A Brooklyn-based startup is betting that the real bottleneck in AI-driven marketing is not the model - it is the fragmented, unusable state of the data those models run on.

Minerva, an AI platform for consumer marketing teams, launched publicly on June 9, 2026, raising $20 million in seed funding and announcing a collaboration with OpenAI that puts frontier models inside the core of its product. The investors include The General Partnership, 8VC, Lingotto Innovation, Topology Ventures, and NBA Investments. The company is based in Brooklyn, New York.

The announcement positions Minerva squarely inside the fastest-moving segment of marketing technology. Sales, marketing, and CRM startups raised approximately $3.7 billion globally in early 2026, with agentic AI companies capturing a disproportionately large portion of that capital. Minerva's raise lands in a period when every significant platform - from The Trade Desk to Adobe to Amazon - has introduced or expanded its own AI agent capabilities. The question Minerva is answering is different from those incumbents: rather than automating the execution of campaigns, it is targeting the structural problem that sits underneath them.

The data problem Minerva is built to solve

Brands accumulate enormous volumes of first-party customer data. Transaction histories, email engagement records, loyalty programme interactions, CRM fields - most large consumer organisations have years of this material. The problem, according to Minerva, is that it almost never sits in a single, coherent place. It is fragmented across systems, inconsistently formatted, and difficult to enrich with external context. That fragmentation limits its utility, both for human analysts and for the AI models that depend on well-structured input.

This is not a new problem. The industry has been building infrastructure around it for years. Google launched a Data Manager API in December 2025 specifically to centralise first-party data uploads across Google Ads, Google Analytics, and Display and Video 360, reducing the engineering complexity of maintaining multiple separate integrations. StackAdapt released a Snowflake native app for first-party data activation aimed at the same underlying challenge. Publicis's $2.5 billion acquisition of LiveRamp in 2026 was explicitly framed around building a data layer for autonomous AI agents - with Publicis CEO Arthur Sadoun describing data co-creation as the mechanism by which high-value datasets are connected across partners in a secure environment.

Minerva's approach is narrower and faster. According to the company, a brand can onboard and begin running end-to-end marketing workflows with AI agents within 24 hours. That claim covers four capabilities the platform delivers after onboarding: unifying and unlocking first-party customer data through the Agentic Data Engineer; enriching that data with Minerva's proprietary identity graph and more than 1,000 consumer attributes; creating, analysing, and optimising campaigns at scale; and measuring performance with detailed reporting on campaign results.

Two AI agents built on GPT-5.5

The collaboration with OpenAI produced two distinct workstreams, each built using frontier models including GPT-5.5. These are not assistants or copilots in the conventional sense - they are automated agents that carry out specific, defined tasks without requiring a practitioner with machine learning expertise to supervise them.

The Agentic Data Engineer addresses what has historically been one of the most time-consuming phases of any data project: understanding an unfamiliar dataset well enough to work with it. According to Minerva, this agent profiles and understands the structure of a customer's first-party data, writes transformation SQL to standardise and restructure it, and validates the output. The company says this collapses what would typically take weeks of human data engineering work into hours.

The Agentic Data Scientist tackles the next stage: generating predictive models from that prepared data. A marketer with no machine learning experience can submit a natural language prompt - the example Minerva gives is "find users likely to book a luxury property in the next 30 days" - and the agent generates, validates, and deploys the predictive model on demand. The practical implication is that audience modelling, which has historically required a data scientist and a substantial lead time, becomes accessible to marketing teams without specialist technical staff.

This architecture connects directly to a structural argument Minerva makes about where the constraint in AI-powered marketing actually sits. As AI agents become more capable, the limiting factor is not the quality of the model - it is the quality and structure of the context those models can act on. The OpenAI collaboration is consistent with this framing: using GPT-5.5's capabilities for reasoning and code generation while directing that power at a specific, domain-relevant infrastructure problem.

The founding team and its background

Minerva was founded by Jackson Engles, Daniel Saedi, and Matthew Joseph. The three met at the University of California, Berkeley, before beginning careers in finance - Engles at Lazard, Saedi at Bridgewater, and Joseph at Citadel. The company grew out of Saedi and Joseph's work using alternative data to trade financial markets, where both the commercial value of consumer data and the difficulty of turning fragmented datasets into reliable signals became apparent firsthand.

