In-DSP measurement is the measurement that a demand-side platform (DSP), the software advertisers use to buy programmatic ads, performs on the campaigns it runs. It covers delivery reporting, conversion attribution, lift experiments and third-party studies that are booked, run or displayed inside the buying platform rather than in a separate tool. It exists because the DSP already sees every bid, impression and click it handles, so it can report on results at very little extra cost. The trade-off is independence: the company that sold the media also counts what the media did.

The phrase is descriptive rather than a product name. Google's Display & Video 360 (DV360), The Trade Desk, Amazon's DSP and Nexxen all offer versions of it, usually set against independent measurement by a separate company.

Three layers inside the platform

Native reporting and attribution. The DSP logs impressions, clicks and spend from its own servers. A tag on the advertiser's site, such as Google's Floodlight, lets the platform credit a sale to an earlier ad within an attribution window. DV360 offers last-click, first-click and data-driven models. Amazon applies its own rules to retail data: until 2026, purchases after an ad view on several formats, including some DSP campaigns, were credited within a 14-day window as standard. This layer shows correlation, not cause: a buyer who would have purchased anyway is still credited.

Platform-run experiments. To estimate cause, DSPs split audiences into an exposed group and a control group that is not shown the campaign. DV360's Help Center says its Conversion Lift control users see "the next ad in the auction" instead of the advertiser's ad. Google asks for at least 3,000 conversions over the study, recommends a budget of at least $5,000 and runs studies for 7 to 56 days. Its Brand Lift product surveys both groups on up to three metrics, such as awareness or ad recall, with YouTube minimums of $5,000 to $15,000 per question by country group.

Sample size governs what such a test can detect. According to the DV360 help pages, 1,200 to 2,800 survey responses detect lift above 4%, while more than 180,000 are needed to detect lift below 0.5%. Google reports "certainty of lift" as one minus the p-value, a standard test of whether a result could be chance. Results appear only when certainty exceeds 70%, and Google describes 90% or higher as suitable for decisions.

A worked example shows why the distinction matters. Suppose a campaign reaches 1 million people and a 20% holdout keeps 200,000 of them from seeing it. If 1,600 of the 800,000 exposed people buy (0.20%) against 240 of the control group (0.12%), the incremental conversion rate is 0.08 percentage points, or about 640 sales caused by the ads. Last-touch attribution in the same platform could credit all 1,600. The figures are illustrative, but the gap between attributed and incremental results is the reason lift testing exists.

Other platforms use different mechanics. The Trade Desk said on August 19, 2026 that Kokai, its buying interface, now applies a 20% holdout to conversion lift studies by default. Nexxen introduced ghost-bidding incrementality in March 2026, building a synthetic control from auctions the advertiser would have entered. The "ghost ads" method was set out by Garrett Johnson, Randall Lewis and Elmar Nubbemeyer in a 2017 paper listed by Northwestern's Kellogg School.

Integrated third-party studies. Outside vendors can also be booked inside the DSP. LoopMe's brand lift joined The Trade Desk's Measurement Marketplace in September 2026, with studies from 1 million impressions against 3 million to 4 million at many rivals. Amazon opened a self-serve Studies page in May 2026 covering more than 50 vendor products, including Kantar Brand Lift, DISQO Brand Lift and NCS offline sales lift. Advertisers choose between "Managed within Amazon DSP", where Amazon configures and invoices the study, and "Managed with Vendor" for existing contracts. The vendor counts, but the DSP controls setup, data flow and billing.

Agency trading desks and in-house teams configure studies; DSP data-science teams build the models and set thresholds. Publishers rarely see in-DSP results, since their counts come from ad servers and supply-side platforms.

Origin and evolution

DSP reporting is as old as the DSP. Google bought Invite Media on June 3, 2010, and its successor, DoubleClick Bid Manager, reached general availability on October 24, 2012.

The causal layer arrived later. Google announced Google Measurement Partners in July 2018, a programme of more than 20 verified vendors covering viewability, brand lift and modelling. Babak Pahlavan, then a senior director at Google, said: "Google strives to build great solutions, (but) some marketers prefer to rely on third-party measurement solutions." DV360 itself had been announced under that name on June 27, 2018. In October 2018 Google said Brand Lift studies could be set up directly in DV360, running continuously through a campaign. The same announcement introduced "lifted users", "cost-per-lifted-user" and a Maximize Lift bid strategy that tied survey results to bidding.

Google announced Nielsen Catalina Solutions sales lift for connected TV in DV360 in July 2021, with no pixel required. Cross-exchange Brand Lift reached connected TV devices in November 2023.

Amazon built on retail data. Its Omnichannel Metrics product, drawing on the Amazon Shopper Panel launched in October 2020, has since July 2026 named individual retailers in DSP reporting. Embedded brand lift then spread to smaller platforms, including illumin in May 2026 and Basis in June 2026, both using Cint.

Why it matters to marketers

In-DSP results now drive the decisions they report on. Lift and conversion data feed automated bidding, so measurement choices become spending choices. Google's cut to a $5,000 incrementality testing threshold put experiments within reach of mid-sized budgets that once relied on attribution alone.

Convenience is the other draw: no separate contracts, tags or log files. Giovanni Gardelli, head of product at Yahoo DSP, argued in January 2026 that "for nearly two decades, measurement has lived in two separate worlds", attribution and incrementality, and predicted they would converge inside platforms.

