← All articles

Alternative Data Investing: A Workflow for Research Teams

How institutional teams turn alternative data into research: dataset types, quality tests, MNPI and vendor risks, and how to connect it to filings and your own work.

AllMind13 min read
Alternative Data Investing: A Workflow for Research Teams

Alternative data investing is the use of nontraditional datasets, such as card transactions, satellite imagery, web-scraped prices and app usage, alongside filings and broker research to form a view on a company or sector. The data itself is rarely the constraint. The constraint is the work around it: testing whether a dataset adds anything, documenting the compliance review, and connecting the signal to the filings, estimates and internal notes that your committee already relies on. We built AllMind for that connective work, and this article sets out the process in the order a research team encounters it.

Allmind
Bring Alternative Signals Into Context
Connect alternative signals with filings, financials, and your firm's own knowledge using AllMind's AI operating system for investment research.

TL;DR

  • Alternative data is information from outside company filings and broker notes. Its value depends on whether it models a specific indicator in your valuation framework, not on how novel it is.
  • Evaluate every dataset on originality, coverage, timeliness and transparency, then on statistical power, economic plausibility and whether it duplicates signals you already have.
  • The main risks are raw-data processing, vendor changes to collection methods or terms, and MNPI or PII exposure. Document due diligence and reassess vendors on a schedule.
  • A signal is only useful in a decision once it sits next to filings, financials and your firm’s own knowledge, with citations a committee can review.
  • AllMind connects 7,000+ premium sources, including alternative data, with a firm’s own knowledge through a proprietary ontology layer.

Table of Contents

What counts as alternative data

The working definition is negative: alternative data is nontraditional information that investors use beyond company filings and broker research. Traditional inputs, such as 10-Ks, earnings releases, consensus estimates and exchange market data, are produced by the company or the market and are visible to everyone at roughly the same time. Alternative data is usually a by-product of other activity: someone paying at a register, a ship leaving a port, a customer reviewing a product.

That origin explains both its appeal and its difficulty. Because it records behavior as it happens, it can move ahead of a quarterly report. Because it was never designed for investors, it arrives unstructured, incomplete or biased toward the panel that produced it.

Common dataset types

The FISD Alternative Data Council guide, published for the Financial Information Services Division, lists web-scraped retail prices and inventory, credit-card and point-of-sale data, geolocation, satellite imagery, social sentiment, search trends, shipping data, app installs and crowdsourced information. Deloitte Insights names a similar set: credit-card transactions, satellite images, geolocation and climate data, social media, shipping trackers, mobile-app usage and product reviews.

The table below maps these to the question each typically informs.

Dataset What it observes Research question it usually informs
Card and point-of-sale transactions Consumer spending, often from panels or payment providers Revenue trajectory for consumer-facing names
Web-scraped prices and inventory Retail pricing, stock levels, assortment Brand and company performance, pricing power
Geolocation Foot traffic and movement Store, venue or site activity
Satellite imagery Physical activity such as parking lots, facilities, shipping Operating activity and capacity
Shipping data Cargo movements Supply chains and trade exposure
App installs and usage Digital engagement User growth and retention
Social sentiment, search trends, reviews Public attention and opinion Demand shifts and brand perception

Adjacent categories matter to many teams as well: clinical trial registries, patents and mining data are not classic consumer-panel products, but they behave the same way in practice. They sit outside the financial statements, they require interpretation, and they gain value when linked to a company or asset. Our alternative data coverage for company and industry research describes how these sit within the platform.

How teams turn alternative data into an investment insight

The most common failure is starting with the dataset. Deloitte’s guidance is to start with an investment requirement and design the data attributes to model a key performance indicator in the firm’s intrinsic valuation model. A retailer’s same-store sales, a software company’s active users and a shipper’s volumes are KPIs. A dataset is relevant only if it explains movement in one of them.

That approach also sets the staffing model. Deloitte notes that data scientists understand the data while investment professionals understand valuation models and investment concepts, so the two groups should define requirements and test the data together. A data scientist alone may produce a statistically clean series that maps to nothing in the model. An analyst alone may not see that a panel is skewed.

