All resources

· 36 min read

12 Benefits of AI in Programmatic Advertising in 2026

The 12 benefits of AI in programmatic advertising in 2026, from smarter bidding to fraud control and agentic buying, and how Lapis turns them into revenue.

Lapis, the YC-backed #1 AI ad generator rated 4.9/5 on G2, uses AI agents to create, launch, and optimize ads on ChatGPT, Google, Meta, and LinkedIn.

AI improves programmatic advertising in 2026 in 12 measurable ways, from pricing every impression on its odds of producing a sale to buying media through software agents, and the advertisers who keep those gains are the ones who judge every model on confirmed revenue rather than on the platform's own report.

The plumbing is already automated at enormous scale, with eMarketer forecasting on June 24, 2026 that US programmatic ad spending would pass $220 billion this year and that walled gardens would take 80.4% of it. The Association of National Advertisers found in its Q2 2026 benchmark, published August 26, 2026, that only 45.1% of tracked programmatic spend reached a measurable, viewable, fraud-free impression outside made-for-advertising sites. Closing that gap is now a modeling problem rather than a staffing problem, which is why The Trade Desk, Google, Amazon, Viant, and StackAdapt each rebuilt their buying products around AI agents this year, and why our 50 AI advertising statistics for 2026 read like a migration of budget toward machines.

Lapis captures these benefits where most performance budgets already sit, in the automated auctions of Meta, Google, LinkedIn, Reddit, and ChatGPT, supplying the creative volume and matched landing pages that AI bidding needs and measuring every channel against confirmed sales through Wayfinder. For open-web and CTV inventory bought through a DSP, Lapis runs beside the DSP as the advertiser's own buyer. As our founder argues in why Lapis is building the AdSense for the AI era, programmatic automated the path an ad travels, and agentic advertising automates the buyer, the institution that decides what the ad should do.

What Are the Benefits of AI in Programmatic Advertising?

AI gives programmatic advertising 12 benefits: smarter bid pricing, predictive audiences without third-party cookies, contextual targeting, dynamic creative, fraud detection, supply path savings, brand suitability, pacing, cross-channel allocation, creative testing at volume, better measurement, and agentic buying. The table pairs each benefit with the job the model does and the strongest 2026 evidence behind it, so every number arrives with its source and date.

#BenefitWhat the AI doesHeadline 2026 evidence
1Smarter bid pricingPrices each impression on predicted conversion valueThe Trade Desk: 32% average CPA improvement from Kokai Zuma modeling (Aug 27, 2026)
2Predictive audiences without cookiesBuilds audiences from first-party and platform signalsAmazon: Brand+ and Performance+ improved CPA by up to 68% (Sep 29, 2026)
3Contextual targetingClassifies pages, video, and conversations in real timeDoubleVerify: DV Neura raised classification output nearly 300 times (Jun 17, 2026)
4Dynamic creativeGenerates and assembles creative per audience and placementPeer-reviewed test: AI banner beat stock photo CTR, 0.80% vs 0.53% (Mar 2025)
5Fraud detectionFilters bots and fake inventory before the bidDoubleVerify: 2.8% fraud unprotected vs 0.6% protected (Jul 29, 2026)
6Supply path savingsRoutes bids through the cleanest, cheapest pathANA: top performers ran on about 19,000 fewer domains and apps (Aug 26, 2026)
7Brand safety and suitabilityScores content adjacency at page levelIAS: brand safety fail rate fell from 3.7x to 1.0x baseline in 2025 (Jul 9, 2026)
8Pacing and budget allocationShifts spend across hours, formats, and frequencyThe Trade Desk: Koa frequency optimization agent (Aug 27, 2026)
9Cross-channel allocationMoves budget between open web and walled gardenseMarketer: walled gardens hold 80.4% of US programmatic spend (Jun 24, 2026)
10Creative testing at volumeGenerates and rotates variants as experimentsAlphabet: Gemini helped create 1.5 billion+ assets in Q3 (Oct 28, 2026)
11Better measurementRuns MMM, geo experiments, and clean-room matchingGoogle: Meridian GeoX open to all advertisers worldwide (Sep 10, 2026)
12Agentic buying and labor savingsAgents plan, negotiate, book, and optimize mediaIAB Tech Lab: Agentic Real Time Framework final and in production (Aug 12, 2026)

Read the table as a ledger of who keeps each gain. The first eight benefits accrue to whoever operates the platform well, while the last four, cross-channel allocation, creative volume, measurement, and agentic buying, pay off only when someone working for the advertiser sits above every platform and decides where the next dollar goes. That split explains why AI has raised the ceiling on programmatic performance without raising the floor for most advertisers, a pattern our guide to how AI is changing advertising traces across every channel.

