Forget the List of Vendors. Just Describe What You Need.

Brad Schneider co-founded Nomad Data roughly five years ago on a simple observation: every existing solution to “where do I find the right data” was some version of a list. Hundreds or thousands of vendors, categorized loosely, described in a paragraph each — useless for a buyer trying to determine whether a specific terabyte dataset solves their specific problem, and stale within months of being compiled.

Nomad’s answer was to abandon categories and lists entirely and rebuild data search as a question engine. A customer describes what they need in plain English — Brad’s example: in the days before the LA fires, insurers repeatedly asked for historic fire risk data going back ten to fifteen years to price their catastrophe models. Nomad’s AI scans across roughly 4,000 data vendors (growing by about 100 a month), guesses which ones likely have relevant data, and then actually asks them. The vendors respond with specifics — geographies covered, years of history, metrics tracked — and that response becomes training data. The system knows more after every query than it did before.

In Episode 127 of InsurTechTalk, Brad and I covered how that search engine actually works, how a churn crisis nearly killed the company in its first eighteen months, and how the same document-reading engine built for data procurement became a claims automation tool almost by accident.

About Brad Schneider

Brad Schneider is the CEO and co-founder of Nomad Data, a New York-based software company that helps organizations find and buy data, and increasingly automates document-heavy workflows using AI. Nomad does not host or sell data itself — it operates as a search and matching layer across roughly 4,000 third-party data vendors, plus a procurement platform and an AI document engine called Document Chat. Insurance clients now span data sourcing teams, underwriting, claims, and risk.

A Search Engine, Not a Marketplace

The distinction matters and Brad was precise about it: Nomad is closer to Google than to a data marketplace.

• Nomad hosts no data itself — it builds and maintains an internal profile of what each vendor has, what it could be used for, and non-public information about the vendor’s business
• There is no public list of who sells what on Nomad’s platform — a data seller, including an insurer selling anonymized claims data, often does not want that advertised
• A user’s plain-English query — “California forest fire data going back 10-15 years” — gets matched by AI against vendor profiles, and Nomad then actually contacts the candidate vendors to confirm
• The vendor’s response — which regions, which years, which metrics — becomes new training data, regardless of whether the initial guess was right

That last step is the actual differentiator. Even when the AI-driven match is wrong, Nomad tolerates it well: if a request goes to four vendors and only one says yes, the buyer only ever sees the yes. The wrong guesses cost nothing and still teach the system something. Brad calls it a crowdsourcing engine for data vendor knowledge — a living index that improves continuously, unlike a static vendor directory that goes stale within a year.

What the Metadata Reveals

One of the more interesting threads in the conversation was what Nomad learns simply from who is asking and why.

• Nomad tracks not just what data exists, but who is searching for it and for what stated purpose
• When a new type of request appears — something never seen before — and then reappears repeatedly within days, it’s usually a signal: a policy change, a news event, a shift in the technology landscape
• That pattern lets Nomad recruit new vendors proactively and anticipate a wave of similar requests before they arrive

This is, in effect, a live index of where an entire industry’s attention is turning, built as an incidental byproduct of the core search product.

Document Chat: From Data Procurement Tool to Claims Engine

Nomad’s expansion into claims automation happened almost as a side effect of solving its own customers’ procurement pain.

Data buyers accumulate enormous volumes of vendor contracts — compliance reviews, legal sign-off, clause extraction, renewal tracking. Nomad built a feature called Document Chat to handle it: an engine that can read roughly a quarter of a million pages a minute, across a single large document or thousands of separate files, and either answer questions or extract structured data — including pulling a full table out of a PDF or reading 20,000 files to produce a CSV.

When Nomad showed this to its existing insurance clients — originally just data buyers — the reaction reframed the whole business. Insurers had enormous internal document repositories and manual processes around them, and wanted to know if the same engine could help.

What followed, over roughly the past year:

• Litigation document review — a 10,000-page filing that would otherwise require page-by-page attorney review at roughly $1,000 an hour gets read by the engine in about 60 seconds, producing a targeted summary or specific extracts
• Demand letter triage — automatically identifying which demand letters carry urgent or high-risk content requiring faster handling
• Claims coverage determination — quickly surfacing whether something is covered, why, and under which policy
• Book-of-business audits — a reinsurer evaluating a block of business can have the engine read 50,000 to 100,000 policies in minutes, producing a structured dataset for downstream analysis, rather than manual sampling

Brad’s framing: this pulled Nomad deeper into insurance-specific automation, where the margin expansion opportunity is real and, in his words, “quite dramatic.”

Vectors Help. The Last Step Is Still Asking.

I pressed him on the underlying technology — whether this runs on vector search, and how Nomad handles data categorization at a scale Google’s own early categorization efforts (around 200 categories at launch) never attempted.

Vectors help narrow the candidate set, but Brad was clear that the decisive step is still the direct outbound question to the vendor. Because the system tolerates being wrong — a rejected guess costs nothing and still generates a useful “no” — Nomad can afford to cast a wide net and let the vendor responses do the real categorization work. It is inherently more current than any static taxonomy, because it updates with every query rather than on a periodic refresh cycle.

What Insurance Buyers Actually Ask For

The buyer profile has broadened over the past year, and the trajectory says something about how insurers are approaching both data and AI:

• Originally: chief data officers, data sourcing and enrichment teams
• Increasingly: chief underwriting officers, chief risk officers, claims technology leaders

Brad’s observation connecting data and AI as businesses is worth keeping: both are inherently amorphous technologies whose value only becomes clear once matched to a specific use case. Nomad’s actual sales motion in both areas is the same — sit with a client, understand where their real operational pain is (a slow claims process, a manual onboarding step, a document bottleneck), and deliver an automated solution rather than “selling technology” in the abstract.

The Churn Crisis That Nearly Ended the Company

Asked for a failure, Brad gave one of the more candid startup stories I’ve heard on the show.

In year one, Nomad was hemorrhaging customers — effectively 100% churn over eighteen months. The root cause was targeting too broadly. Nearly everyone in a company needs data for something, but very few people have a recurring need for it. Customers would use the platform once, find exactly what they needed, thank Nomad, and cancel.

The fix had two parts:

• Rebuild the product to be sticky — this is what birthed the procurement platform, giving customers reasons to return beyond the initial search
• Retarget entirely — instead of anyone who might need data once, focus on teams whose job is finding data on a recurring, ongoing basis: data sourcing and acquisition teams at insurers, consulting firms, and investment firms, who have a live need again next month and next quarter

That single retargeting decision, more than any product change, is what removed the churn and let the business grow consistently.

Key Takeaways

• Static vendor lists go stale within months; Nomad’s model treats every search as a training signal that keeps the index continuously current
• The decisive step in matching buyer to vendor is still a direct question to the vendor — AI narrows the field, but confirmation comes from a human response
• What buyers search for, and why, is itself valuable metadata — spikes in a specific query type are an early signal of a shifting industry trend
• A document engine built to solve procurement’s own paperwork problem became a claims and litigation automation product once insurers saw it in action
• Reading a 10,000-page litigation file in 60 seconds replaces work that otherwise costs roughly $1,000 an hour in attorney time
• Targeting broadly (“everyone needs data sometimes”) produces catastrophic churn; targeting recurring need is what makes a data business viable
• Data and AI are more alike as businesses than most people assume — both require matching an amorphous capability to a specific, painful use case to create real value