Alternative Data for Hedge Funds:
What It Is, Main Categories,
and How Investors Use It

What is alternative data?

Alternative data is information used in investment research that comes from outside the conventional sources of market prices, company financial statements and sell-side estimates. The broad categories are card transactions, foot traffic, website visits, app downloads and scraped product listings, usually aggregated and scrubbed of personal data (PII). These data are organized and resold to hedge funds by vendors or sometimes by companies that produce the data as “exhaust” or byproduct of normal operations.

Another source is nontraditional analysis of familiar documents, such as scoring sentiment in earnings call transcripts, or reading the changes in a company’s SEC filings from one quarter to the next. The category is defined by where the information comes from and how it is processed.

For a hedge fund the purpose is fairly narrow: they are looking for an edge. Alternative data should tell the investment team something about a company that its existing process would otherwise miss or notice too late. We made the case for why public filings are an underused example of this in The News Gap, and the same standard applies to every type of data in this guide.

A dataset earns its place in your stack when it tells you something new and valuable, something your process would otherwise miss

How popular is it and how many investors use alternative data?

Alternative data is now ubiquitous among sophisticated investors. Lowenstein Sandler’s 2025 Alternative Data Report surveyed 107 professionals at hedge funds, PE and VC firms in late 2025 and found that 90 percent of respondents currently use alternative data, up from 67 percent a year earlier and 62 percent in 2023. More than two-thirds reported alternative data budgets above $1 million, and only 4 percent said they do not expect to use it in the future.

Spending is keeping pace with adoption. Neudata, which tracks the market, estimates that spending on alternative data rose 34 percent in 2024, well above its 21 percent average annual growth since 2020, and that the market for investment managers could reach nearly $40 billion by 2030. In Neudata’s 2024 buyer survey, the average firm subscribed to 20 datasets and spent about $1.6 million a year, while the largest firms averaged 43 datasets. When managers use more than one dataset they refer to it as their “data mosaic” since each piece answers a different set of questions about their companies.

Keep in mind, surveys such as this one draw from a few dozen to a few hundred respondents, and firms that already buy data are the likeliest to answer them. So, there may be some selection bias. The same Lowenstein report also found that use of some sources, consumer transactions among them, actually declined as firms were more selective, perhaps scrutinizing which dataset has a place in the process. It’s also possible that some managers who have been paying for transaction data were not able to see a meaningful ROI over the past few years as the data requires some work for it to consistently translate to alpha.

The main types of alternative data

The market has thousands of datasets with new datasets released frequently, but most of what hedge funds buy falls into a handful of categories. The table below summarizes six of them, and the sections that follow explain what each one is good for.

Geolocation data

Geolocation data aggregates anonymized location signals from mobile devices to estimate visits to stores, restaurants, factories and distribution centers. An analyst covering a retail chain can compare visit trends across its locations before same-store sales are reported, and an industrials analyst can watch activity at a competitor’s plant. Panels shift as apps and privacy settings change, so the first diligence question is how the provider keeps its sample consistent over time. There are many other issues with this data such as demographic bias, the inability to identify shopping locations in multi-storied buildings, panel fluctuations due to regulations, etc. Still, investors commonly find value in the data for certain companies.

Web traffic data

Web traffic data estimates visits and engagement on company websites, usually drawing on browser panels and clickstream sources. It is most useful for businesses that sell online or acquire customers online, where traffic trends can shed light on reported revenue weeks before the report. Estimates for smaller sites carry wide error bars, and a rise in visits does not always mean a rise in purchases. The blind spot is companies that do not rely on a website for revenue, which limits the universe mostly to consumer names.

Web scraping

Web scraping collects what companies publish on their own sites and on third-party platforms, including prices, product assortments, stock availability, job postings and customer reviews. A fund can track discounting at a retailer or hiring at a software company almost as it happens. Many firms build scrapers in-house, which lowers the cost of collection but leaves the fund responsible for reviewing each site’s terms of use. This data source is extremely popular and has a low bar to entry, especially as technology to write scrapers has become widely accessible and cheap. This data, however, is limited to very niche questions about a company or product and rarely shows strong correlations to KPIs that drive the business (with some exceptions).

Credit card data

A very popular source for hedge funds, transaction data drawn from anonymized card, bank and receipt panels is the most established way to nowcast revenue for consumer-facing companies. Because so many funds buy similar panels to forecast the same quarterly print, the edge has narrowed over time, which could explain Lowenstein Sandler’s finding that fewer respondents used consumer transaction data in 2025 than in earlier years. Bias toward particular demographics and card types remains the main modeling risk and funds need a sophisticated way to deal with bias, or hire consultants to do it for them.