How Does AI Work Inside Programmatic Advertising?

AI works inside programmatic advertising by replacing the hand-set rules that DSPs, SSPs, and exchanges used to apply in each real-time auction with models that predict the value of every impression and act on that prediction in milliseconds.

The mechanics have not changed since real-time bidding arrived: a publisher's supply-side platform offers an impression to exchanges, each demand-side platform decides whether and how much to bid for its advertisers, and the winning ad renders before the page or stream finishes loading. What changed is who makes the decision inside that fraction of a second. The Trade Desk said on August 27, 2026 that its Koa AI analyzes more than 20 million impressions every second, a volume no trading desk can inspect by hand. Google said on March 23, 2026 that Gemini models now power the marketplace, campaign setup, and reporting inside Display & Video 360.

LayerRule-based programmaticAI-driven programmatic in 20262026 example
BiddingFixed CPM bids and manual multipliersPredicted conversion value per impressionKokai Zuma forecasting and Koa agents
AudiencesThird-party cookie segmentsFirst-party lists, clean rooms, model-built audiencesAmazon DVA+ on shopping and streaming signals
ContextKeyword block listsMultimodal classification of pages, video, conversationsGemini bid optimization in DV360; DV Neura
CreativeA few static banners trafficked by handGenerated variants per audienceStackAdapt AI Video Builder; MNTN QuickFrame AI
SupplyEvery exchange and every pathInclusion lists, direct paths, partner code in the auctionViant Direct Access; IAB Tech Lab ARTF
MeasurementLast-click and platform reportsMix models, geo experiments, clean roomsGoogle Meridian GeoX; AdCP 3.2
WorkflowTraders building line itemsAgents that read media plans and book dealsDV360 Ads Advisor; Viant Outcomes; AAMP agents

The human job in programmatic has moved up a level, from setting bids and building line items to choosing goals, supplying inputs, and checking results. Analysts at eMarketer noted on June 24, 2026 that buyer and seller agents are taking over the exchange of briefs and spreadsheets, so teams can stop managing individual line items. Our analysis of the future of ad agencies and ad tech shows how that shift is reshaping the intermediaries that used to sell the labor.

The 12 Benefits of AI in Programmatic Advertising

The 12 benefits below run from the auction outward, starting with how AI prices a single impression and ending with agents that run the whole buy, and each one carries its dated evidence and the action it calls for.

1. Smarter Bid Pricing on Every Impression

AI bid pricing values each impression on its predicted chance of producing the outcome you pay for, rather than on a flat CPM, and it is the best-proven AI benefit in programmatic. The Trade Desk said on August 27, 2026 that the upgraded forecasting and modeling in its Kokai Zuma release had produced an average 32% improvement in CPA in initial results. Viant chief executive Tim Vanderhook told investors on August 12, 2026 that the company's autonomous Outcomes product beats human-run campaigns by 65% on average.

Advertisers capture those gains by pointing the bidder at a conversion your own systems confirm, such as a booked demo in your CRM or a paid order in your store, and holding out 10% of the audience to prove what the model added. A bidder told to chase cheap clicks will find them reliably and report success while revenue stays flat.

2. Predictive Audiences Without Third-Party Cookies

Predictive audiences replace the third-party cookie segment with models trained on first-party data, logged-in platform signals, and clean-room matches, so they work whatever browsers decide. Chrome kept third-party cookies, with Google confirming on April 22, 2025 that it would not roll out a new cookie prompt and retiring most Privacy Sandbox ad APIs on October 17, 2025. Safari and Firefox still block cross-site tracking by default, which leaves cookie audiences thin on a large share of devices.

Amazon said on September 29, 2026 that DVA+, its new AI-first way to buy display, video, and audio, draws on trillions of shopping, browsing, and streaming signals, and that Brand+ and Performance+ optimization had improved cost per acquisition by up to 68%. Google said on March 23, 2026 that its audience segments in Display & Video 360 reach 96% of ad-supported CTV households through Confidential Publisher Match. Upload your customer lists, add a clean room once your volume justifies one, and use our answer to whether retargeting is dead for the full cookieless playbook.

3. Contextual Targeting That Reads the Page or Conversation

Contextual AI reads the meaning of the page, video, or conversation beside an ad, reaching intent without identifying anyone. Google said at its NewFront on March 23, 2026 that Gemini's multimodal understanding now informs bid optimization in Display & Video 360, weighing what a creative shows and what a live broadcast contains. DoubleVerify said on June 17, 2026 that its DV Neura engine had raised content classification output nearly 300 times, enough to score the bid stream itself rather than a sample of pages.

