Government contract analysis: measure demand, not row counts.

Government contract analysis should begin with an observable market definition—not a download—and measure opportunity demand, awards, signed contracts, and financial actions at their own grains. Raw notice counts are not contract counts. Estimated tender value is not spending. A framework value repeated beside five suppliers is still one reported value unless the source allocates it.

A defensible analysis states:

• the business question;
• the jurisdictions and official sources;
• the included procurement stages;
• the object being counted;
• the date field, window, and retrieval cutoff;
• the value basis and currency policy; and
• the missing, unresolved, or excluded share.

The result is best described as observed procurement recorded in the selected sources and window. It is not automatically the total government-contract market. Publication thresholds, exemptions, reporting delays, source transitions, below-threshold practices, vehicle-only competitions, and missing values all shape what can be observed.

Government contract analysis: the short answer

Use this sequence:

1. Write one market question that names the buyer, category, geography, stage, and period.
2. Choose one native population: opportunities, award decisions, signed contracts, or financial transactions.
3. Deduplicate to the object that answers the question: procedure, lot, award, contract, order, transaction, or supplier relationship.
4. Keep NAICS, PSC, and CPV in their original schemes and versions.
5. Resolve buyers and suppliers with source identifiers before grouping names.
6. Calculate counts and values separately, using one compatible value basis per measure.
7. Attach value completeness, supplier-identity coverage, unallocated multi-supplier value, and source coverage to every ranking or concentration result.
8. Reconcile the result against the source and retain the exact filters, cutoff, and calculation version.

The government contracts database guide defines the underlying opportunity, award, contract, transaction, and supplier layers. This article turns those definitions into a quantitative market-sizing method.

Define the observable market before calculating it

“Government IT market” is not an analysis specification. It leaves the level of government, category boundary, procurement stage, period, value measure, and data coverage undefined.

A useful market definition follows this pattern:

Buyer universe × geography × category rule × procurement stage × event-date window × source coverage × retrieval cutoff

Examples:

• US federal prime-contract obligations for selected PSC groups, grouped by awarding sub-tier agency, using transaction action dates in fiscal year 2026 and data retrieved after the stated disclosure cutoff;
• EU-level competition notices for cybersecurity CPVs, counted at bid-ready lot grain, open at the final day of each month in a rolling 12-month window; or
• UK construction contracts entered into under the Procurement Act regime, using UK7 contract details notices, contract dates, and the latest accepted OCDS record as of the extraction timestamp.

Each example describes a different observable market. None should be silently combined with the others.

Separate the decision from the available fields

Start with what the team needs to decide:

• Which buyers show recurring demand in our category?
• Is the visible pipeline growing or merely publishing more amendments?
• Which suppliers have the broadest buyer relationships?
• How concentrated are signed awards within one buyer-category slice?
• Is a category dominated by one large vehicle or supported by repeat contracts?
• Which jurisdictions have enough classification, value, and identity coverage for a comparison?

Then decide whether the available data can answer it. A dataset that lacks bidder identities cannot calculate supplier win rate. A portal that publishes estimated opportunity values but not payments cannot measure realized spending. A source with incomplete supplier identifiers can still support a count, but its concentration result needs an identity-coverage denominator.

Do not confuse disclosed procurement with national market size

Public portals observe a legally and operationally defined publication universe. US federal contract-action reporting, TED publication, and UK notice rules cover different buyers, thresholds, stages, and exceptions. State and local US procurement is outside SAM.gov’s federal boundary; TED is strongest for EU-level regulated publication; and UK coverage depends on the notice regime, territory, threshold, and procedure.

If the question is total general-government procurement as an economic share, use national-accounts measures designed for that purpose. The OECD’s general-government procurement concept includes intermediate consumption, capital formation, and certain social transfers in kind; it is structurally different from adding portal award values. (OECD: size of public procurement)

Use portal analysis for observable opportunities, disclosed awards, contracts, suppliers, and transaction activity. Call it that.

Choose opportunity, award, or financial activity

Three analytical populations answer three different market questions.

Opportunity activity measures prospective demand

Opportunity data can include pipeline notices, market engagement, sources sought, planned procurement, solicitations, tender notices, amendments, cancellations, and direct-award transparency. Define which stages qualify.

Useful opportunity measures include:

• distinct new procedures published during a period;
• bid-ready lots published during a period;
• opportunities open at a stated cutoff;
• median remaining response time;
• reported estimated or maximum value, with its basis preserved;
• buyer-category recurrence; and
• the share of records amended, extended, cancelled, or unresolved.