App data

App data measures downloads, active users, session time and in-app spending across mobile platforms. For consumer apps, fintech and gaming companies, engagement often turns before revenue does, which makes it a useful early warning in either direction. Coverage is strongest for large apps on the major platforms and much thinner for niche or enterprise software.

Filing intelligence

Filing intelligence applies machine analysis to SEC filings, the record every other dataset in this list is eventually checked against. Rather than measuring behavior outside the company, it measures what the company itself disclosed: new risk factors, changed language, omitted assurances and shifts in emphasis between one 10-Q and the next. The source is public and free, so the value lies entirely in the processing, which means detecting changes across thousands of companies, judging which ones matter and doing both quickly enough to act.

How hedge funds put alternative data to work

Fundamental managers run concentrated books and use alternative data to test a thesis, checking whether a product launch is gaining traction, whether a turnaround is showing up in store visits, or whether the risk a short depends on is growing. They also use the data to run competitive analysis. Sometimes managers use more than one data source to build what is called a data mosaic to have a complete picture of the company’s positioning. It helps shape the story an analyst pitches to the PM and helps guide the assumptions inside the forecast model. The large majority of Lowenstein Sandler’s respondents said they combine alternative data with fundamental analysis.

Systematic managers were the first buyers of alternative data, and more than half of the data buyers in Neudata’s 2024 survey follow quantitative strategies. A quant team needs breadth and history, meaning consistent coverage across hundreds or thousands of names, point-in-time records long enough to test, and a delivery format that fits an existing pipeline. The question it asks of any new feature is whether it adds predictive power once the factors already in the model are accounted for.

Quantamental teams sit between the two, using systematic screens built on alternative data to decide where their analysts spend time and then applying fundamental judgment to the short list. A screen that flags forty companies with a change worth investigating is useful to them only if an analyst can trace each flag back to the evidence behind it.

Five questions to ask before you acquire the next dataset

Whatever your strategy, a dataset has to clear the same tests before it earns a place in your process. These are the questions we would want any client to ask us.

1. What did the data show at an earlier date?

Point-in-time data means each historical value reflects what was known on that date. Ask for original timestamps, preserved data vintages and the policy for restatements, and check whether delisted companies remain in the history, otherwise there will be survivorship bias.

2. Does it add something your existing factors miss?

Spending data on a stock that is already rising may mostly show priced-in momentum and that can overlap with your other signals. Try to understand what adds a unique signal given your standard factors and your own existing data.

3. Has it held up outside the backtest?

A backtest tests an idea against history, while a live record shows what a system produced before the outcomes were known (out of sample). Anyone can produce an impressive backtest by overfitting. Ask when production began, what has changed in the method since, and how the data behaved in weak periods and as more clients began trading on it.

4. What will it cost to run in production?

The license fee is only part of the cost, since researchers still have to map identifiers, handle gaps and revisions, and monitor delivery. Many managers have to hire highly paid consultants to work on the data first. Request a sample in the real delivery format and check coverage and arrival time against your own universe and decision schedule.

5. Can you trace and explain the output?

Every input to a live process will eventually have to be explained to a PM, a risk officer or a regulator. Ask where the data came from, what rights govern its use, what can change between runs, and whether a finding can be followed back to its source.

Where filing intelligence fits

Filing intelligence answers several of these questions by construction, which is why it complements the other types rather than competing with them. The source is public, so there is no personal data to license and none of the provenance questions that Lowenstein Sandler’s respondents rank among their top concerns. Every filing carries a timestamp, which makes a point-in-time record straightforward to verify. And because disclosures describe a company’s own risks and plans, they often capture changes that transaction or traffic data will show only later, if at all.

Increase Alpha applies filing intelligence in two ways. For systematic teams, our daily signal runs an independent model for each of 2,456 U.S. equities and produces forecasts over a ten-trading-day horizon, with timestamped production predictions going back to July 2021 and delivery as a daily flat file. The models are deterministic and purpose-built, with no general-purpose LLM in the forecasting process. For fundamental teams, the News Gap Report checks each filing theme against the press release, the earnings call, financial news and analyst coverage, and traces every finding to the filing section it came from. Coverage is U.S. equities only, and the signal is an input to test inside your own process rather than a strategy on its own.

Put it to the test

The most useful evaluation of any alternative dataset happens inside your own process, and ours is no exception. Systematic teams can request the signal specification and a point-in-time sample to test against their own factors, and the Agent Explorer shows company-level factor attribution in the meantime. Fundamental teams can request a News Gap Report on a ticker they cover, compare the findings with their own research, and put the five questions above to us directly.

For institutional use only. Past performance is not indicative of future results.