IAB Tech Lab also folded Mixpeek's contextual enrichment into its AAMP 2.3 agent release so that buyer and seller agents describe context in one taxonomy, as IAB Canada summarized on August 14, 2026. Inside ChatGPT the conversation is the context, as our guides to where ads appear in ChatGPT and other AI assistants and what ChatGPT ads are explain. Brief contextual categories with the same care as audiences, and test them as separate line items so the results stay readable.

4. Dynamic Creative Optimization

Dynamic creative optimization uses AI to assemble or generate a version of the ad for each audience, product, and placement, and in 2026 generation moved inside the buying platforms. Amazon's DVA+ makes reviewing AI-generated creatives one of its four setup steps, and StackAdapt opened a beta of its AI Video Builder for video and CTV ads on May 20, 2026. MNTN said on its August 4, 2026 earnings call that its free QuickFrame AI creative tool had passed 73,000 signups this year.

The single-asset case is settled: a peer-reviewed field test by Hartmann, Exner, and Domdey, published in the International Journal of Research in Marketing in March 2025, found a DALL-E 3 banner beat a professional stock photo on click-through rate, 0.80% against 0.53%. The trap is variation without difference, because swapping colors on one idea teaches the model nothing about why people buy. Feed the optimizer distinct angles, as our guide to AI ad angles versus variations shows, and send each angle to a matching page built with dynamic AI landing pages.

5. Fraud and Invalid Traffic Detection

AI fraud detection scores every bid request for bots, data-center traffic, spoofed apps, and fake sites before money changes hands. DoubleVerify's 2026 Global Insights report, distributed on July 29, 2026, found a fraud and sophisticated invalid traffic rate of 2.8% in unprotected campaigns against 0.6% in protected ones, and its Fraud Lab found that AI bots generated 15% of clicks in unprotected media.

The threat is moving to television, where DoubleVerify said on May 7, 2026 that CTV fraud schemes rose 140% year over year in the first quarter. Unprotected CTV campaigns ran nearly 9% fraud against less than 1% when protected, DoubleVerify found, costing unprotected advertisers about $1.8 million per billion CTV impressions. Integral Ad Science put global invalid traffic at 1.1% for 2025, and 1.4% in the United States, in its Media Quality Report on July 9, 2026. Turn on pre-bid filtering in every DSP, and strip invalid traffic from reports before anyone judges creative or channels on them.

6. Supply Path Optimization and Lower Fees

Supply path optimization uses AI to choose the cleanest, cheapest route to each impression, cutting resellers, duplicate auctions, and long-tail sites that absorb fees without adding reach. The ANA's Q2 2026 Programmatic Transparency Benchmark, published August 26, 2026, found that transaction costs took 27.2% of programmatic spend, and that its higher performers ran across about 19,000 fewer domains and apps than lower performers while converting 52.3% of spend into qualified impressions against 31.1%.

The ANA's original study, published December 5, 2023, found the average campaign ran on 44,000 websites and recommended working with 75 to 100 trusted sellers instead. Viant said on August 10, 2026 that more than 80% of its CTV spend runs through its Direct Access program. IAB Tech Lab's Agentic Real Time Framework, declared final on August 12, 2026, lets a buyer run its own bidding logic in a sealed container inside an SSP. Ask every DSP for log-level data, buy from an inclusion list, and treat any path you cannot see as a cost.

7. Brand Safety and Suitability

AI brand safety and suitability tools classify the content around each impression against a brand's rules, and they now catch AI-generated junk sites that keyword lists never saw. Integral Ad Science reported on July 9, 2026 that global brand safety fail rates fell from 3.7 times its baseline in January 2025 to 1.0 by December, while mobile web display carried 54.9% of suitability violations and 71.9% of made-for-advertising impressions despite holding 45.1% of impressions.

DoubleVerify's Fraud Lab exposed a network of more than 200 domains on March 4, 2026 that used large language models to mass-produce clickbait. The ANA said on May 27, 2026 that made-for-advertising exposure rose to 1.1% of benchmark spend after holding between 0.4% and 0.6% through 2025, naming AI slop as the emerging cause. Set suitability tiers by category, apply them before the bid, and review mobile web inventory first, because that is where the waste concentrates.

8. Pacing and Budget Allocation

AI pacing spreads budget across hours, formats, and frequency levels according to where the next dollar works hardest, instead of spending evenly until the money runs out. The Trade Desk added a Koa frequency optimization agent in Kokai Zuma on August 27, 2026, and Google said on March 23, 2026 that its live sports suite in Display & Video 360 uses AI to absorb broadcast bidding spikes. Amazon says DVA+ handles inventory selection, format allocation, bidding, and optimization once you set an objective and a budget.

