← Blog

Investment Thesis Data: What to Gather and How to Use It

Which company, ownership, financial and risk data a strong investment thesis needs, how to test its quality, and how to turn it into kill criteria and monitoring.

Investment Thesis Data: What to Gather and How to Use It

Abstract evidence structure for investment thesis data

Investment thesis data is the evidence that turns a hunch about a company or sector into a claim you can test: who the company is, who owns it, how it earns and spends, what risks surround it, and which signals would prove you wrong. The strongest theses rest on one consistent, source-verified view of each company rather than on scattered tabs and stale spreadsheets. That is the reason we built Milda.ai as a single company intelligence platform covering 420 million companies and 50 million verified contacts.

Milda
milda.ai
Build a More Complete Investment Thesis
Milda brings together company, ownership, financial, risk, and growth signal data to help investment teams assess opportunities with more context.
Visit

TL;DR

  • A thesis is a falsifiable claim plus evidence, a valuation view, a risk boundary and a review plan. Each part needs its own data.
  • Core data falls into four groups: company identity and ownership, business and industry, financial quality, and risk.
  • Judge any dataset on freshness, coverage, source and bias before it carries weight in your argument.
  • Write kill criteria as observable, dated data points, then track them after you invest.
  • Private and early-stage targets are where coverage gaps hurt most, so check how many of your candidates a source actually sees.

Table of Contents

What investment thesis data actually is

An investment thesis is a strategic statement, backed by research, explaining why an investment is likely to succeed and fit the investor’s goals. Investopedia lists its usual contents: the investment itself, goals, viability and supporting trends, potential downsides and risks, and costs against potential returns and losses.

Investment thesis data is whatever supports, sizes or challenges each of those parts. Investment research data is the broader pool: filings, estimates, market statistics, alternative sources. The difference is selection. Research data is everything you could read. Thesis data is the small set of facts your claim depends on, chosen before you look, so the evidence can actually disagree with you.

Research evidence filtered into balanced thesis inputs

The line between the two is also blurring on the vendor side. Integrity Research reports that alternative data providers increasingly deliver KPIs, analytics and finished research insights rather than raw feeds, partly because many asset managers lack the resources to integrate raw data into a fundamental process. For a lean team, that means the useful question is not “how much data can we buy?” but “how quickly can a data point become a decision?”

The parts of a thesis and the data each one needs

Frameworks differ in wording but agree on the skeleton. Equity Analysis Lab separates six questions: what must be true, what evidence supports it, how valuation changes the reading, what could break the thesis, how the position fits a portfolio, and what to monitor afterward. Kubera’s guide makes the same point from the other direction: most weak theses fail because they contain only the first two items, the claim and the evidence.

Here is how those parts map to data you need to collect.

Thesis part What it asks Data that answers it
Core claim What will happen, why, over what period Segment and market size, growth history, company firmographics
Business evidence Is the economics durable? Revenue model, customers, pricing, patents, trade flows, competitor map
Financial quality Do profits turn into cash? Revenue, margins, cash flow, debt, working capital, share count
Valuation context What is already priced in? Peer multiples, historical ranges, ownership and recent funding
Risk boundary What would invalidate it? Legal, regulatory and financial risk indicators, leadership changes
Review logic When do we re-test? Growth signals, filings, events, milestone metrics

The sharpest claims are specific. Kubera’s example is a company that can hold net revenue retention above 110% for five years, supporting margins the current price does not reflect. Notice that this sentence names its own metric and horizon, so the data you need is obvious before you start.

Company, ownership, financial and risk data

The CFA Institute’s equity research report essentials is a good checklist for a practitioner. It groups supporting data into four buckets.

Company and ownership data

The guide opens with identifiers: ticker, exchange, sector, price, market capitalization, liquidity, float and major shareholders. Float can be far smaller than market cap, which matters for large institutions that need to build a position. For private targets there is no ticker, so identity and ownership become the foundation of everything else: legal entity, parent and subsidiaries, headquarters, headcount, and who holds the equity.

Ownership is more than a footnote. It tells you who can force a sale, who is likely to block one, and who has incentives that differ from yours. Our guides on reading who owns a company and on shareholder data explain how to read cap tables and holding structures. In the platform, ownership and shareholder records sit beside the firmographics, so you do not rebuild a company’s structure from separate lookups.

