How to Estimate Private Company Revenue for M&A

Executive Summary

Estimating private-company revenue for an acquisition screen is an evidence-classification problem, not a search for a magic database field. U.S. private companies generally do not make the Securities and Exchange Commission filings expected of public issuers ^[1]. A useful output is therefore a dated range, with a named metric, entity, period, currency, method, and confidence status. A point estimate without those fields is less decision-useful than an honestly unresolved answer.

The evidence hierarchy starts with a filed financial statement, followed by an official filing that actually contains a relevant revenue figure or band, a dated company statement, a credible transaction or credit document, a documented vendor estimate, and a transparent analyst model. An unexplained database value belongs at the bottom. Form D is not revenue evidence: the SEC calls it a notice of an exempt securities offering ^[2]. Commercial fields also need labels. Crunchbase, for example, describes its field as an estimated revenue range ^[3]. That is a useful lead, but not the same thing as filed revenue.

Metric normalization comes before modeling. Revenue, gross receipts, billings, bookings, annual recurring revenue (ARR), gross merchandise volume (GMV), run rate, and systemwide sales are not interchangeable. International Financial Reporting Standard 15 ties revenue recognition to transfer of promised goods or services ^[4]. SUSB, the Census Bureau's Statistics of U.S. Businesses program, defines receipts as operating revenue from goods or services ^[5], but warns that summed establishment receipts can duplicate amounts for large multi-establishment enterprises ^[6].

Proxy models should produce ranges, not false precision. Revenue per employee is common, but even its proponents note that company headcount is difficult to observe ^[7]. Official industry ratios are benchmarks, not target facts, and SUSB can lag its reference year by roughly 2 to 2.5 years ^[8]. The screening rule is simple: provisionally pass when the low bound clears the mandate floor, provisionally fail when the high bound is below it, and keep the target pending when the range straddles it. Apply a separate evidence-quality gate. Unknown is a research status, not a rejection.

Introduction and Background

The practical question for a private-equity origination analyst is not, “What is the company's revenue?” It is, “Does the available evidence justify classifying this candidate against the mandate today?” That reframing separates acquisition target revenue screening from financial due diligence, quality of earnings work, valuation, or a management representation.

Private-company financial data is uneven by design. The Library of Congress notes both that a private company's own website may supply the most information and that general web information can be dated or wrong ^{[9] [10]}. A university research guide makes the key distinction explicit: private-company revenues may be estimated, while public-company revenues may come from filings ^[11].

For Docket's private-equity audience, that means preserving provenance throughout the screen. Docket states that each researched answer includes a source, supporting excerpt, and collection date (Source: docket.capital). Its Audit stage separately reviews collected evidence and leaves unresolved questions visible (Source: docket.capital). Those controls do not make an estimate true. They make the reasoning inspectable.

This report provides a repeatable method to accept, bracket, model, or leave revenue unknown. It covers private-company financial data sources, revenue estimation methods, entity and period normalization, proxy failures, contradiction handling, threshold decisions, and escalation. Every modeled result remains an estimate until better evidence replaces it.

Define the Screening Decision and Evidence Taxonomy

The decision is narrower than diligence

A revenue screen answers whether a company appears clearly above, clearly below, or near a mandate threshold. It should not imply audited accuracy. The analyst should record two gates:

• Economic gate: Does the supported low and high range sit above, below, or across the mandate floor?

• Evidence gate: Is the source quality adequate for the action being considered?

• Freshness gate: Does the reference period still represent the business being screened?

• Scope gate: Does the figure cover the legal entity, consolidated group, locations, or franchise system intended by the mandate?

• Metric gate: Is the value revenue, receipts, ARR, GMV, bookings, billings, or another measure?

A target can clear the economic gate but fail the evidence gate. That company remains pending. Conversely, a reliable filed number below the floor can support a provisional fail without spending time on a proxy model.

Evidence ladder

Table 1 ranks source types by what they can support. “Typical lag” is directional because filing regimes, fiscal years, and publication schedules differ.

The ladder prevents flattening. A current regulator-hosted revenue statement and an undated vendor point estimate may both occupy a spreadsheet cell, but they should not receive the same confidence code. The SEC says Form D becomes public on EDGAR after filing ^[12]; public availability does not change the meaning of the offering fields.