Pacing models are excellent at spending money well and indifferent to whether you should be spending it at all. Write the guardrails before you switch them on: a daily ceiling, a maximum cost per confirmed sale, and a stop rule that pauses spend when that cost runs 30% above target for three consecutive days, the discipline our AI ad optimization playbook applies to budgets and bidding.

9. Cross-Channel Allocation Across Open Web and Walled Gardens

Cross-channel allocation carries the most money and the least independence, because each platform's AI moves budget only among its own placements. The research firm eMarketer forecast on June 24, 2026 that walled gardens would take 80.4% of US programmatic ad spending this year, so most automated buying already happens inside Meta, Google, and Amazon. Amazon opened Full-Funnel Campaigns to US advertisers on September 29, 2026, combining sponsored ads, display, video, and streaming TV in one AI-run campaign, and Google said on March 23, 2026 that advertisers who added another Google Marketing Platform product saw a 76% lift in ROAS.

Those gains are real, but each is earned inside one company's inventory, and none of these systems will move a dollar from Amazon to Meta, or from the open web to ChatGPT, when the next sale is cheaper there. That decision needs a buyer above every platform that counts conversions once, the role our breakdown of an AI tool versus an AI employee describes, alongside the settings in our guides to Meta Ads AI features and Google Ads AI features.

10. Faster Creative Testing at Volume

AI creative testing turns the brief into a stream of live experiments at a volume no design team can match. Alphabet said on October 28, 2026 that Gemini helped advertisers generate more than 1.5 billion campaign assets in the third quarter, up more than seven times year over year. Meta said on October 21, 2026 that nearly 10 million advertisers used its generative AI creative tools in the same quarter.

Volume pays only when the variants test different ideas, and Meta researchers showed why in a paper revised December 15, 2025: AdLlama, an ad text model trained on performance data, lifted click-through rate 6.7% across nearly 35,000 advertisers because it learned from results rather than imitating past copy. Give every variant a hypothesis and enough delivery to reach 30 conversions before you judge it, as our playbooks for AI ad creative testing at scale and why creative volume testing matters lay out.

11. Better Measurement: Incrementality, MMM, and Clean Rooms

AI measurement replaces last-click reporting with models and experiments that estimate what advertising caused, the one number that justifies every other benefit. Google said on September 10, 2026 that Meridian GeoX, its open-source geo-experiment library, is now available worldwide and calibrates its Meridian marketing mix model with lift results from any ad platform. OpenAI added geo-based incrementality tests with Haus, Measured, and WorkMagic for ChatGPT Ads on October 5, 2026.

AgenticAdvertising.org said at its October 1, 2026 town hall that AdCP 3.2 supports clean-room activation across Snowflake, Databricks, Google Cloud, Azure, and AWS as an experimental feature, while The Trade Desk said on August 6, 2026 that its new measurement framework was still in alpha. StackAdapt said on May 20, 2026 that 41% of conversions happen without a paid click, which is why click-based attribution undercounts programmatic. Measure every channel against confirmed sales or leads from your CRM, store, or billing system, counted once, which is what Lapis Wayfinder does, and keep holdouts running, as our ROI measurement framework details.

12. Agentic Buying and Labor Savings

Agentic buying hands the trader's workflow to software agents that read a brief, find inventory, negotiate, book, and optimize, and 2026 is the year its standards reached production. IAB Tech Lab declared its Agentic Real Time Framework final on August 12, 2026, after several companies had already deployed it, and its AAMP program publishes open-source buyer and seller agents that book programmatic guaranteed deals, private marketplaces, and open-auction buys.

AgenticAdvertising.org, which publishes the Ad Context Protocol, ran the first agent-to-agent media buy on October 16, 2025 and released AdCP 3.2 on October 1, 2026. Yahoo DSP opened an Agent Network of 23 partners on June 18, 2026, and eMarketer reported on June 24, 2026 that PubMatic expects agents to execute 25% of ad buys by 2028. Our agentic ads guide shows how to write permission rules first, covering what an agent does alone, what needs approval, and what it never touches, and our forecast on whether AI will replace media buyers by 2030 covers what happens to the people.

What Are the Risks of AI in Programmatic Advertising?

The risks of AI in programmatic advertising are black-box decisions, platforms grading their own results, AI-made junk inventory, fees that grow with every feature, and privacy exposure, and each one has a direct fix.