Do not count every notice version as new demand. “Published during July” is a flow. “Open on 31 July” is a snapshot. A procedure first published in June and amended in July belongs in the July publication activity only if the metric explicitly counts amendments; it can still belong in the July month-end open snapshot.

Use the source-comparison guide to set the SAM.gov, TED, Find a Tender, and legacy Contracts Finder boundaries before comparing volume.

Award and signed-contract activity measures disclosed outcomes

An award decision and an executed contract can be separate stages. Open Contracting Data Standard guidance models awards and signed contracts separately because an award might not become a contract and the value, duration, items, or parties can differ. One process can contain several awards, and one award can relate to several contracts. (OCDS awards and contracts)

For TED, a result notice is a publication container. It can describe several lots, tenders, winners, and contracts. For current UK Procurement Act records, UK6 announces the award decision before signature, while UK7 follows contract entry. Count UK7 contract objects when the measure is signed contracts; retain UK6 as a separate intended-award stage. (Find a Tender notice types)

Useful outcome measures include:

• distinct awards and signed contracts;
• no-award or terminated procedures;
• named supplier relationships;
• reported award or contract value at a named grain;
• time from competition publication to decision or signature; and
• the share of opportunities with an evidence-backed outcome link.

Financial activity measures reported commitments or payments

US federal award analysis has a particularly useful transaction layer. USAspending describes a prime award summary as a roll-up of related transactions, while each base action or modification is a transaction. Its analyst guide distinguishes federal_action_obligation on a transaction from total_obligated_amount on a summary. Negative obligations are de-obligations and belong in the net history. (USAspending Analyst’s Guide)

For a period flow, a defensible measure is:

net obligations = sum(federal_action_obligation for transactions whose action dates fall in the window)

State that the result is net prime-contract obligations, not payments or contract value. Keep IDV ceilings and child-order obligations separate. Do not count modifications as new contracts. USAspending’s source guide also documents different submission cadences, incomplete linkage between financial and award files, and a 90-day public delay for DoD and US Army Corps of Engineers contract data. (USAspending data sources)

Set the analysis grain before counting rows

The same procurement can generate many legitimate rows:

1. one procedure;
2. several notice publications and versions;
3. several lots;
4. zero, one, or several awards per lot;
5. several contracts per award;
6. orders or calls under a vehicle;
7. base and modification transactions; and
8. one or more supplier relationships per award or contract.

The row grain is not an implementation detail. It determines the denominator.

Use separate fact tables for objects and relationships

A practical analysis model includes:

• a procedure table;
• a notice-version table;
• a lot table;
• an award table;
• a contract or order table;
• an action or transaction table;
• a party table; and
• award–supplier or contract–supplier relationship tables.

The award table can carry one aggregation-safe award value. The award–supplier table can carry every named winner. If the source does not allocate value by supplier, the relationship rows should not inherit the complete award value.

Keep source IDs and relationship evidence. In OCDS, award and contract IDs are scoped to the contracting process, or OCID, rather than being global. In US data, PIID, agency context, modification number, and parent-vehicle relationships matter. A globally unique key invented from a title is not a substitute.

Distinguish cohort and snapshot analysis

A cohort follows objects selected by an event in a period, such as contracts signed in 2025. A snapshot describes state at a cutoff, such as opportunities open on 31 July 2026. Revisions after the cutoff can change a later reconstruction.

For reproducibility, retain the source versions accepted at the original cutoff. If only the latest current record is available, label the analysis “reconstructed as of retrieval” rather than pretending it reproduces a historical snapshot.

The government contract tracker guide explains how to retain those versions and derive material events.

Measure buyer demand without inflating activity

Buyer analysis has two jobs: preserve institutional identity and measure demand at a consistent grain.

Resolve the buyer hierarchy

Group on codes and hierarchy relationships, not only display names. For US award data, distinguish:

• top-tier agency;
• sub-tier agency;
• awarding office where available;
• awarding agency, which creates or administers an award; and
• funding agency, which supplies the money.

USAspending uses the GSA hierarchy for award data and a different OMB hierarchy for account data. Its analyst guide notes that awarding and funding agencies are often—but not always—the same. Select one role for the primary ranking and expose the other as a separate dimension. (USAspending agency fields; SAM.gov Federal Hierarchy API)

For TED and UK data, retain the published organization identifier scheme and value, normalized name, jurisdiction, and parent relationship where supported. A renamed authority should not appear as a new buyer merely because its label changed. Conversely, two similar department names should not be merged without evidence.