Normalize the Metric, Entity, Period, and Currency

Translate the metric before using it

Table 2 is a metric translation table. It tells the analyst what a field can and cannot establish.

The practical rule is never translate by label alone. Ask how the source defines the numerator, when it is recognized, whether taxes and pass-through amounts are excluded, and whether the amount is gross or net. A claimed $50 million of GMV may imply a much smaller revenue amount, while $50 million of systemwide sales may include franchisee sales outside the franchisor's income statement.

Resolve entity and period

Entity mistakes can overwhelm model error. Consolidated statements treat a parent and subsidiaries as one economic entity ^[22]. Census, by contrast, defines an establishment as one physical business location ^[23]. Confirm whether the target record represents a legal entity, parent group, subsidiary, trade name, franchise system, or single location.

Then normalize the period and currency:

• Fiscal basis: A fiscal year is 12 consecutive months and need not align with the calendar year ^[24].

• Short periods: A short tax year is less than 12 months ^[25]; annualization should be explicit.

• Acquisition perimeter: Separate continuing operations from revenue added by acquisitions during the reference period.

• Currency: Store the nominal currency, source-period average or closing conversion basis, conversion date, and original amount.

• Vintage: Preserve both the source publication date and the economic reference period.

• Classification: Use the operating activity and production process, not the company label.

Build Proxy Models Without False Precision

Revenue per employee

The common model is employee range × industry receipts-per-employee range. Grata describes revenue per employee as the most common investor method ^[26], but also warns that employee counts are not readily accessible and that some sectors are underrepresented in professional-network data ^[27].

The model should retain four corners:

• Low case: low employee count multiplied by low sector receipts per employee.

• Headcount case: high employee count multiplied by low sector ratio.

• Productivity case: low employee count multiplied by high sector ratio.

• High case: high employee count multiplied by high sector ratio.

Do not compress those corners to a midpoint unless the decision rule specifically requires one. Also reconcile what “employee” means. The Bureau of Labor Statistics' Quarterly Census of Employment and Wages (QCEW) excludes proprietors, unincorporated self-employed workers, and unpaid family members ^[28]. Contractors, part-time workers, owners, and duplicate profiles can make a target's observed count noncomparable with an official denominator.

Unit, customer, pricing, and payroll models

Alternative models can be stronger when the operating unit is visible:

• Locations × sales per location: Useful for homogeneous site networks; adjust for openings, closures, ramp, local mix, and franchise ownership.

• Customers × average price: Grata notes that a customer, member, subscriber, or unit count can be multiplied by observed pricing ^[29]. Adjust for discounts, free tiers, seat counts, churn, usage, and partial periods.

• Capacity × utilization × price: Appropriate where rooms, vehicles, procedures, or production lines constrain output. Each factor needs its own range.

• Payroll ÷ payroll-to-receipts ratio: Official statistics can support sector priors, but SUSB itself uses payroll-to-receipts ratios to estimate some out-of-scope multi-establishment receipts ^[30]. Treat that as a modeled benchmark, not circular confirmation.

• Public comparable scaling: Match size, scope, business model, geography, and maturity. S&P Global cautions that public benchmarks may differ materially in scale or maturity ^[31].

• Digital activity model: Traffic, reviews, app use, or job postings can indicate scale or direction, but should not become revenue without a calibrated relationship. Grata says web traffic may feed e-commerce revenue models ^[32].

Academic research supports uncertainty-aware output. Work on scaleups characterizes financial data as proprietary, costly, and scarce ^[33] and describes a revenue-extrapolation model that produces confidence estimates ^[34]. A separate thesis tested 12 models on data for 2,245 private companies (Source: aaltodoc.aalto.fi) (Source: aaltodoc.aalto.fi). These studies show that modeling is possible, not that a model transfers unchanged to every sector or target.

Triangulate Independent Evidence and Preserve Contradictions

Triangulation is not counting how many databases display the same value. Several vendors may ingest the same company statement, public directory, or upstream provider. Correlated repetition is still one input.

A practical reconciliation sequence

• Create a claim ledger: Store low, high, metric, period, currency, entity, source, source date, method, and exact supporting passage.

• Trace originators: Follow a vendor citation back to the filing, company statement, official dataset, or research producer that created the number.

• Test independence: Ask whether each source was independently observed or merely redistributed.