Black boxes come first because agents now act on real money. IAB Tech Lab's AAMP 2.3 release, summarized by IAB Canada on August 14, 2026, removed a failure in earlier agent versions that let them fabricate CPM values when pricing data was missing, and added a pricing provenance field so every negotiated price traces back to a source. Demand the same from any vendor: a change log, a reason for every budget move, and a permission model that separates what the agent does alone from what it proposes.

Self-grading is the structural risk that no setting inside a platform can fix. With 80.4% of US programmatic spend inside walled gardens in eMarketer's June 24, 2026 forecast, most of the money runs through platforms whose AI both sells the impression and counts the sale, and none of them deduplicates a customer who also clicked another channel's ad. Junk inventory is the newest risk, with the ANA's May 27, 2026 benchmark showing made-for-advertising exposure doubling to 1.1% as AI slop sites multiply.

Fees and trust remain the oldest risk in the open-web supply chain. The Trade Desk's 2025 revenue of $2.9 billion on $13.4 billion of gross spend, reported February 25, 2026, implies a 21.6% take rate. Publicis advised clients in March 2026 to avoid the platform after an audit it commissioned questioned fees, which The Trade Desk denied, as MediaPost reported on March 18, 2026. Adweek had reported on February 19, 2026 that Dentsu and WPP had stepped back from its OpenPath product over transparency concerns. Privacy risk grows as agents touch more data, which is why IAB Tech Lab's real-time framework runs partner code in containers with no external network access by default.

The fix for all five is the same structure: independent measurement in confirmed revenue and an AI employee that works for the advertiser rather than for the seller of the impression. That is the operating model behind the best AI employees for marketing, and it is why Lapis reports every channel through Wayfinder instead of through the platforms' own dashboards.

Open Web Programmatic vs Walled Gardens vs ChatGPT: Where Does AI Deliver the Most?

AI delivers the largest performance gains in walled gardens and ChatGPT, where one company controls the data, auction, and measurement, and the largest savings on the open web, where fees and junk inventory leave the most money to recover.

FactorOpen web programmatic (DSP)Walled gardens (Meta, Google, Amazon, LinkedIn, Reddit)ChatGPT Ads
Share of US programmatic spend, 202619.6%, the remainder outside walled gardens80.4% (eMarketer, Jun 24, 2026)New channel; $1 billion run rate in under 200 days (OpenAI, Aug 31, 2026)
Whose AI buysKokai, DV360 with Gemini, Viant Outcomes, StackAdapt IvyAdvantage+, AI Max and Performance Max, DVA+OpenAI Ads Manager with AI creative suggestions
Core signalContext, first-party IDs, clean roomsLogged-in first-party behaviorThe conversation itself
Fee layer27.2% transaction costs (ANA, Aug 26, 2026)No separate tech fee; margin sits inside the auctionNo separate tech fee; $25 minimum daily budget in the US
Fraud and junk exposureHighest; needs pre-bid verificationLow on owned surfaces; partner networks varyOwned surface only
MeasurementThird-party verification plus your attributionThe platform grades itselfPixel, Conversions API, incrementality partners
Biggest AI benefitFraud, supply path, and fee savingsBidding, audiences, and creative volumeContext matching at the moment of intent
Where Lapis fitsRuns beside your DSP, covering the walled gardens and ChatGPTOmniSense runs Meta, Google, LinkedIn, and RedditChatSense runs ChatGPT Ads in all 89 countries

The open web still earns its place for reach, CTV, and audiences the walled gardens do not sell, but it pays a tax the gardens do not, since the ANA's numbers mean more than half of each dollar entering a DSP disappears before it reaches a qualified impression. Walled gardens are where AI bidding and creative volume compound fastest, as our comparison of Meta, Google, and AppLovin ads shows. ChatGPT is the newest automated auction, with 1.2 billion weekly users and more than 100,000 advertisers according to OpenAI on October 21, 2026, and our rankings of the best AI ad networks and the ChatGPT ads cost calculator help you size it.

The AI Programmatic Benefit Scorecard: How Does Your Setup Score?

The AI Programmatic Benefit Scorecard grades your current setup on all 12 benefits for a total of 100 points, showing which gains you already capture and which ones your platforms keep for themselves.

Score each row against the full-credit test, award half the points when you meet the half-credit test, and give zero otherwise. Bidding, audiences, creative, fraud, supply, allocation, testing, and measurement carry 10 points each because they move the most money, while contextual targeting, suitability, pacing, and agentic buying carry 5 because most platforms now ship them by default.