That financial markets background is relevant context. Quantitative finance has spent decades building infrastructure to extract tradeable signals from messy, inconsistent external data sources - satellite imagery, credit card transaction feeds, geolocation patterns. The underlying technical challenge is similar to what Minerva is attempting for marketing: taking data that exists in inconsistent, noisy form and transforming it into structured context that a model can reason over reliably. The founding team's experience on that problem in one domain shapes their approach to solving it in another.

"Marketing teams are under pressure to deliver better outcomes with more complexity, more channels and more data than ever before," said Jackson Engles, co-founder and CEO of Minerva. "Minerva gives marketers the context and infrastructure to understand their customers more deeply and act on those insights faster. Our goal is to give repetitive, operational work to AI so our customers can spend more time on the work that requires real human judgment."

Early results and initial customer base

The company has signed three dozen customers ahead of its public launch. The roster includes the NBA, Juicebox, Luxury Presence, Trust and Will, and Wander. The NBA relationship is described as working to identify opportunities to help teams deepen fan engagement - a use case that involves predicting future behavioural patterns in an existing customer base rather than purely prospecting for new ones.

In early deployments, Minerva reports two headline performance figures. Brands using the platform improved paid media ROAS by 3.4x and direct mail MQL rates by 2.5x, both achieved by rebuilding how customer data is used to acquire customers. These figures are attributed to Minerva's own account of its early deployments.

A 3.4x ROAS improvement is a large number. For context, a survey published by Affinity Solutions in May 2026 found that agency marketers expecting gains from real-time purchase signal optimisation anticipated ROAS improvements of between 16% and 30% - a far more modest range than what Minerva is claiming from first-party data restructuring alone. The comparison is not entirely direct, since the mechanisms differ, but it illustrates why the claim will face scrutiny. Results from early deployments with a small number of customers do not necessarily generalise to a broader, more diverse customer base.

What investors are saying about the context layer

Phin Barnes, co-founder and Managing Partner at The General Partnership, articulated the investment thesis in specific terms. "AI agents are context-hungry, and whoever structures the right context for a domain wins that domain," Barnes said. "Minerva is building the context layer for marketing. Most brands have valuable first-party data, but it's fragmented and hard to use. Minerva turns that data into something AI can actually reason over, so then marketers can act on it."

The "context layer" framing is worth examining. It implies that the competitive moat Minerva is building is not the AI model itself - which is licensed from OpenAI and therefore available to competitors - but the proprietary infrastructure for structuring and enriching marketing data into a form that models can use effectively. The identity graph with more than 1,000 attributes and the Agentic Data Engineer's profiling and transformation capabilities are the components that would be difficult to replicate quickly.

The General Partnership has co-invested alongside 8VC, the firm co-founded by Joe Lonsdale. Lingotto Innovation and Topology Ventures complete the syndicate alongside NBA Investments, the latter of which connects to the NBA's direct customer relationship with the platform.

The agentic AI context

Minerva's launch arrives as the ad tech industry has been bracing for AI agents in a sustained way. The Trade Desk launched Koa Agents in April 2026 as its first in-platform AI agent, with Stagwell as pilot partner. Adobe introduced AI agents for digital marketing and ad strategy automation the same month. Amazon has run Ads Agent - which lets advertisers instruct the system with natural language commands to pause campaigns below a given ROAS threshold - since November 2025.

What distinguishes Minerva is the targeting of the data preparation layer rather than the campaign execution layer. As PPC Land has documented through multiple surveys, 79% of marketing professionals report that managing AI and automation is already one of the fastest-growing parts of their daily work, and 81% name AI tools as the most essential item in the modern marketer's toolkit - yet a structural gap persists between AI tool adoption and the audience understanding that should underpin it. Minerva's product is, in essence, an argument that this gap exists because the data feeding those tools is broken.

The OpenAI dimension of the announcement also carries meaning for the broader market. OpenAI opened its ChatGPT Ads Manager to all US businesses in May 2026, introducing CPC bidding alongside the existing CPM model, as it builds out its advertising infrastructure. ChatGPT advertising crossed $100 million in annualised revenue within six weeks of the pilot's launch, expanding to over 600 advertisers. A startup that holds an active collaboration with OpenAI - using its frontier models as the engine for autonomous data and science agents - occupies an interesting position as that relationship between OpenAI and the marketing technology industry continues to develop.

The verticals and expansion plan

Minerva's current customer base is concentrated in sports, hospitality, and financial services - verticals characterised by rich but fragmented customer data and a strong commercial need to predict future behaviour. The NBA is a natural fit: a league trying to use fan interaction data to predict which fans are most likely to convert on merchandise, ticket upgrades, or streaming subscriptions. Luxury hospitality brands face similar challenges with stay histories, preference data, and booking patterns spread across multiple systems.