Retail media has sharpened the question. Incremental estimates US retail media spend at about $70 billion in 2026, with Amazon taking more than three-quarters, and much of that spend is judged by the retailer's own DSP reports. Justin Manus, chief technology officer at the agency Tinuiti, said in Nexxen's March 2026 announcement: "Attribution modelling only goes so far in understanding our return on investment."

Limitations and disputes

The central objection is conflict of interest. A DSP paid from media spend designs the experiment, chooses the control group, sets the reporting threshold and publishes the headline. The Trade Desk said in October 2026 that retail data delivered 41% higher return on ad spend, without sample size or control design.

Rules can also change beneath the numbers. Amazon shortened view-through attribution from January 1, 2026, keeping the old 14-day figures in a separate metric family. Google expects its move to begin-to-render billing to cause a "slight decrease in overall billed impression volume."

Independent audit coverage is thin. Amazon DSP's Industry Standard report gained additional Media Rating Council (MRC) accreditation in April 2024, and IAS earned accreditation for its Amazon DSP integration in November 2025. The MRC's fourth-quarter 2025 status update lists no new decisions for DV360 or The Trade Desk. Accreditation, where it exists, covers counting (impressions, viewability, invalid traffic), not the lift models that estimate cause, and no common standard governs how platforms build control groups or report significance. Google's choice to show results from 70% certainty, below the 90% it recommends for decisions, illustrates how much is left to each platform.

Siloed results also miss other channels: Incremental, a measurement firm, found platform attribution misses 36% to 53% of true impact. And Adelaide attention scores in DV360 bidding let an outside metric shape the auction it is meant to judge.

Not the same as

  • Third-party verification. Firms such as IAS and DoubleVerify count impressions, viewability and invalid traffic with their own tags. Shown in a DSP dashboard, the count stays independent; the integration does not.
  • Ad server reporting. An ad server such as Campaign Manager 360 records delivery across many DSPs and publishers, often as the advertiser's reference count. Discrepancies are settled under contract; the IAB's 2026 Direct Buy framework uses a 10% threshold.
  • Clean room measurement. A clean room, such as Amazon Marketing Cloud, lets an advertiser run its own queries on joined data, beside the DSP rather than in its standard reports.
  • Other meanings. In engineering, "DSP measurement" refers to digital signal processing chips, and in music distribution, DSP means digital service provider. Neither relates to advertising.

Recent developments

Measurement has become a selling point in 2026. The Trade Desk simplified Kokai lift studies in August and added LoopMe in September, with its shares still weighed down by a 24% fall on weak third-quarter guidance. Google set February 2027 for begin-to-render billing in DV360 and Campaign Manager 360. Amazon added weekly brand penetration indices for nine countries on October 2, 2026. According to AdExchanger, Amazon rebranded its DSP at unBoxed on September 29, 2026, under the DVA name; PPC Land coverage still says Amazon DSP, so naming is unsettled as of October 2026.

Timeline

  • June 3, 2010 - Google acquires Invite Media, the basis of its DSP.
  • October 24, 2012 - DoubleClick Bid Manager reaches general availability.
  • 2017 - Johnson, Lewis and Nubbemeyer publish the "ghost ads" method for online ad experiments.
  • June 27, 2018 - Google announces Display & Video 360 within Google Marketing Platform.
  • July 2018 - Google Measurement Partners is announced with more than 20 vendors.
  • October 2018 - Brand Lift studies become available directly in DV360.
  • October 2020 - Amazon Shopper Panel launches in the United States.
  • July 2021 - Google announces NCS sales lift for connected TV in DV360.
  • November 2023 - DV360 cross-exchange Brand Lift extends to connected TV.
  • April 2024 - Amazon DSP announces additional MRC accreditation.
  • September 2024 - Enhanced attribution becomes mandatory in DV360.
  • November 13, 2025 - IAS gains MRC accreditation for Amazon DSP measurement.
  • January 1, 2026 - Amazon applies shorter view-through attribution.
  • March 24, 2026 - Nexxen unveils ghost-bidding incrementality in its DSP.
  • May 15, 2026 - Amazon opens self-serve access to more than 50 third-party studies.
  • July 7, 2026 - Amazon names individual retailers in Omnichannel Metrics.
  • August 19, 2026 - The Trade Desk sets a default 20% holdout for Kokai lift studies.
  • September 2, 2026 - LoopMe brand lift joins The Trade Desk's Measurement Marketplace.
  • September 29, 2026 - Amazon rebrands its DSP at unBoxed, according to AdExchanger.
  • October 2, 2026 - Amazon adds brand penetration indices to its DSP tools.
  • February 2027 - DV360 and Campaign Manager 360 billing moves to begin-to-render counting.

Summary

Who. DSP operators such as Google (DV360), The Trade Desk, Amazon, Yahoo, Nexxen and Viant; measurement vendors such as Kantar, Cint, LoopMe, NCS and IAS that plug into them; and the agencies and advertisers who read the results.

What. Reporting, attribution, lift experiments and embedded third-party studies run or shown inside the platform that bought the media, as opposed to measurement by an independent party.

When. Delivery reporting dates to the first DSPs around 2010 to 2012. In-platform Brand Lift reached DV360 in 2018, sales lift followed in 2021, and self-serve study marketplaces spread through 2026.

Where. In DSP dashboards and study pages, including DV360's Experiments section, The Trade Desk's Measurement Marketplace and Amazon's Studies page, across display, video, connected TV and audio.

Why. The DSP already holds the bid and impression data, so measurement inside it is fast and cheap and can feed bidding directly. The same position makes it a party to the result, which is why independent audit and outside measurement remain contested.