• Normalize first: Convert comparable periods, currencies, scopes, and metrics before comparing bounds.

• Record contradictions: Keep both claims and explain why they differ. Do not silently overwrite the older field.

• Widen when incompatible: If two credible sources cover different entities or metrics, create separate ranges or widen the combined range rather than averaging.

• Down-weight staleness: A precise but old figure may deserve less decision weight than a current operating signal.

• Stop when marginal value falls: Escalate once another public search is unlikely to change the threshold classification.

Vendor documentation supports this discipline but remains self-description. PrivCo says no single feed is trusted on its own and that estimates require convergence across signals ^{[35] [36]}. It also says researchers investigate disagreement instead of averaging it ^[37]. Gartner describes market-share estimation as combining primary surveys and vendor briefings with secondary research ^[38]. The transferable lesson is methodological diversity, not automatic acceptance of a vendor field.

Commercial values should therefore carry at least four labels:

• Estimated or reported: Never present one as the other.

• As-of date: Apollo notes that a field without a refresh timestamp has unknown vintage ^[39].

• Range and confidence: PrivCo says its profiles show when financial information was updated and confidence in each estimate ^[40].

• Method family: Headcount, web activity, company statement, transaction data, or a blended model.

Apply Threshold Rules Under Uncertainty

The output should be a status, not a forced answer. Let L be the defensible low bound, H the high bound, and F the mandate floor:

• Clearly above: If L exceeds F and the evidence gate passes, mark “provisional pass.”

• Clearly below: If H is below F and the evidence gate passes, mark “provisional fail.”

• Straddles threshold: If L is at or below F and H is at or above F, mark “pending.”

• Stale: If the newest support predates material changes in the company or market, keep the economic range but fail freshness.

• Unknown: If no defensible bounds can be formed, record “unknown.” Unknown is not “no.”

Table 3 shows a sensitivity grid for a Hypothetical Example, a fictional business-services company. The employee bounds and sector-ratio bounds are analyst-selected assumptions for demonstrating the method, not claims about a real company or industry. The fictional mandate floor is $10 million.

The range is $5.6 million to $13.2 million, so it crosses the fictional $10 million floor. Headcount alone cannot flip the result under the low productivity assumption, while the combined high assumptions do. The correct status is pending, not a midpoint-based pass or fail. Because the inputs are hypothetical, the table is a calculation example only. In a live screen, each bound would link to a dated source and an evidence-quality code.

Data Analysis and Evidence

Official datasets are valuable because they define populations and supply sector priors. They do not disclose an individual target's revenue. The 2022 Economic Census represents more than 8 million business locations, with approximately 4.2 million asked to complete surveys ^{[41] [42]}.

Coverage does not remove measurement limitations. The 2022 Economic Census unit response rate was just over 60%, and basic statistics combine reported data with administrative records or imputation ^{[43] [44]}. Its remaining 2022 releases were complete by March 2026, illustrating the reference-period lag ^[45].

The QCEW offers a fresher labor denominator. Employer reports cover more than 95% of U.S. jobs, with quarterly county employment and wage data released within six months of the reference quarter ^{[46] [47]}. National reporting typically exceeds 90% of establishments, while BLS says the remaining 5% to 10% of data are imputed ^{[48] [49]}. This can update employment assumptions, but QCEW wages are not revenue.

SUSB adds receipts but requires extra caution:

• Modeled values: Some multi-establishment receipts are estimated with payroll-to-receipts ratios.

• Possible duplication: Summed establishment receipts may duplicate enterprise receipts.

• Population mismatch: Employer datasets exclude nonemployers; Nonemployer Statistics instead draws primarily from IRS business income-tax records ^[50].

For each live model, the range worksheet should store:

• Low and high estimate, without a manufactured midpoint.

• Reference period and publication date, separately.

• Original currency and conversion basis.

• Entity perimeter, including parents, subsidiaries, and locations.

• Metric definition, including gross-versus-net treatment.

• Model family and formula.

• Independent inputs, each with source and excerpt.

• Contradictions and resolution status.

• Evidence grade and freshness grade.

• Threshold classification and review owner.

The evidence quality itself is data. Crunchbase says public data comes from contributors, public sources, and partners ^[51]. Apollo characterizes private-company annual revenue as a modeled range ^[52]. Those labels should remain attached when values enter a screening system.

