Programmatic Display Advertising: Costs, Where the Money Actually Goes, and the CPVD Alternative

How programmatic display actually works (the supply chain)

Programmatic display is what happens when you stop buying ads from a publisher's sales team and start buying impressions from a real-time auction. A user loads a webpage; the publisher's SSP (supply-side platform) sends a bid request describing the slot, the user, and the page; DSPs (demand-side platforms) bid on behalf of advertisers; the highest bid wins and the creative renders. The whole transaction takes roughly 100 milliseconds.

There are at least four distinct counterparties between an advertiser's dollar and a publisher's pageview, and each takes a margin: the agency or trading desk running the buy, the DSP bidding into the auction, the ad exchange / SSP running the marketplace, and the publisher's tech stack (ad server, header bidder, identity vendor) on the way out. Stack on top of that data providers, verification vendors (viewability, brand safety, IVT detection), and identity resolution services, and a single impression can be routed through six to ten paid layers.

This is the supply chain the ANA and the IAB have spent years auditing. The audits did not produce reassuring numbers. See the Middleman Tax for the structural reason this happens: auction-based ad-tech architecturally rewards intermediation.

From the 2023 study's 36 cents to today: what the audit found, and what changed

The ANA Programmatic Media Supply Chain Transparency Study (December 2023, conducted with PwC, TAG TrustNet, Fiducia and Kroll) was the first fully itemized log-level audit published of US programmatic display. Twenty-one advertisers and twelve supply-chain companies shared bid- and impression-level data covering $123 million in spend and 35.5 billion impressions between September 2022 and January 2023. On $1,000 of open-web spend it found $360 of TrueAdSpend: an impression that was viewable, measurable, not invalid traffic, and not on Made-for-Advertising inventory. TrueAdSpend is a quality bar the impression clears, not evidence that a person looked at it. The ANA put the available efficiency gain at $22 billion. That study is the right source for the structure of the supply chain, and it is six quarters stale as a level. We date-stamp it every time and never quote 36 cents as today's number.

The current number comes from the same body. The ANA now publishes the benchmark quarterly. Its current headline covers all environments including connected TV: 45.1% of spend became a benchmark-qualified impression in Q2 2026, with transaction costs at 27.2%. In Q3 2025, the last quarter it itemized web and mobile excluding connected TV, transaction costs took 26.9% of the ad dollar, media-productivity losses took 26.1%, and TrueAdSpend was 47.1%. The two bases are not comparable and never belong on one trend line. On the ex-connected-TV basis the figure rose from 36% in the 2023 study to 47.1%, worth roughly $13.6 billion of recovered efficiency. The Q4 2025 release then split participants at the median of TrueAdSpend: the higher-performing half converted 56.7% of spend into benchmark-qualified impressions, the lower-performing half 37.5%. That is a median split on the outcome, not a split between advertisers who run a transparency program and advertisers who do not.

• Transaction costs, 26.9% today. In the Q3 2025 benchmark, SSP platform fees are the single largest line at 13.2%, DSP data fees are 3.6%, and DSP platform and other costs make up the balance. The December 2023 study itemized the same layer at 29% on $1,000 of open-web spend: $130 SSP platform, $80 DSP platform, $60 DSP data, $20 DSP additional.
• Non-measurable impressions, 15.2%. The largest single loss. Impressions where viewability or fraud could not even be evaluated, because verification tags weren't present, weren't supported on the inventory, or were stripped in the supply path.
• Non-viewable, 10.1%. Impressions served but never meeting the MRC threshold, which is 50% of pixels in view for at least one continuous second for display and two seconds for video. The MRC sets that standard; it does not publish a rate.
• Made-for-advertising inventory, 0.5%. MFA sites are low-quality publishers built specifically to harvest programmatic dollars. On the December 2023 waterfall they took $100 of every $1,000; on the Q3 2025 benchmark they are down to half a point of the dollar among participating advertisers, who are the ones actively filtering.
• Invalid traffic, 0.3%. Bots, click farms and declared fraud, measured at log level. It is the smallest line in the waterfall, and it is why this page does not lead with bots.