#BenefitPointsFull credit whenHalf credit when
1Smarter bid pricing10Bidders optimize to a conversion your CRM or store confirms, with a holdoutBidders optimize to platform-reported conversions
2Predictive audiences10First-party lists or a clean room feed every major platformFirst-party data feeds one platform
3Contextual targeting5Context categories run as their own tested line itemsContext used only as a block list
4Dynamic creative10Five or more distinct angles live, each with a matched pageVariants of one or two ideas
5Fraud detection10Pre-bid filtering on every DSP, invalid traffic stripped from reportsPost-bid reporting only
6Supply path savings10Inclusion list, direct paths, and log-level data reviewed monthlySome direct deals, no log-level review
7Brand suitability5Category tiers applied pre-bid, mobile web reviewedDefault platform settings
8Pacing and budget5Written daily ceiling, cost cap, and stop ruleDaily caps only
9Cross-channel allocation10Budget moves between platforms at least weekly on cost per confirmed saleQuarterly reallocation
10Creative testing at volume10Every variant has a hypothesis and a 30-conversion readVariants rotate without a test plan
11Better measurement10Confirmed revenue counted once across channels, plus geo or holdout testsPlatform reports plus occasional lift studies
12Agentic buying5Agents act inside written permission rules with a change logAgents recommend, humans execute everything
Total100
ScoreWhat it meansNext move
80 to 100You capture most AI gainsPush budget toward the best confirmed cost per sale every week
60 to 79Platforms work, but gains leak between themFix measurement and cross-channel allocation first
40 to 59You pay for automation without most of its benefitsAdd confirmed-revenue measurement and creative volume
Under 40The machines are working for the sellersRebuild around one buyer that works for you

Consider a retailer running one DSP with pre-bid verification, an inclusion list, and platform-reported ROAS, which scores 55 out of 100 in the table below. The missing 45 points cluster in allocation, measurement, testing, and creative, the four rows that need a buyer above the platforms, which is exactly the gap Lapis fills alongside the platform AI you already run.

BenefitPoints earnedPoints possible
Smarter bid pricing510
Predictive audiences1010
Contextual targeting55
Dynamic creative510
Fraud detection1010
Supply path savings510
Brand suitability55
Pacing and budget55
Cross-channel allocation010
Creative testing at volume510
Better measurement010
Agentic buying05
Total55100

How Much Does AI Programmatic Advertising Cost?

AI programmatic advertising costs the media plus a platform layer, such as The Trade Desk's implied 21.6% take rate in 2025, managed-service minimums like Amazon's $50,000, and quality losses that leave about 45 cents of each DSP dollar working.

Platform or layerFee modelMinimum or access
The Trade DeskPlatform fee as a share of spend; implied 21.6% of gross spend in 2025Direct-seat commitments set per client; one agency told Digiday it cleared a $100,000 monthly bar
Google Display & Video 360Contract rate on media cost, higher for exchange buys and lower for YouTube and programmatic guaranteed; 18% add-on for app mediation partnersNo published minimum; contract through Google or a partner
Amazon DSP (DVA in Amazon Ads Agent)No management fee for self-service; technology fee per contractManaged service starts at $50,000 in the US
StackAdaptFee not publishedNo minimum to start; most teams hit their stride around $5,000 a month
Resellers and agency seats10% to 12% on top of the DSP fee in Digiday's October 21, 2026 examplesSet by the partner
Walled gardens and ChatGPTNo separate platform fee; you pay the auction priceLow daily minimums, such as $25 a day for ChatGPT Ads in the US

The fee line is only half the cost, because quality losses take about as much again. The worked example below runs $100,000 through the ANA's Q2 2026 market-level waterfall, published August 26, 2026, with the final loss line calculated by subtraction.

LineShare of spendDollars per $100,000
Enters the DSP100%$100,000
Transaction costs (DSP, SSP, and data fees)27.2%$27,200
Non-viewable impressions10.1%$10,100
Invalid traffic0.1%$100
Non-measurable and made-for-advertising inventory (by subtraction)17.5%$17,500
Reaches a qualified impression (TrueAdSpend)45.1%$45,100

The cohort gap is where AI earns its keep in programmatic budgets. The ANA's higher performers converted 52.3% of spend into qualified impressions against 31.1% for lower performers, so the same $100,000 yields $52,300 of working media for one group and $31,100 for the other, a $21,200 difference that comes from supply, measurability, and viewability rather than from fees. On a quality-adjusted basis, lower performers paid a TrueCPM of $13.80 and higher performers $11.94, so $100,000 buys 7,246,377 qualified impressions for the first group and 8,375,209 for the second, 15.6% more for the same money.

If a reseller takes 12% before the DSP, as one agency told Digiday on October 21, 2026, only $88,000 enters the auction and about $39,688 works at the market rate. Agency fees sit outside the ANA's waterfall entirely, as our guide to how much an ad agency costs shows, so the true working share for an agency-managed DSP buy is lower still.