Build a buyer-demand score from transparent measures

Avoid an opaque score until the component metrics are useful on their own. Start with:

• distinct bid-ready procedures or lots;
• open-at-cutoff pipeline;
• distinct awards or signed contracts;
• aggregation-safe reported value;
• US net obligations where applicable;
• median rather than only mean value;
• number of active categories;
• repeat-procurement intervals;
• amendment and cancellation rates; and
• value, category, and outcome-link completeness.

Rank each measure separately. If a combined score is necessary, publish its formula, scaling, caps, missing-value behavior, and version.

Measure recurrence without calling it intent

A buyer with eight comparable procedures over four years has an observed recurrence pattern. That pattern can support a research window, not a promised procurement.

Record the median and range between comparable publication or contract dates, then attach current contract-end, pipeline, forecast, or market-engagement evidence where available. Do not convert an IDV ceiling, an option end date, or annual seasonality into a guaranteed recompete.

Keep NAICS, PSC, and CPV analytically distinct

Classifications are not interchangeable category labels.

• NAICS is an industry classification. In US federal acquisition, the contracting officer assigns the code that best describes the acquisition’s principal purpose, and the code connects to small-business size standards. (FAR 19.102; US Census NAICS)
• PSC describes the predominant product, service, or research-and-development category bought on a federal contract action. Preserve the code edition because codes are added, changed, and end-dated. (Product and Service Code Manual)
• CPV is the EU’s hierarchical vocabulary for describing the subject of public procurement. It includes a main vocabulary and supplementary vocabulary; eForms can distinguish a main CPV from additional classifications. (TED Common Procurement Vocabulary)

Aggregate within each source scheme first

Roll codes up within their native hierarchy:

• NAICS sector, subsector, industry group, industry, and national industry;
• PSC portfolio, category, and individual code where the selected manual supports the relationship; and
• CPV division, group, class, category, and more specific code.

Retain the full source code beside the roll-up. Historical analysis should use the code vintage reported on the record, then provide a separate concorded field if the question requires a current taxonomy.

Treat crosswalks as analytical models

A NAICS–PSC–CPV crosswalk is normally many-to-many and lossy. Version the mapping, retain the original codes, record the mapping method, and publish the unmapped and ambiguous shares.

Do not assign a lot’s full value to every additional CPV. Use the main classification for a mutually exclusive value total, or treat additional classifications as non-additive tags. If a multi-category allocation is inferred, label the allocation method and include a sensitivity case.

Measure supplier competition with resolved identities

Supplier rankings are relationship analysis, not a GROUP BY supplier_name exercise.

Choose the entity level

Decide whether the analysis concerns:

• the legal entity named on the award;
• a consortium or joint venture;
• a registered branch;
• an operating subsidiary; or
• an ultimate corporate parent.

The US Unique Entity ID is a strong legal-entity key for entities registered in SAM.gov, but it does not automatically identify the complete corporate group. TED and Find a Tender can publish national-registry identifiers or other organization identifiers, but name-only records and source-scoped party IDs still occur. Never silently merge subsidiaries into a parent.

Publish legal-entity and parent-group views separately only when both are supported. Record the source identifiers, name normalization, match method, confidence, and effective dates behind a consolidation.

Use metrics whose denominator exists

Useful supplier measures include:

• distinct award or signed-contract wins;
• distinct buyer relationships;
• category and geography breadth;
• aggregation-safe value;
• median award or contract value;
• share of award count;
• share of usable value; and
• repeat relationships over a stated period.

Do not calculate a supplier win rate unless the denominator contains that supplier’s submitted or eligible bids. Public award data usually reveal winners, not every bidder’s participation across the same opportunity population. “Ten awards divided by 100 market opportunities” is not a 10% supplier win rate.

Keep shared award values out of supplier totals

If a €20 million framework names four suppliers without allocations, the market can report:

• one €20 million framework award;
• four award–supplier relationships; and
• €20 million of unallocated multi-supplier value.

It cannot report €20 million won by each supplier, €5 million won by each supplier, or €80 million of supplier awards without further evidence.

Calculate CR4 and HHI with defensible denominators

Concentration measures summarize the distribution of supplier shares inside a defined population.

• CR1 is the largest supplier’s share.
• CR4 is the sum of the four largest supplier shares.
• HHI is the sum of each supplier’s squared percentage share: HHI = Σ shareᵢ².