Working media (TrueAdSpend)

47.1%

Of the ad dollar, web and mobile excluding connected TV, Q3 2025. Last quarter the ANA itemized that cut; current all-environments headline 45.1% (Q2 2026)

ANA Programmatic Transparency Benchmark, Q3 2025

Transaction costs

26.9%

SSP platform 13.2%, DSP data 3.6%, DSP platform and other costs the balance

ANA Programmatic Transparency Benchmark, Q3 2025

Median split on TrueAdSpend

56.7% vs 37.5%

Higher-performing half of benchmark participants versus the lower-performing half, split at the median (Q4 2025)

ANA Programmatic Transparency Benchmark, Q4 2025

The 2023 structure, date-stamped

36¢

$123M of spend, 35.5B impressions, 21 advertisers; $22B of available efficiency gain. Structure only, not the current level

ANA Programmatic Media Supply Chain Transparency Study (Dec 2023)

Viewability first, fraud second

Ad fraud is the story the industry likes to tell about programmatic waste, and it is the smallest part of it. DoubleVerify's 2025 Global Insights Report, published July 2025 and based on more than one trillion measured impressions, puts fraud and sophisticated invalid traffic at 1.0% in North America and 0.8% in the US. The ANA's Q3 2025 benchmark, measuring at log level against a different sample, puts invalid traffic at 0.3% of spend. Two independent measurements, both about one percent or less. Anyone quoting you a double-digit invalid-traffic rate or a headline dollar total for global ad fraud is quoting a projection, not a measurement, and we do not repeat those on this site.

The real tax is viewability, and its bigger sibling, measurability. The MRC standard is 50% of pixels in view for at least one continuous second for display, two seconds for video. DoubleVerify measures the authentic viewable rate in North America at 71% (73% display, 79% video), so on roughly three impressions in ten nobody had a real chance to see the ad. The ANA's waterfall says the same thing from the money side: 10.1% of the dollar buys non-viewable impressions and another 15.2% buys impressions where viewability could not even be evaluated. A buyer who isn't paying attention is paying for impressions that, by the industry's own definition, did not deliver a viewable ad.

The third thing worth knowing is that the audience is opting out. DoubleVerify's consumer survey puts ad-blocker use among North American consumers at 41%. An interruptive impression is not only leaking value through the chain; a large minority of the people it is aimed at have installed software specifically to refuse it.

Fraud and sophisticated invalid traffic

1.0%

North America, measurement year 2024 (0.8% in the US)

DoubleVerify 2025 Global Insights Report

Invalid traffic, ANA benchmark

0.3%

Of spend, Q3 2025, measured at log level

ANA Programmatic Transparency Benchmark, Q3 2025

Authentic viewable rate

71%

North America; display 73%, video 79%

DoubleVerify 2025 Global Insights Report

Non-measurable plus non-viewable

25.3%

Of the ad dollar: 15.2% non-measurable, 10.1% non-viewable (Q3 2025)

ANA Programmatic Transparency Benchmark, Q3 2025

Where programmatic display still wins

Honest comparison demands honest credit. There are categories where programmatic display does what no other channel can, and where the working-media loss is still acceptable to the buyer because the alternatives are worse.

1. National-scale brand reach. A national CPG, streaming service, or insurance carrier targeting tens of millions of US adults at high frequency cannot achieve that footprint through any single publisher direct buy. Programmatic exchanges aggregate the long tail of inventory at a unit cost no closed network can match, even after the supply-chain rake.
2. Audience modeling at huge volumes. Advertisers with first-party customer files in the tens of millions can use DSPs (The Trade Desk, Google DV360, Amazon DSP) to do lookalike modeling, frequency capping, and cross-device reach measurement at a scale closed environments don't replicate. The math works because the working-media loss is amortized across audiences large enough to absorb it.
3. Real-time creative optimization and dayparting. Programmatic supports dynamic creative, weather-triggered messaging, and granular dayparting that static media buys can't. For categories where creative-context fit moves response rates double-digit percentages (retail, travel, automotive) the optimization upside can offset a chunk of the supply-chain loss.
4. Premium private marketplaces (PMPs). Programmatic display increasingly routes through curated PMPs and direct-deal IDs, where take rates are more disclosed and inventory quality is higher. The ANA measured 81.6% of benchmarked spend running through private marketplaces in Q3 2025, and 99.1% of it in low-risk environments. PMP inventory carries a CPM premium over the open exchange, which buyers absorb because the viewability and brand-safety lift more than offsets it. No source at our sourcing bar publishes an open-exchange or PMP CPM average, so we quote neither.
5. CTV and OLV (online video) inventory. Connected TV inventory has grown into a meaningful share of programmatic spend, and on premium CTV apps (broadcaster-owned, MVPD-distributed) the working-media efficiency is materially better than on long-tail open-web display.

Where programmatic display bleeds money

Programmatic was architected for advertisers whose KPI is reach × frequency at a CPM target. It was not architected for advertisers whose KPI is cost per acquired customer in a defined service area. When you run the latter on a system built for the former, you pay the supply-chain rake without harvesting the reach-and-frequency upside that justifies it.

