home screens and FAST (Free Ad-Supported Streaming TV) channels are the most programmatically accessible inventory.

The frequency management challenge: in CTV, frequency is measured at the household level, not the individual level. A household with four viewers might see the same ad four times in one evening across four viewing sessions — all registering as four impressions for the same household but each experienced by a different person. Household-level frequency caps are the only practical control available, and even these are limited by cross-platform (streaming service vs. device OS vs. DSP) frequency coordination gaps.

A complete programmatic measurement architecture integrates: ad server impression data (Campaign Manager 360 or equivalent as the system of record for delivery); independent verification (IAS or DoubleVerify for viewability, brand safety, and IVT); attribution platform (GA4 or independent MTA platform); incrementality testing (DSP built-in or third-party tools like Measured); and brand lift studies (annual or per-major-campaign panel research). Each layer answers a different question and none of them should be used in isolation.

The reconciliation process between layers reveals the measurement gaps. When DSP-reported impressions and ad server impressions diverge by more than 10%, there is a discrepancy worth investigating — usually a trafficking error, a verification vendor blocking rate, or a DSP counting methodology difference. When attributed conversions significantly exceed modelled incremental conversions, the channel is harvesting organic demand rather than generating it. These insights are only visible when the measurement layers are designed to cross-reference each other.

The documented trend toward programmatic in-housing — brands managing DSP buying directly rather than through agency trading desks — reflects real economic and transparency incentives: agency fee structures are not always transparent; data access may be restricted by agency contracts; and the learning curve for sophisticated programmatic is now manageable for most organisations with a dedicated team.

The case for in-housing: full access to campaign data (agencies sometimes restrict raw data export); no agency margin on media (10–15% saving on media cost); direct relationship with DSP and SSP partners; faster iteration on campaign settings; and cumulative institutional knowledge that stays with the brand rather than an agency.

The case for staying with an agency: agencies have scale advantages in fee negotiation (£50M agency spend gets better DSP rates than £2M brand spend); specialist expertise in niche formats (CTV, DOOH, audio) that may not justify a full-time in-house specialist; and operational capacity during peaks that may exceed a lean in-house team's bandwidth. The decision is rarely binary — a hybrid model (brand owns strategy and measurement; agency handles execution) increasingly characterises sophisticated programmatic organisations.

Further Reading

Go deeper with these reference guides from the Digital Codex library.

Sources & References