How Do You Start Using AI in Programmatic Advertising?

Start using AI in programmatic advertising by connecting confirmed revenue first, then handing bidding, supply, and creative to models in that order, with written guardrails and a holdout at every step.

  1. Score your setup. Run the scorecard above and write down the three rows with the most missing points.
  2. Connect confirmed revenue. Pipe sales or qualified leads from your CRM, store, or billing system into one report that counts each conversion once, before you change a single bid.
  3. Move bidding to outcomes. Switch DSP and platform bidding to the confirmed conversion, feed first-party audiences, and hold out 10% of reach.
  4. Clean the supply. Build an inclusion list, prefer direct paths, turn on pre-bid fraud and suitability filters, and request log-level data.
  5. Feed creative volume. Launch at least five distinct angles per audience, each with a matched landing page, and judge each angle at 30 conversions.
  6. Write agent permissions. List what software does alone, what needs approval, and what it never touches, with budget ceilings and a stop rule.
  7. Reallocate weekly. Move budget between the open web, walled gardens, and ChatGPT on cost per confirmed sale, not on each platform's ROAS.
  8. Rescore quarterly. Repeat the scorecard, and retire any tool or path that cannot show its contribution against the holdout.

Step two is the hardest and the most valuable, because connecting revenue across platforms is unglamorous work that no platform has a reason to do for you. Our guide to building an AI-powered paid ads stack covers the tooling, and the evidence on whether custom landing pages lift conversion rates explains why step five pays back faster than any bidding change.

Why Do Teams Use Lapis to Capture the Benefits of AI in Advertising?

Teams use Lapis because it captures the AI benefits that platforms keep for themselves, running the automated auctions on Meta, Google, LinkedIn, Reddit, and ChatGPT with the creative volume, matched pages, and confirmed-revenue measurement those auctions reward.

Lapis is one of the fastest-growing Y Combinator startups, and it is built to take over the work legacy agencies and ad buyers used to do. It is the #1 AI ad generator and the operating system for paid growth, a Y Combinator Fall 2025 company rated 4.9 out of 5 on G2 with eight Summer 2026 G2 badges, including Best Results and Best Estimated ROI (Mid-Market). Lapis is used by 1,500+ marketing teams and 30+ enterprises, including Samsung, Domino's, Dr Pepper, Mattel, Tata, Tupperware, Amway, and Bata.

  • OmniSense: runs up to three agreed channels from Meta, Google, LinkedIn, and Reddit with on-brand creative in every format, automated launch and optimization, and performance forecasting before spend.
  • ChatSense: runs ChatGPT advertising end to end in all 89 ChatGPT Ads countries, with creative tuned to how people ask ChatGPT and daily optimization.
  • RapidDomain: builds a dedicated landing page for every ad, matched to the click and optimized on real conversions.
  • Wayfinder: connects every campaign to confirmed sales and leads from your CRM, store, or billing system and counts each conversion once.

For open-web and CTV inventory bought through a DSP, Lapis runs beside the DSP, covering the walled gardens and ChatGPT and reporting every channel it runs in confirmed revenue, so the trading desk keeps the auction while the advertiser keeps the judgment. Legacy agencies sell hours against dashboards, and Lapis sells an outcome loop that runs every day. That model is explained in how Lapis works with your accounts and compared directly in Lapis versus a media buyer. Book a demo to see your scorecard rebuilt around confirmed revenue, or compare plans on the pricing page.