Run the math for a local service operator. A roofing contractor with a $5,000 monthly programmatic display budget pushes that money through a DSP. At the ANA's Q3 2025 ratio, roughly $2,355 of it arrives as working media. Of that working media, an unknown share is targeted at homeowners with a failing roof in the contractor's actual service area. The rest lands on news sites read by people in adjacent metros, on devices that don't belong to a buyer with a need, in dayparts when nobody is shopping for a roof. The supply-chain tax for every $1 of intended reach compounds against an audience-precision tax that programmatic targeting was never tuned to solve at the local-tunnel level.

There is also no reason to assume a small local advertiser lands in the better half of the benchmark. Every mechanism behind the industry's improvement since 2023 (private-marketplace curation, supply-path optimization, log-level feeds, a verification stack from vendors like DoubleVerify or IAS, and staff whose job is supply-chain governance) is something a business spending a few hundred dollars a month cannot buy. On the ANA's Q4 2025 median split the lower-performing half converted 37.5% of spend into benchmark-qualified impressions, which is statistically where the whole industry sat in 2023. That is our read, not the ANA's definition, and it is the honest planning assumption for a local buy. See what is impression fraud for the fraud-mechanism breakdown.

Major DSPs and SSPs as industry context

The US programmatic stack is concentrated. On the demand side, two independent DSPs (The Trade Desk and Google DV360) plus Amazon DSP route the majority of open-web buying. On the supply side, Magnite, PubMatic, OpenX, Index Exchange, and Google AdX dominate exchange-side flow. Naming them is industry context, not a head-to-head challenge.

The Trade Desk (TTD) is the largest independent DSP. It charges a platform fee as a percentage of gross spend, negotiated per advertiser and not published as a rate card. The analyst estimates that circulate for it are not verifiable at our sourcing bar, so this page does not carry one.

Google DV360 is bundled inside Google Marketing Platform. DV360 doesn't disclose a discrete platform fee. Its margin is embedded in product pricing and into the AdX exchange that often clears DV360 demand.

Magnite (NASDAQ: MGNI) is the largest independent SSP, and PubMatic (NASDAQ: PUBM) is one of its main competitors. Both take a share of what the publisher earns. Analyst reconstructions of those take rates from filings vary widely and are not published by either company, so instead of quoting an estimate, this page uses the one figure that was measured rather than inferred: on the ANA's Q3 2025 benchmark, SSP platform fees are 13.2% of the ad dollar, the largest single line in the 26.9% transaction-cost layer.

None of these companies are the problem. The architecture is: every layer that can charge a fee will charge a fee, and the auction model rewards adding layers (data, identity, verification, supply-path optimization) faster than it rewards removing them.

CPVD as the precise alternative

Cost Per Verified Delivery (CPVD) is what you build when you start from "the operator should pay for one verified delivery to a real person driving in a chosen geography" instead of "the operator should pay a CPM and trust the supply chain." WilDi Maps' delivery is a GPS-confirmed person driving entering a geography the operator owns. Not an impression, not a click, not a bid request.

WilDi runs a three-tier model rather than one flat unit. Each tier maps to a different precision and intent profile:

• Tunnels: a 1-mile road strip the operator owns. Hyper-local, premium tier. Built for arrival routes, exit ramps, and neighborhood-tunnel targeting where the buyer knows exactly which mile of road matters to their service area.
• Zones: a neighborhood-sized area of about 0.28 square miles, one H3 hexagon. Hyper-local, premium tier. Built for service-area saturation when the operator wants everyone driving through a specific neighborhood reached.
• Background: city-wide rotation at tier-based per verified delivery. Built for breadth, brand presence, and lower-cost reach across a metro.

How CPVD changes the unit economics

Three things change versus programmatic display: location is reported by the device itself rather than inferred from a bid-stream signal that may have been stripped or guessed; the unit is one verified person driving in your chosen geography during your flight, not a thousand impressions trying to model an audience; and there is no auction-rake cascade. No DSP fee, no SSP fee, no data fee, no MFA arbitrage layer, no unknown delta.

When a person claims a delivery, they can direct-drive to the operator's location, click through to the operator's website, or open the operator's app page. CPVD pricing starts from $0.25+ (background, tier-based); tunnels and zones are priced for hyper-local precision. Every dollar you spend maps to a logged delivery; every dollar that didn't deliver is a dollar you didn't spend.

For a local service operator running a measurable CAC model, the architecture difference matters more than the unit-cost difference. See what is Cost Per Verified Delivery for the full model and the Middleman Tax for the structural reason CPVD exists.

CPVD vs programmatic display open exchange vs PMP

Side-by-side on the dimensions a local service operator (or a national brand evaluating diversification) actually weighs.