Case Studies and Real-World Examples

Labor-intensive services (Hypothetical Example)

Harbor Field Services is fictional. Public pages indicate several branches, but observed headcount varies by source and includes contractors. The analyst should:

• define whether the target is the operating subsidiary or its parent;

• create a low and high employee count after de-duplication;

• select a six-digit NAICS activity based on production process;

• calculate a sector receipts-per-employee range from official tables;

• widen the range for contractors and owner labor excluded from official denominators;

• cross-check branch count multiplied by revenue per branch;

• keep the classification pending if the range crosses the mandate floor.

This sector is labor sensitive. A small change in worker count or mix may move the result materially. County Business Patterns excludes self-employed people and businesses without employees ^[53]. That makes an owner-heavy target especially prone to denominator mismatch.

Software marketplace (Hypothetical Example)

Northstar Workflow Exchange is fictional. Its website discusses subscription ARR and transaction volume, but neither equals recognized revenue. The analyst should:

• separate subscription contracts from marketplace transactions;

• translate ARR into a revenue range using activation timing, churn, and services assumptions;

• translate GMV using an evidenced take-rate range;

• exclude customer funds and pass-through amounts where appropriate;

• test the output against headcount only as a secondary model;

• preserve separate subscription and transaction ranges before combining them;

• request management data if the threshold decision remains sensitive to take rate.

The public-company definition used earlier shows why: take rate can be defined as net revenue divided by GMV ^[54]. A marketplace's large transaction volume can coexist with much lower recognized revenue. Public comps should inform the parameter range, not be copied as the fictional target's economics.

Implementation Guidance for PE Origination Teams

A reproducible screening workflow

1. Resolve identity. Confirm domain, legal name, parent, subsidiaries, and operating locations before collecting numbers.

2. Search direct evidence first. Check filed accounts, regulator records, tax-exempt returns where applicable, company statements, and transaction documents.

3. Reject category errors. Do not use a Form D offering amount as revenue. The SEC requires the filing within 15 days after the first securities sale, which dates the offering notice, not the company's operating period ^[55].

4. Name the metric. Record revenue, receipts, ARR, GMV, bookings, billings, or systemwide sales exactly as the source does.

5. Normalize scope and time. Align entity, fiscal period, currency, acquisition perimeter, and continuing operations.

6. Build at least two models when possible. For example, headcount × sector ratio and locations × revenue per unit.

7. Trace independence. Count originators, not database appearances.

8. Carry four corners. Preserve low-low, high-low, low-high, and high-high model outputs.

9. Apply two gates. Classify economics and evidence quality separately.

10. Write a stop note. State what was checked, what remains unknown, and which document could resolve it.

Some entities have unusually good public records. For tax-exempt organizations, IRS tools provide individual and bulk access to covered Form 990-series filings received in 2017 or later ^[56]. For a standard commercial company, that route usually does not exist. SBA size rules offer another useful definition, but not a universal revenue answer: they define receipts broadly and generally rely on federal tax returns ^{[57] [58]}.

When to stop and escalate

Escalate to a human reviewer or management-provided data when:

• The range straddles the floor and no independent public input is likely to narrow it.

• The entity perimeter is unresolved, especially after a recent acquisition or reorganization.

• The metric is nonstandard, adjusted, or not reconcilable to recognized revenue.

• The source is stale relative to visible operating change.

• Two credible sources conflict and their scopes cannot be reconciled.

• One assumption drives the result, such as take rate, utilization, or contractor count.

• The only available value is unexplained or circularly repeated.

• The expected outreach cost or opportunity value justifies requesting direct information.

The escalation package should contain the range worksheet, source excerpts, model formulas, sensitivity result, and precise unresolved question. That enables a reviewer to improve the estimate without repeating the entire search.

Implications and Future Directions

Revenue screening will become more automated, but automation increases the importance of explicit semantics. A system that can populate more cells faster can also propagate a mislabeled GMV, stale ARR, or offering amount across a larger target universe. The durable advantage is therefore structured uncertainty, not maximum fill rate.

Three practices are likely to matter most:

• Field-level provenance: Every value should retain source, excerpt, access date, period, metric, entity, and transformation history.

• Range-native workflows: Screening systems should store low and high values plus evidence grade, not force all evidence into a single numeric field.