Business and industry data

The CFA guide points to revenue and expense drivers, competitive analysis through Porter’s Five Forces, production capacity, pricing, distribution and the stability of market share. In practice this means answering: who are the real competitors, how concentrated is the field, and where does this company sit in it. Competitor mapping, patents, publications and trade data are useful here because they show capability and traction that a press release does not.

Financial data

The guide calls for historical results, forecasts, footnote-based earnings-quality checks and industry-specific ratios. Equity Analysis Lab adds that a thesis weakens when accounting profit rises while cash conversion falls, so revenue alone never settles the question. Look at margins, cash flow, debt, working capital and share count together.

For listed companies this is filing work. For private companies it is harder, and we cover what to trust and what to discount in our guide to private company financials.

Risk data

The CFA guide lists operational, financial, regulatory and legal risks, plus auditor qualified opinions and material weaknesses in internal control. Treat these as thesis inputs, not an appendix. If your thesis depends on a regulatory outcome, that outcome belongs in your kill criteria, not in a closing paragraph of caveats.

When the target is private or early-stage

Most screening tools are built around listed companies. Funds hunting for growth, corporate innovation teams scouting partners and government strategy offices mapping an industry often care about firms that file nothing public. There, the evidence is indirect: hiring, new products, patents, trade activity, ownership changes and funding events.

These event-driven signals are what we call growth signals, and we explain how to read and act on them in our guide to company growth signals. Our private market intelligence guide covers how to assemble a view of a company outside public disclosure.

Market sizing is the other half. A thesis about a niche, for example in life sciences, needs a bottom-up estimate of the addressable market that you can validate rather than a top-down headline number. For that step, sizing and validating a market is worth working through before you commit to a number.

As a rule, write the market-size assumption as a claim with a source and a range, and let your company-level data confirm that enough real companies exist to support it.

How to judge the data before it carries weight

Bad evidence is worse than none, because it feels like rigor. Test every source on four properties.

  1. Freshness. When was each record last verified? A firmographic from three years ago can describe a company that has been acquired, renamed or wound down. Ask how often records refresh and whether events update them or batch jobs do.
  2. Coverage. How many of your candidate companies does the source actually contain? Take a list of 50 targets you already know and count the hits. Missing companies bias a screen silently: you never see what was never in the database. Our own claim is that our multi-source approach surfaces up to three times more companies than others, and our data coverage page shows what the database holds.
  3. Provenance. Can you trace a number to its origin: a registry, a filing, a website, a model estimate? Source-verified records let you tell reported facts from inferred ones.
  4. Bias. Who gets covered, and who does not? Large firms, certain countries and certain sectors are over-represented in most sources. A thesis about small suppliers in a mid-size market is exactly where a gap in coverage distorts the conclusion.

The discipline of checking a model against outcomes applies here too. The CFA Institute Research Foundation’s model validation guide describes backtesting as comparing predictions with actual outcomes, and notes that regular backtesting lets managers adapt as conditions change. Apply the same habit to your data: periodically check whether the signals you relied on in past decisions preceded the outcomes you expected.

BlackRock’s From data to deals whitepaper makes a related point. Public markets act as a standardized, multi-cycle laboratory where alternative signals can be validated against observable financial outcomes, such as product-release activity matched against later revenue growth or margin expansion. It describes a data-informed framework as augmenting fundamental diligence, not replacing it. That is the right mental model: data sharpens judgment and speeds screening, and the thesis remains yours.

Turn the data into kill criteria

The most valuable and most commonly missing component of a thesis is the finding that would kill it. L3VLUP’s framework says good kill criteria are observable, dated and outside your control. Its examples: subscriber growth below four per cent in either of the next two quarters, or gross margin failing to reach 41 per cent by the second half. It also notes that a falling price is not information, and that “the thesis not playing out” is not a criterion.

A practical way to write them:

  1. Pick the two to five conditions that must hold for your claim to be true.
  2. Phrase each as a positive statement with a number and a date.
  3. Name the data source that will report it, so the check is mechanical.
  4. Write the opposite as the trigger for review.

