Koah’s ad network for AI

John Ebbert

Nic Baird of Koah

As artificial intelligence moves beyond general-purpose chatbots and into consumer apps, developers and publishers of AI apps face a more immediate version of an old internet problem: how to pay for the product.

Even as token costs reportedly plummet, AI features still cost far more to operate than traditional web and app experiences. In February, Koah announced a $20.5 million Series A and describes itself as “the ad network built for AI”: a marketplace connecting advertisers with AI apps through native, real-time intent matching. The company believes that makes advertising a necessity much earlier in the life of an AI product.

In a conversation with tipsheet yesterday, Co-Founder and CEO Nic Baird discusses the company’s growth, why conversational ads should be treated as a consideration channel rather than a new version of Search, the measurement challenge created when AI acts as a thought partner and how dynamically generated ad formats could support a wider AI app ecosystem.

Topics covered include:

  • Koah’s growth and marketplace momentum
  • Koah’s ad network model for AI apps and advertisers
  • Why consumer AI monetization is arriving earlier
  • The opportunity beyond ChatGPT and Claude
  • What ChatGPT Ads is teaching the market
  • Which advertiser categories are working today
  • AI as a consideration and thought partner channel
  • Attribution for conversational media
  • Why inference costs constrain AI publishers
  • Dynamic formats and generated ad copy
  • Koah’s supply-side strategy
  • What conversational advertising could look like in three to five years

Read the full interview: Koah CEO Nic Baird on building the ad network for AI (Sept. 11) – tipsheet

LLMs & CHATBOTS

Amazon DSP opens managed path to ChatGPT ads

Yesterday, Amazon Ads and OpenAI announced an integration that lets select U.S. advertisers test conversational ads in ChatGPT. Delta Vacations is among the first brands participating.

Chris Conetta, Director of Omnichannel Supply at Amazon DSP, said in a statement:

“Conversational ads represent the fastest growing engagement opportunity for brands to reach new and existing audiences. The AI-driven ad experience is not just a vision of what’s to come, but an opportunity for advertisers to meet relevant users where they spend time today. With access to ChatGPT Ads, advertisers can now extend their campaigns further into where their customers are actively spending time in conversational chat platforms.”

Read: Amazon Ads Announces New Integration to Support Advertising in ChatGPT (Sept. 10) – Amazon Ads

An Amazon spokesperson clarified the announcement’s details with tipsheet:

  • The pilot is U.S.-only to start.
  • ChatGPT conversational ad inventory is available through Amazon DSP, but campaigns are initially set up and managed by the Amazon Ads team through managed service.
  • Amazon Ads says it plans to open self-service buying through Amazon DSP; meanwhile, its team helps optimize setup, including bidding strategies, context hints and max bids.
  • OpenAI’s ads system controls all ad delivery decisions.
  • Amazon Ads is integrating directly with OpenAI’s ChatGPT ad platform, offering CPM and CPC bidding models.
  • Ads are text and image units and appear beneath ChatGPT responses, clearly labeled as sponsored and visually separated from organic answers, for logged-in adult users on ChatGPT’s Free and Go subscription tiers.
  • Product feed ads can auto-generate creative from product names, images and attributes.
  • Advertisers receive aggregated performance reporting including impressions, clicks, CTR, CPC and CPM.

From tipsheet: The first leg of the long-awaited integration between Amazon Ads and ChatGPT is here. The buying interface may be shared, but the intelligence determining each ad’s delivery remains with OpenAI.

MEASUREMENT

Google turns measurement into a decision loop

When causality and attribution collide…

Google has rolled out a series of measurement updates designed to move measurement from reporting what happened toward influencing what marketers do next.

On Thursday, the company said the biggest change comes to Meridian, Google’s open-source marketing mix model. Google is adding agentic capabilities that can audit data quality, resolve errors and guide marketers through building models in real time. Meridian can also now incorporate brand signals such as Branded Google Query Volume (GQV) to measure how upper-funnel advertising contributes to future sales.

On the data front, Google announced:

In addition, Google is making Meridian GeoX generally available globally. GeoX lets marketers run causal geo experiments across any advertising platform, then feed incrementality results into Meridian to improve the MMM. The system increasingly connects three measurement approaches: attribution, incrementality experiments and marketing mix modeling.

Read: Google’s Upcoming Advertising Measurement Tools (Sept. 10) – Google Ads & Commerce blog

From tipsheet: Google is building a measurement feedback loop that extends beyond Google media. First-party data feeds the system, attribution observes performance, GeoX tests causality and Meridian models the larger allocation decision. Now agents are entering the loop to help construct the models themselves. Google doesn’t have to own every advertising dollar to influence how the next one gets spent.

Q&A with Nipoon Malhotra

Following the announcement, Google VP/GM of Measurement Nipoon Malhotra responded to a selection of questions from tipsheet via email.

tipsheet: How autonomous will Meridian’s new agentic capabilities be? For example, will the agent only identify problems and recommend changes or can it actually make changes to data inputs or model construction?

Nipoon Malhotra: We’re giving Meridian agentic assistance. You can think of these skills as your AI companion. They can audit data, make recommendations to resolve errors, and provide modeling instructions, which the agent can act on with your permission.

The goal is to speed up time-consuming processes of building and refining models and give our customers greater access to Meridian expertise.