Consumer-spending data illustrates the translation step. The FISD guide explains that such datasets can come from representative consumer panels or payment-technology providers and may be aggregated and extrapolated to produce investment insights. Extrapolation is where judgment enters: the panel’s representativeness, its treatment of cash and its changes over time all affect how far the extrapolation can be trusted.

A practical sequence:

  1. State the question. Name the company or sector, the KPI and the decision it informs.
  2. Identify candidate datasets that observe the KPI directly or through a close proxy.
  3. Run legal and compliance review before the data is used in research.
  4. Test the data against the KPI history, jointly between the investment and data-science teams.
  5. Record the result, including datasets that failed, so the work is not repeated.
  6. Integrate the signal into the model and the research record with its source and vintage.

How to evaluate a dataset before you rely on it

BlackRock’s systematic investing team separates data quality from signal quality. This is a useful split for any team, discretionary or quantitative.

Data quality

Assess four properties first:

  • Originality: is this a distinct source or a repackaging of something you already license?
  • Completeness and coverage: which companies, geographies and periods are included, and how many gaps exist?
  • Update speed and timestamp reliability: when did the event occur, and when did the data become available? Backtests built on dates that postdate availability overstate results.
  • Transparency: can the vendor explain the source, the processing and the version history?

Transparency has a research dimension as well. A 2023 peer-reviewed data paper on insider trading notes that public SEC insider-trading reports have been available in structured form since mid-2003, and it argues that commercial databases with opaque aggregation, or with historical records the provider can change, make replication difficult. Its dataset is built directly from the original regulatory filings and updated daily without altering the reported information. The lesson for any vendor evaluation is that reproducibility is a property you can ask about: how was the series built, and can history be restated?

Signal quality

Once the data passes, test whether it predicts. BlackRock describes testing prospective signals with quantitative measures including Information Coefficient, Predictive R-squared and Horizon-Decayed Information Ratio, alongside checks for economic plausibility and for additivity or redundancy. In plain terms:

  • Statistical predictive power: does the signal correlate with subsequent outcomes, and does it persist across the horizon you care about?
  • Economic plausibility: is there a believable mechanism linking the data to the KPI, or is the relationship a coincidence of the sample?
  • Additivity: does it add information beyond what your existing model captures?

The FISD guide makes two points that deserve attention at this stage. First, a dataset can carry a strong signal and still add little value if its output resembles signals your model already generates. Second, broader coverage of the investable universe can be preferable to a stronger signal limited to a single security, because it allows more capital to be allocated and better risk management. A dataset that is brilliant for one ticker and silent on the rest of the book has a narrow job.

Risks and compliance in alternative data investing

Deloitte identifies three categories of risk: the difficulty of processing raw data, vendor changes to collection methods or terms of use, and regulatory concerns involving material nonpublic information (MNPI) and personally identifiable information (PII). MNPI is information a reasonable investor would consider important that has not been made public. PII is data that identifies individuals.

An article in the Brooklyn Journal of Corporate, Financial & Commercial Law explains that Rule 10b5-1 prohibits trading on material nonpublic information in breach of a duty of trust or confidence. The practical consequence is that the question is not only what the data contains but how it was obtained. Data gathered under terms that restrict its use, or sourced through someone who owed a duty of confidentiality, raises a different question from data compiled from public activity. The author recommends written compliance policies, employee training, documented reporting of potential MNPI and ongoing monitoring of alternative-data use. Your counsel and compliance function own the legal judgment; the research process should make their review easy to perform and easy to evidence.

MNPI and the legal duty

Vendor risk

Deloitte observes that when a vendor changes its collection methods or terms of use, comparability across history can be impaired and an earlier risk assessment can be undermined. Its recommended controls are documented due diligence, periodic reassessment of vendors, and procedures to pause affected data after an adverse event so legal and compliance can review it. A vendor file that records what you were told, when, and what was tested is the document that keeps a committee comfortable in the second year of a subscription as well as the first. For a structured approach, see our guide to alternative data due diligence for institutional investors.