A worked example. Suppose your claim is that a regional logistics software company can grow revenue at 15% a year because carriers are consolidating onto fewer platforms. Your conditions might be: at least three of the five largest carriers in the region are already customers; no top-three competitor has been acquired by a deep-pocketed strategic buyer; and ownership stays free of a shareholder with a competing product. Each condition maps to data you can pull: customer and competitor lists, ownership records, and event signals such as acquisitions. The kill criteria then write themselves.

The Equity Analysis Lab framework also notes that a risk boundary without an invalidation condition, and a thesis treated as static, are the typical weaknesses of those layers. Build both into the first draft.

Monitoring: update facts, do not defend the opinion

After you invest, the thesis becomes a monitoring plan. AI Vista Capital’s learning center says monitoring should update facts rather than defend the original opinion, and that indicators, filings or events, not price moves alone, should trigger a review. Equity Analysis Lab’s research process lists the monitoring inputs: earnings releases, guidance, filings, competitive changes and thesis milestones.

Two habits keep this honest:

  • Separate noise from mechanism. A small quarterly miss rarely matters. A change in the underlying mechanism, such as a new entrant or a lost anchor customer, does.
  • Set a review cadence. Quarterly is common, with event-driven reviews between. Ownership changes, leadership departures, acquisitions and new filings are the events most worth alerting on, and they are the kind of M&A and ownership movement we describe in our guide to M&A signals.

For deal teams, the same data supports diligence once a thesis survives screening. Our financial due diligence checklist lists eight checks for buyers and sellers.

Our view: the thesis is only as good as the company record behind it

We build Milda.ai on a simple belief: most thesis failures begin as data failures. A team picks a promising segment, screens with whatever source is nearest, and never learns how many relevant companies the source missed. By the time ownership, financials and risk are checked, the idea has momentum and the checks become a formality.

We designed the platform to reverse that order. Coverage comes first: a database of more than 420 million companies, assembled from proprietary and multi-source data, with patented technology behind the accuracy of the datasets. Then depth: firmographics, ownership and shareholders, financials, risk assessments, trade data, patents, publications and competitor maps on one company record. Then movement: event-driven growth signals that tell you when a thesis input has changed.

The same record is available in the web app, the browser extension and the API, so an analyst sketching a thesis, a business development lead finding the right contact among 50 million verified contacts, and a data team feeding an internal model all work from the same facts. Our data is source-verified and GDPR-compliant, which matters when the output goes into an investment committee paper.

A good thesis asks to be proven wrong. We want the evidence it meets to be complete enough that the answer is trustworthy.

Vilius Stanciauskas, Founder and CEO

Put the thesis data in one place

If you are building or stress-testing a thesis, Milda.ai gives you the company, ownership, financial, risk and signal data described above in one workspace. Browser Analyst, Strategy Terminal and Data Solutions cover the web app, extension and API routes, and pricing adapts to team size and data usage. See the options on our pricing page.

Milda

For market mapping specifically, our market research use case shows how teams build the company universe behind a thesis.

FAQ

What does an investment thesis look like?

It is a short document, often one to three pages, that states a falsifiable claim, the evidence for it, a valuation view, the main risks, and what would change your mind. Frameworks such as Kubera’s ask the claim to be one sentence covering what you expect, why, over what period and through what payoff mechanism. The rest of the page exists to test that sentence.

What data do I need first?

Start with company identity and ownership, then business and industry data, then financial and risk data, following the CFA Institute’s equity research essentials. Gather only what bears on your core claim. Anything else can wait until the idea earns diligence.

How is thesis data different from general research data?

Research data is the whole pool of filings, statistics and alternative sources. Thesis data is the subset you chose in advance because your claim depends on it. Choosing before you look keeps you from collecting only the facts that agree with you.

How do I know if my data is good enough?

Test freshness, coverage, provenance and bias. Run 50 companies you already know through the source and see how many appear, how recent the records are, and whether each figure traces to an origin. Repeat the check each time you enter a new sector or country.

When should I change my mind about a thesis?

When a kill criterion you wrote in advance is met, or when the underlying mechanism changes. A falling share price alone is not a reason, as L3VLUP notes. Tie each review to a dated, observable data point.

Sources

See it against your own question

Bring a market, a list or a counterparty you are already working on, and we will run it through the platform on the call.

© 2026 Milda.ai All rights reserved.