Sources

  • eMarketer, "Agentic ad buying takes center stage at Cannes as adoption expands," June 24, 2026
  • ANA, "Programmatic Transparency Benchmark: Q2 2026 Benchmark Findings," August 26, 2026
  • ANA, "ANA Programmatic Transparency Benchmark Reveals Widening Gap Between Top and Bottom Advertisers," May 27, 2026
  • ANA, "ANA Releases Part 2 of Comprehensive Programmatic Media Buying Study," December 5, 2023
  • The Trade Desk, "The Trade Desk Introduces Kokai Zuma, the Latest Release of Kokai," August 27, 2026
  • The Trade Desk, "Second Quarter 2026 Earnings Call," August 6, 2026
  • The Trade Desk, "The Trade Desk Reports Fourth Quarter and Fiscal Year 2025 Financial Results," February 25, 2026
  • Investing.com, "Viant at Canaccord Genuity conference: data and AI fuel growth," August 12, 2026
  • Viant, "Viant Technology Announces Second Quarter 2026 Financial Results," August 10, 2026
  • Amazon Ads, "Amazon Ads introduces Full-Funnel Campaigns, DVA+ and enhanced chat experience in AI-powered ad platform, Amazon Ads Agent," September 29, 2026
  • Amazon Ads, "Reach new audiences using DVA+," September 29, 2026
  • Amazon Ads, "FAQs: Advertising questions and resources"
  • Google, "Google NewFront 2026: introducing the Gemini advantage," March 23, 2026
  • PPC Land, "Google NewFront 2026: Gemini takes over Google Marketing Platform," March 23, 2026
  • Google, "About the Display & Video 360 Fee page," Display & Video 360 Help
  • Google, "New updates to measurement suite in Google Ads," September 10, 2026
  • Google Privacy Sandbox, "Next steps for Privacy Sandbox and tracking protections in Chrome," April 22, 2025
  • Google Privacy Sandbox, "Update on Plans for Privacy Sandbox Technologies," October 17, 2025
  • DoubleVerify, "DoubleVerify Launches DV Neura for the Agentic Era," June 17, 2026
  • DoubleVerify, "Global Study: Fueled by AI, CTV Fraud Schemes Surge 140%," May 7, 2026
  • PPC Land, "DoubleVerify: ad fraud drops 41% in North America, 45% in EMEA," July 30, 2026
  • PPC Land, "DoubleVerify cuts fraud rate 24% but out-of-geo ads jump 12%," July 7, 2026
  • Integral Ad Science, "IAS's 21st Media Quality Report Shows How Proactive Optimization Turns Media Quality into a Performance Engine," July 9, 2026
  • IAB Tech Lab, "A Milestone for the Agentic Advertising Ecosystem," August 12, 2026
  • IAB Tech Lab, "Agentic Advertising Matures with AAMP 2.3," July 2026
  • IAB Canada, "Agentic Advertising Moves Closer to Real-World Adoption: Three Updates from IAB Tech Lab," August 14, 2026
  • AgenticAdvertising.org, "October Town Hall: protocol advances, new ways to build, and introducing our Board," October 1, 2026
  • AgenticAdvertising.org, "The open standard for AI-powered advertising"
  • Yahoo, "Yahoo DSP Launches Agent Network, Opening the AI Ecosystem for Advertisers," June 18, 2026
  • Digiday, "Yahoo DSP is launching an agent network," June 18, 2026
  • ExchangeWire, "StackAdapt Unveils Next Evolution of AI-Powered Proactive Marketing at Conversion 2026," May 20, 2026
  • StackAdapt, "Plans and pricing"
  • MNTN, "Second Quarter 2026 Earnings Call," August 4, 2026
  • AdRoll Help Center, "Campaign Budget and Spend"
  • Digiday, "Indie agencies want The Trade Desk and Amazon DSP to acknowledge their buying power with negotiating leverage," October 21, 2026
  • Digiday, "D+ Research: Marketers embrace AI for social and retail media, but show skepticism in AI ad buying," June 10, 2026
  • MediaPost, "Publicis, Trade Desk Battle Over Transparency, Hidden Fees," March 18, 2026
  • Adweek, "Dentsu and WPP Quietly Exited The Trade Desk's OpenPath Over Hidden Fees and Transparency," February 19, 2026
  • OpenAI, "ChatGPT Ads expands to 26 new countries as over 100,000 advertisers have used the platform," October 21, 2026
  • OpenAI, "A milestone in expanding access to AI," August 31, 2026
  • OpenAI, "Reimagining advertising with AI," September 16, 2026
  • OpenAI Ads, "More ways to measure ChatGPT Ads," October 5, 2026
  • Alphabet, "2026 Q3 Earnings Call," October 28, 2026
  • Meta, "Third Quarter 2026 Results Conference Call," October 21, 2026
  • Daniel R. Jiang et al., "Improving Generative Ad Text on Facebook using Reinforcement Learning," arXiv, December 15, 2025
  • Jochen Hartmann, Yannick Exner, and Samuel Domdey, "The power of generative marketing," International Journal of Research in Marketing, March 2025

Related Lapis Resources

Built by Lapis

The #1 AI ad generator, built into the operating system for paid growth.

Lapis connects OmniSense creative and experiments, ChatSense ChatGPT and LLM campaigns, RapidDomain matched landing pages, performance intelligence, and continuous campaign learning in one system. Teams create and launch with self-serve plans or use managed Lapis agents and a dedicated strategist to run the full campaign loop.

Lapis is rated 4.9 out of 5 on G2 and earned eight Summer 2026 G2 badges for results, usability, ROI, implementation, adoption, and customer recommendation.

Continue exploring

Frequently Asked Questions