These skills are designed to work with your preferred set up — meaning you can deploy them to work with any AI agent you’re using.

tipsheet: Given the agentic capabilities, what requires human approval?

Nipoon Malhotra: We built this tool to empower marketing teams that want to build their own systems, keep full control, and save time by letting AI handle the tedious manual work. This frees up marketers to focus on the bigger picture, helping them make faster, better decisions that drive business growth.

The marketer remains in the driver’s seat. Anything the agent acts on requires their permission.

tipsheet: As Meridian brings together MMM, incrementality and signals such as Branded Google Query Volume, what does Google see as attribution’s role in that measurement system?

Nipoon Malhotra: To get a complete view of performance, advertisers need to look at a full portfolio of signals. While tools like Marketing Mix Modeling (MMM) give you the big picture view of your investments, other metrics like attribution and incrementality play different, critical roles: Signals like traditional attribution and our innovation in areas such as Qualified Future Conversions (QFCs), act as leading indicators; providing the real-time insights.

PROTOCOLS

PMG pilots sell-side decisioning

Can an agency’s buying intelligence run inside an exchange?

Agency PMG is activating Bedrock Platform‘s containerized DSP inside Index Cloud through its Alli Buyer Cloud operating system, allowing buying logic to execute at the exchange rather than from a remote bidder.

Read: PMG Leverages Agentic Buying of Streaming TV with Bedrock’s Containerized DSP on Index Cloud (Sept. 10) – Index Exchange

Index Exchange VP of Product Michael Richardson explained the architecture in an email follow-up with tipsheet:

“PMG gains the ability to apply its own bidding logic — via Bedrock — against the full stream of supply rather than a sample.

A conventional integration connects in from outside the exchange, and that connection is metered, so the buyer evaluates a fraction of what’s available. Because Bedrock’s DSP runs as a container inside Index Cloud, co-located with the auction, PMG decisions at the origin of the impression. That’s the sell-side decisioning model, and this is the first time an independent agency has run it from its own operating platform.

Subjective’s content and audience signals feed the decision at bid time, so the judgment happens against the actual programming across more than 3,000 media owners, without PMG maintaining any external bidder infrastructure.”

The pilot begins with premium CTV inventory. Mobile app and display are planned next.

Bedrock launched its containerized DSP on Index Cloud in April.

From tipsheet: This reflects the architectural direction envisioned by IAB Tech Lab’s ARTF: bringing buying intelligence closer to where each impression originates. The infrastructure hosting those impressions becomes part of the decisioning process.

Related:

  • What Is the Agentic Real-Time Framework (ARTF)? – Index Exchange
  • ‘A trader won’t need to leave our platform’: PMG builds its own CTV buying platform (December 2025) – Digiday

CONNECTED TV

AWS turns video pipeline into yield layer

AWS has introduced four AWS Elemental (media workflow) capabilities for live video, including AI-generated contextual metadata and a new MediaTailor yield optimization product that automatically backfills unfilled streaming TV ad inventory with Amazon Publisher Services (APS) demand.

The contextual-metadata tool analyzes live video in real time for objects, actions, brands, on-screen text and sentiment. AWS says the resulting signals can support content discovery, brand suitability and contextual ad decisioning.

New commercial opportunity

The more commercially consequential launch is Yield Optimization: streaming TV publishers using MediaTailor can set price floors and category rules while AWS fills eligible unsold slots with APS demand. AWS says it requires no additional setup cost.

Read: Personalize live viewing, monetize every impression, and reach every device with AWS Elemental at IBC 2026 (Sept. 10) – Amazon Web Services blog

From tipsheet: Amazon is becoming involved in more of the publisher’s ad serving workflow: AWS can generate contextual signals from live video, insert ads into the stream through MediaTailor and help fill unsold slots with APS demand. The video pipeline is becoming part of the ad marketplace.

Also: AWS’s near-term competitive target is not every streaming ad server. It is the separate plumbing publishers use to route unsold impressions to fallback demand.

PLATFORMS

Case study: Changing the case

Yesterday, Criteo Chief Customer Officer Ed Dinichert wrote on LinkedIn:

Criteo, Flywheel and Amazon DSP
Times change: Amazon and Criteo were competitors, not collaborators, a decade ago. Now Omnicom Commerce Group’s ecommerce agency Flywheel is activating Criteo Commerce Graph audiences through Amazon DSP curation.

LLMs & CHATBOTS

Developments

  • Introducing DeepSeek-V4.1-Flash (Sept. 10) – DeepSeek on X
  • How Emmanuel Macron brought France’s AI darling to the world stage (Sept. 9) – Politico
  • Want to Succeed in an AI World? Study Anthropology (Sept. 10) – The Wall Street Journal

MORE

  • Days of arb: Publishers Are Spending Hundreds of Millions a Month Buying Traffic (Sept. 10) – Adweek
  • Open web: Decision Fabric: The Intelligence Layer of the Open Internet (Sept. 10) – Pubmatic’s blog
  • More spending: IAB Raises 2026 U.S. Ad Spend Forecast to +12.3% YoY Growth, as Strong First Half Lifts Expectations (Sept. 10) – IAB
  • More partners: Adform is One of OpenAI’s Technology Platform Partners as ChatGPT Ads Arrive in Europe (Sept. 10) – Adform
  • New app: I Tested Muse, Meta’s Agentic Commerce App (Sept. 10) – Debra Aho Williamson on The AI Ad Economy Substack