• Contradiction-aware review: Conflicting observations should remain visible until a documented resolution changes status.

Commercial providers increasingly describe similar principles. S&P Global recommends evidence-backed ranges supported by multiple signals ^[59]. Datasite describes private-market data as fragmented and notes that analysts may assemble government records across sources ^[60]. These are vendor-affiliated observations, not independent accuracy benchmarks.

The important design choice for a PE screening rubric is to reward warranted abstention. A system should be able to say “unknown after specified sources were checked.” That status preserves a potentially attractive target while preventing an unsupported estimate from steering outreach priority. Better public and commercial data can narrow ranges, but management-provided records and later diligence remain the validation path.

Frequently Asked Questions (FAQs)

How can an analyst find private company revenue?

Start with direct evidence: filed accounts where the jurisdiction makes them public, regulator records, company statements, transaction documents, and tax-exempt filings when applicable. Then consult commercial databases, but retain whether the field is estimated, its range, update date, and method. The Library of Congress notes that information may also appear in a public acquirer's SEC filings after a transaction ^[61].

Which private company financial data sources are most reliable?

Reliability is claim-specific. A filed statement is strong for the defined entity and period. An official record is strong only for the field it actually contains. A company statement is first-party but may use a nonstandard metric. A vendor estimate is useful when method, date, scope, and confidence are disclosed. A transparent analyst range can be better than an unexplained point value. PrivCo says estimates should be labeled as estimates ^[62].

Can employee count estimate company revenue?

It can support a range when headcount and sector productivity are both bounded with comparable definitions. It should not produce a precise point. Official ratios may mix firms of different size, geography, capital intensity, and business model. Professional-network counts may omit or duplicate workers. Use at least one independent operating model where possible.

Which revenue estimation methods work for private companies?

The strongest method depends on what can be observed. Employee count multiplied by a sector productivity range suits labor-linked businesses. Locations multiplied by revenue per unit can fit site networks. Customers multiplied by realized price can fit subscriptions, while capacity multiplied by utilization and price can fit constrained services. Payroll ratios and private-company or public-company peers are secondary checks. Every method should preserve low and high assumptions, not force a precise answer.

How does this support private equity target company research?

The workflow turns an uncertain private company revenue estimate into a research status that a screening-rubric owner can use. It separates direct evidence from modeled evidence, identifies the assumption that changes classification, and records what additional source could resolve the answer. This supports acquisition target revenue screening before outreach without presenting the result as diligence-grade financial fact.

Is Form D revenue?

No. Form D is an exempt-offering notice, and its amount fields concern securities offered or sold. The form itself instructs filers to enter the dollar amount of securities being offered ^[63]. Offering size is financing activity, not operating revenue.

How should a revenue estimate be verified?

Verify identity, source text, metric, entity scope, fiscal period, currency, and whether sources are independent. Recalculate every transformation. Preserve contradictions and compare at least two models. If the decision remains sensitive, request management financials or leave the field pending. Gartner's documented method begins by estimating vendor revenue in each active market segment, illustrating why segmentation matters before aggregation ^[64].

What should happen when the estimate crosses the mandate threshold?

Keep the target pending. Report the low and high bounds, identify the assumption that changes classification, and apply a separate evidence-quality gate. Do not average incompatible estimates merely to force a decision. Unknown or pending should trigger targeted follow-up, not automatic rejection.

Conclusion

The defensible way to estimate private-company revenue is to preserve the distance between evidence and inference. Start with the strongest direct source, name the metric exactly, align the entity and fiscal period, and retain source dates. Treat vendor ranges as estimates, financing amounts as financing amounts, and aggregate industry statistics as model inputs rather than company facts.

When direct evidence is absent, build multiple transparent ranges. Revenue per employee can be useful, but only when both headcount and sector productivity are bounded consistently. Unit count, pricing, capacity, payroll, customer, and public-comparable models provide independent checks when their assumptions are observable. A wider honest range is more useful than a precise unsupported point.

For acquisition target screening, the final classification should combine economics and evidence quality. A low bound above the mandate floor supports a provisional pass. A high bound below it supports a provisional fail. A straddling range stays pending. Stale or irreconcilable inputs remain unresolved. This approach gives PE origination teams a reproducible research status while reserving financial diligence and valuation conclusions for the documents and processes designed to support them.