Processing burden

Raw alternative data is rarely ready to model. Entities need to be mapped to securities, histories need to be aligned, and definitions need to be reconciled across sources. This is often the largest hidden cost of a dataset, and it is why the question “what does it take to use this?” belongs next to “what does it predict?” Our note on what makes financial data ready for AI research covers the structural requirements.

Where the signal meets the rest of your research

A satellite series showing rising activity at a facility is a data point. It becomes research when it is read against the company’s guidance in the latest filing, the segment margins in its financials, what a broker note assumed, and what your own team wrote last quarter. In most firms those inputs sit in different systems and formats, and the connection is made by an analyst reading across tabs.

This is the problem a financial ontology addresses. An ontology defines entities such as companies, products, facilities and trials, and the relationships between them, so that a shipping record, a patent and a line in a 10-K can be recognized as pertaining to the same business. We explain the concept in What Is a Financial Ontology?. Without that layer, every new dataset adds another silo. With it, a new dataset can be connected to the entities and documents already in the research record.

Three workflow points follow for teams building this capability:

  • Tie every dataset to a KPI and a coverage assignment. An analyst should know which names a dataset supports and which it does not.
  • Keep citations attached to outputs. A conclusion that cannot be traced to its underlying records will not survive committee or compliance review.
  • Automate the recurring work. Weekly dataset refreshes, sector updates and portfolio monitoring are repetitive, and repetition is where analyst time is lost. Our guide to AI portfolio monitoring with sourced filings shows how a recurring workflow can be structured.

For teams comparing approaches to AI in research more broadly, How Hedge Funds Are Using AI in 2026 and our AI vendor landscape for institutional investment teams provide context.

A checklist for adopting a new alternative dataset

Stage Question to answer Evidence to keep
Requirement Which KPI in the valuation model does this inform? Written hypothesis
Sourcing How was the data collected, and under what terms? Vendor documentation, compliance sign-off
Quality Is it original, complete, timely and transparent? Coverage and timestamp review, version history
Signal Does it predict, make economic sense and add to current signals? Test results, including failures
Integration Where does it sit in the model and the research record? Linked sources and citations
Ongoing Have collection methods or terms changed? Scheduled vendor reassessment, pause procedure

Using AllMind for alternative data research

AllMind is an AI operating system for institutional investors. It integrates insights from more than 7,000 premium data sources with a firm’s own knowledge, including company filings, financial statistics, clinical trial results, patents, broker research, academic research, news, mining and alternative data. A proprietary ontology layer connects those datasets, and AI agents handle complex inquiries and recurring workflows so analysts can concentrate on the analysis.

The outputs are cited answers and reports suitable for compliance and committee review, and the platform maintains permissions and audit controls and data privacy standards throughout the research process. Teams that want to see how alternative data fits alongside filings and their own notes can start with the AllMind research platform.

Allmind

For a broader view of how the stack fits together, see AI software for buy-side teams and selecting an AI research platform for institutional investors.

FAQ

What is an example of alternative data?

Web-scraped retail prices and inventory are a common one, since they can help assess how a brand or company is performing. Others include credit-card and point-of-sale data, geolocation, satellite imagery, shipping data, app installs and social sentiment. Each is a by-product of activity outside the company’s own reporting.

How is alternative data different from traditional investment data?

Traditional data comes from company filings, financial statements and broker research, and is published on a schedule. Alternative data comes from other sources, often as a by-product of consumer or commercial activity, and is typically less structured. That makes it potentially earlier, but it also requires more processing and testing.

How do we know if a dataset is worth using?

Check originality, coverage, timeliness and transparency of source and processing. Then test predictive power, economic plausibility and whether the signal duplicates what the existing model already produces. A dataset that fails the last test adds cost without adding information.

What are the main compliance concerns?

The principal concerns are MNPI and PII. Rule 10b5-1 prohibits trading on material nonpublic information in breach of a duty of trust or confidence, so the way data was obtained matters. Written policies, training, documented review and periodic vendor reassessment are the controls the research recommends, with legal and compliance making the final call.

Can alternative data be combined with our own research?

Yes, and that is where much of the value lies. A signal is more useful when it is read against filings, financials, estimates and the team’s prior work, with sources attached. An ontology-based platform makes those connections systematic rather than manual.

Sources