There is a familiar shape to most AI pitch decks in 2026: a chat window, a clever prompt, a demo that dazzles for ninety seconds. Predictive, a seed-stage venture firm run out of New York and Latin America, spends its days looking for the opposite. Its founder, Kevin Alvarez-Fung, would rather fund the company routing a freight truck through customs, or scoring a small-business loan, or catching a fraudulent lease application - the un-glamorous middle of the economy where the software is hard to build and even harder to copy.
The wager is simple to state and difficult to execute. In a market where nearly every startup can rent the same frontier model by lunchtime, Predictive argues that the model is not the moat. What lasts is proprietary data and product precision - the parts a competitor cannot summon with an API key.
What Predictive actually does
At its core, Predictive is a check writer. It backs pre-seed and seed companies, usually at the moment a founder has a strong product hypothesis but little proof. But the firm's own pitch to founders leans on what happens after the wire clears. Predictive positions itself as an operating partner: it helps scope the first version of a product so it fits into a customer's existing workflow, secures design partners, and then works to turn those early pilots into paid contracts.
That framing matters because it targets the exact place most early startups stall. A signed pilot feels like victory. In practice it is often a free trial that quietly expires. Predictive's platform is built to close that gap - matching portfolio companies to real buyers and staying involved until a pilot becomes revenue.
The mechanics of that help are worth spelling out, because they are unusual for a fund of this size. Predictive describes four repeatable services it runs for portfolio companies: securing design partners from its corporate network, scoping products so they slot into a customer's existing workflow rather than demanding a new one, converting pilot contracts into paid revenue, and identifying and engaging target customers at scale. None of that is glamorous, and none of it shows up in a headline valuation. But for a seed company, an introduction to a buyer who will actually sign is worth more than another slide of encouragement.
Figures as stated by Predictive.
The industries it hunts in
Predictive keeps a deliberately narrow lens. It concentrates on what it calls foundational industries - logistics, trade, financial services, healthcare and manufacturing - sectors that move physical goods and real money, and that have been comparatively slow to absorb application-layer AI. The theory is that these markets generate enormous, messy, proprietary datasets, and that a startup which becomes the system of record for one of them builds an advantage that compounds every day it operates.
Where Predictive concentrates - relative portfolio emphasis
Illustrative weighting drawn from disclosed portfolio companies, not a formal allocation.
The data moat, explained
"Data moat" gets thrown around loosely, so it is worth being concrete about what Predictive means. A model is rented; anyone can rent the same one. A data moat is earned. It is the proprietary information a company accumulates by being embedded in a workflow - the shipment histories, the underwriting outcomes, the machine-tolerance readings - that make its predictions sharper than a rival starting from scratch. The firm's name is the thesis: the companies it wants are the ones whose product gets more predictive the longer it runs.
This is also why Predictive is comfortable in sectors other AI investors find unfashionable. A restaurant-operations platform or a global-trade engine will never trend on social media, but each sits on a data stream no competitor can buy off the shelf.
There is a second-order effect the firm is clearly counting on. When a startup owns the data that trains its own models, its product improvement and its defensibility become the same motion. Every additional customer sharpens the predictions, which wins the next customer, which sharpens them again. In consumer AI, where users can switch tools on a whim, that flywheel is fragile. In freight or underwriting, where switching a system of record is painful and rare, it tends to hold.
The founder behind the fund
Alvarez-Fung did not arrive in venture from a spreadsheet. He was an early team member at Kensho, the machine-intelligence company later acquired by S&P Global, and at Digits, and he co-founded the payments startup Mural Pay. That operating history shows up in the firm's product-first language - the emphasis on scoping, workflows and first revenue reads like it was written by someone who has shipped software, not just funded it.
He describes himself as a third-culture person with Puerto Rican, Chinese and Boston roots, splitting time between New York City and Latin America. Away from deals, he has mentioned working on a debut novel, running the Hudson, and practicing jazz piano and salsa - the sort of detail that hints at why the firm's cultural center of gravity sits between two continents rather than inside one zip code.
The bench
The other half of Predictive's differentiation is people. The firm has assembled a network it describes as 75 founders and product leaders who, between them, have raised roughly $3.5 billion and backed more than 35 unicorns from the seed stage. Names attached to that bench include Jeff Seibert, a co-founder of Digits, and Zoe Barry, who founded ZappRx. For a founder deciding where to take a first check, the calculus is straightforward: capital is fungible, but access to operators who have already built and sold in your category is not. That bench is the asset Predictive cannot be easily out-bid on.
The portfolio
Predictive's track record is the clearest argument for its approach. Earlier bets include several companies that have gone on to command billion-dollar valuations, and a Fund II lineup aimed squarely at the foundational-industries thesis.
- Esusu - financial services building credit access for underserved Americans
- Owner.com - AI-driven marketing and operations for independent restaurants
- Pomelo - fintech infrastructure and cross-border money movement
- Huspy - a homeownership and mortgage platform
- Machina Labs - robotic, AI-guided sheet-metal manufacturing
- Wander - technology-enabled luxury travel
- Nauta · Oway · Dirac · Spur - Fund II bets across trade, logistics, manufacturing and software QA
Two continents, one thesis
Geography is part of the strategy. Predictive invests across both the United States and Latin America, and keeps a foot in Miami alongside New York. The firm runs a Latin America-focused publication, latamvc.co, and its corporate network spans companies in both regions. The bet embedded in that footprint is that category-defining companies increasingly emerge outside Silicon Valley, and that a fund fluent in both markets can source and support founders others overlook.
How it makes money - and where it fits
The business model is conventional venture economics: Predictive takes equity in the companies it backs and, like most funds, earns through management fees on committed capital and carried interest on gains when those companies exit. What is less conventional is the surface area of help it offers around the check - the design-partner network and operator bench are the differentiators it sells.
In the crowded field of seed AI investors, that positions Predictive against both generalist operator-led funds and regional specialists. Firms competing for the same founders range from application-layer-focused shops to Latin America stalwarts. Predictive's pitch is the intersection the others rarely occupy at once: deep operating help, a data-moat filter, and genuine reach across two markets.
The risk in the strategy is the mirror image of its appeal. Foundational industries are slow. Enterprise sales cycles in logistics and healthcare are measured in quarters, not weeks, and the data advantages Predictive prizes take years to compound into something a competitor cannot catch. A fund built on that patience is making a bet not just on which companies win, but on its own willingness to wait. For founders who have watched flashier peers raise faster and flame out, that patience may be exactly the point.
Whether the thesis pays off will be settled the way every venture thesis is - slowly, in a decade of outcomes. But the logic is coherent, and refreshingly unfashionable. As the market crowds into the flashiest corners of AI, Predictive is quietly buying tickets to the parts of the economy that still run on trucks, invoices and factory floors, betting that is where durable data - and durable companies - are still up for grabs.