Using percentage shares gives an HHI approaching zero for many similarly small suppliers and 10,000 for one supplier with 100% of the defined market. The US Department of Justice uses HHI in merger analysis and publishes the formula and interpretive thresholds. Those thresholds presume a properly defined relevant antitrust market; a portal/category/year slice is not automatically one. (US DOJ: Herfindahl-Hirschman Index; DOJ market-share methodology)

Publish count-based and value-based concentration separately

Award-count share answers how frequently a supplier appears. Value share answers how much allocation-safe value it received. Obligation share answers how much net committed federal funding its transactions represent in the selected window. These can produce different rankings.

A supplier can win many small orders while another wins one large contract. Show both views and name the denominator in the chart title.

Run identity and allocation sensitivity cases

At minimum calculate:

1. legal entities only;
2. supported parent groups where available;
3. count-based concentration across resolved award–supplier relationships;
4. value-based concentration across allocation-safe records; and
5. lower and upper sensitivity cases for unresolved or shared records without pretending one is fact.

Do not combine all unknown suppliers into one entity: that artificially raises HHI. Do not simply discard them without showing the resolved share: that makes the result look more complete than it is.

Make time, currency, and value basis explicit

Select one event date for each measure

Possible dates include:

• notice publication or update date;
• response deadline;
• award decision date;
• contract signature or start date;
• transaction action date;
• performance end date; and
• first-observed or retrieval time.

Use half-open windows such as start ≤ event_date < end. They avoid overlap when adjacent monthly or annual periods are concatenated. Always publish the timezone and retrieval cutoff.

Do not compare an incomplete current period with a complete prior period. For recent sources, disclose publication lag and rerun the same cohort after the expected delay. Keep a frozen version when a result is used for a decision.

Retain native value before conversion

Each analytical value needs:

• native amount and currency;
• source field;
• value type, such as estimated, maximum, award, signed contract, obligation, or payment;
• tax basis where known;
• event date; and
• source grain.

Convert into a separate field. For event-level nominal analysis, use a documented rate tied to the award, signature, or action date. For annual flow comparison, a consistent period-average rate is often easier to reproduce. Retain native totals beside converted totals.

The European Central Bank publishes daily euro reference rates and methodology, but describes them as informational reference rates rather than transaction rates. Record the rate series, observation date, direction, and calculation version. (ECB exchange-rate methodology)

Currency conversion does not make unlike value bases comparable. A euro-denominated framework maximum and a dollar-denominated net obligation remain different measures after conversion.

Separate nominal, constant-currency, and real analysis

For a multi-year trend, label whether the series is:

• native-currency nominal;
• converted nominal using event-date or period-average FX;
• constant-currency using one reference period; or
• inflation-adjusted using a named price index.

Do not switch policies mid-series. Purchasing-power parity can support macroeconomic volume comparisons; it should not replace an exchange rate on individual contract records.

Compare markets without creating false equivalence

Cross-market analysis is safest as a staged comparison.

1. Publish native-source measures

Show US, EU, and UK populations separately with their sources, stages, thresholds, date fields, and coverage. Preserve the native currency and value basis.

2. Align lifecycle stages

Compare bid-ready lots with bid-ready lots, award decisions with award decisions, and signed contracts with signed contracts. Do not compare all SAM.gov notice types with TED competition notices or UK7 contracts.

3. Align the analytical unit

Choose procedure, lot, award, contract, or transaction. If a source cannot supply the chosen unit, mark the metric unavailable rather than substituting notice rows.

4. Align time and lag

Use the same event concept and complete periods. State the as-of timestamp and known late-reporting boundary.

5. Compare ratios before totals

Ratios can be more informative when source universes differ:

• supplier-identified relationship share;
• value completeness;
• median value within a compatible stage;
• top-four supplier share;
• buyer-category share;
• amendment or cancellation rate; and
• outcome-link coverage.

Even ratios require equivalent filters and denominators. A country with more complete low-value publication can still look different from one whose central portal primarily captures above-threshold records.

6. Use an index for trend shape

If the question is momentum rather than monetary scale, calculate an index within each market—such as the rolling 12-month distinct-procedure count divided by its own baseline. This compares direction while avoiding a claim that raw portal totals represent equal coverage.

The European Commission’s Public Procurement Data Space is designed to improve access and analysis across TED and national procurement data. Its existence does not remove current coverage, quality, and stage differences from a specific analysis. (European Commission: Public Procurement Data Space)

Publish the quality and coverage denominators

A ranking without its denominator is an assertion. A market analysis should ship with a coverage table.