The $20 million will fund expansion of Minerva's engineering, research, and go-to-market teams, build out the self-serve platform, and extend the company's footprint beyond its initial three verticals into broader consumer categories. A self-serve component matters for a company at this stage: it suggests the platform is designed to be operated by marketing teams directly, without requiring a managed service layer or extended implementation cycles.

The platform's positioning for CMOs and marketing teams specifically - not data engineers or data scientists - reflects the product assumption that the operational complexity should be absorbed by the agents themselves. The Agentic Data Engineer and Agentic Data Scientist are named that way deliberately: they are presented as performing the functions of specialist roles without requiring those specialists to be on staff.

Whether that claim holds across the diverse range of data environments that a broader customer base will represent remains to be tested. The early deployments were concentrated in a small number of accounts where Minerva could work closely with customers to validate outputs. Scale introduces data quality problems, edge cases, and integration challenges that are difficult to anticipate from a narrow initial sample.

Timeline

  • Before 2024 - Daniel Saedi and Matthew Joseph work on alternative data models for financial markets trading, observing the structural difficulty of turning fragmented consumer datasets into reliable signals.
  • 2024 (founding) - Jackson Engles, Daniel Saedi, and Matthew Joseph found Minerva in Brooklyn, New York, drawing on their finance backgrounds at Lazard, Bridgewater, and Citadel respectively.
  • October 1, 2025 - LiveRamp launches agentic orchestration, giving AI agents governed access to identity resolution, segmentation, and measurement tools; PPC Land coverage.
  • November 11, 2025 - Amazon announces Ads Agent at unBoxed, enabling natural language campaign management commands including ROAS-based pause instructions; PPC Land coverage.
  • December 9, 2025 - Google launches the Data Manager API, consolidating first-party data uploads across Google Ads, Google Analytics, and Display and Video 360; PPC Land coverage.
  • February 9, 2026 - OpenAI begins testing advertising in ChatGPT with a minimum advertiser commitment of $200,000 and a $60 CPM rate.
  • March 26, 2026 - ChatGPT advertising crosses $100 million in annualised revenue within six weeks of launch; PPC Land coverage.
  • April 21-22, 2026 - Adobe introduces AI agents for digital marketing automation; The Trade Desk launches Koa Agents with Stagwell as pilot partner; PPC Land coverage.
  • May 5, 2026 - OpenAI opens ChatGPT Ads Manager self-serve beta to all US advertisers, introducing CPC bidding; PPC Land coverage.
  • May 8, 2026 - Crunchbase data shows sales, marketing, and CRM startups raised $3.7 billion globally in early 2026, with agentic AI companies capturing a disproportionate share; PPC Land coverage.
  • May 20, 2026 - Publicis announces its $2.5 billion acquisition of LiveRamp, framed as building the data layer for autonomous AI agents; PPC Land coverage.
  • June 9, 2026 - Minerva launches publicly with $20 million in funding from The General Partnership, 8VC, Lingotto Innovation, Topology Ventures, and NBA Investments, alongside an OpenAI collaboration using GPT-5.5 agents.

Summary

Who: Minerva, a Brooklyn-based AI platform company founded by Jackson Engles, Daniel Saedi, and Matthew Joseph, backed by The General Partnership, 8VC, Lingotto Innovation, Topology Ventures, and NBA Investments.

What: Public launch of an AI platform for consumer marketing teams, raising $20 million in seed funding and announcing a collaboration with OpenAI. The platform uses two AI agents built on GPT-5.5 - an Agentic Data Engineer and an Agentic Data Scientist - to unify first-party data, enrich it with a proprietary identity graph of more than 1,000 attributes, and run end-to-end marketing workflows within 24 hours of onboarding.

When: The launch was announced on June 9, 2026. The company was founded in 2024.

Where: Minerva is headquartered in Brooklyn, New York. Its initial customer base spans sports, hospitality, and financial services, with clients including the NBA, Luxury Presence, and Wander.

Why: The company is addressing a structural problem in marketing: brands accumulate large volumes of first-party customer data, but its value is constrained by fragmentation and the absence of external consumer context. Minerva's argument is that the limiting factor for AI-powered marketing is not the quality of the model but the quality and structure of the data those models act on. The $20 million will fund engineering and research expansion, self-serve platform development, and entry into consumer categories beyond the initial three verticals.