Minimum coverage measures

Report by source, stage, period, and jurisdiction:

• source editions or partitions expected and received;
• distinct canonical objects;
• latest-version coverage;
• buyer-identifier completeness;
• supplier-identifier and resolved-identity coverage;
• classification completeness and crosswalk coverage;
• usable-value coverage by object count;
• unallocated multi-supplier value;
• opportunity-to-outcome link coverage;
• original-currency and tax-basis completeness;
• records excluded by validation rule; and
• source publication and ingestion lag.

Use both row counts and object counts. A missing value repeated over four supplier rows is still one missing award value.

Keep three denominators visible

For supplier value analysis, show:

1. total in-scope award or contract objects;
2. objects with one valid aggregation-safe value; and
3. value-bearing objects whose supplier allocation and identity are usable.

The first supports coverage. The second supports market-level value. The third supports supplier share. They should not be treated as one denominator.

Validate with invariants

Useful checks include:

• notice versions do not increase distinct procedure count;
• award–supplier rows do not increase distinct award value;
• child-order obligations are not added to parent ceilings;
• positive and negative transactions reconcile to the selected net obligation;
• each mutually exclusive category allocation sums to one object or 100%;
• supplier shares sum to 100% of the allocation-safe denominator;
• CR4 does not exceed 100%;
• HHI remains within the declared scale;
• original and converted values retain the same source object count; and
• period partitions are complete, non-overlapping, and reproducible.

Apply the web scraping data quality framework to completeness, validity, freshness, duplication, and source fidelity before treating the output as decision-ready.

Build a reproducible government contract analysis

1. Write the analysis contract

Record the question, scope, grain, stage, date field, cutoff, source list, value basis, entity level, classification rule, currency method, exclusions, and quality thresholds.

2. Acquire the native source objects

Use official search for exploration, APIs for bounded repeatable queries, and bulk data for history and reconciliation. Retain request filters, pages or cursors, raw files, checksums, and retrieval time.

SAM.gov Contract Awards exposes US procurement actions and modifications; USAspending exposes award summaries, transactions, recipients, agencies, classifications, and downloads; TED provides search and open-data routes; and Find a Tender exposes OCDS releases and compiled records. (SAM.gov Contract Awards API; USAspending endpoints; TED Search API; Find a Tender record packages)

3. Build source-native fact tables

Model each source before harmonizing. Preserve source grain, IDs, stages, relationships, codes, amounts, currencies, dates, and version semantics.

4. Canonicalize objects and relationships

Create stable internal keys without discarding source keys. Resolve buyer and supplier identities with method and confidence. Keep candidate and unresolved links visible.

5. Create mutually exclusive measures

Choose the one table and one value basis that answer each metric. Generate separate non-additive tags for secondary classifications and relationships.

6. Calculate coverage before rankings

Reject or qualify a measure if source, value, category, or identity coverage falls below the written acceptance rule. Do not discover the denominator problem after publishing the top-ten table.

7. Produce native and aligned outputs

Publish source-native results first. Add aligned cross-market measures only where the grain, stage, date, and value basis can be made comparable.

8. Run sensitivity cases

Test legal entity versus supported parent group, count versus value share, main category versus mapped portfolio, and exclusion versus bounded treatment of unresolved records.

9. Reconcile and freeze the release

Compare counts and samples with official interfaces or source totals where available. Retain the input manifests, transformation version, calculation code, coverage report, output checksums, and reviewer decision.

Choose official data, a normalized view, or a managed feed

Use official portals and downloads when the analysis is one-off, one jurisdiction is enough, and the team can model the native source. They remain the authoritative evidence.

Use a normalized view when the question recurs and the published scope, fields, history, and quality evidence are sufficient. WebTruffle’s free government tenders dataset includes opportunity market summaries and buyer-intelligence views across its stated US, EU, and UK sources. The free government contract awards dataset includes aggregation-safe award, buyer, supplier, and market views for its stated EU and UK scope. The award dataset does not claim US federal award coverage.

These are observed public editions, not complete worldwide market totals. Keep the release manifest, real retained window, source coverage, field completeness, and known gaps with the analysis.

Use a managed feed when the question requires different sources, longer history, organization resolution, proprietary category portfolios, outcome linking, alerting, service levels, or delivery into a warehouse, CRM, or product. The government procurement intelligence use case describes that next step without changing the official source’s authority.

Frequently asked questions