Public procurement demand: August 2–8, 2026.

Construction led the CPV-coded notice mix; manufacturing led the NAICS-coded mix. They are not one global ranking. From August 2 through August 8, 2026, WebTruffle first observed 12,880 normalized opportunity/amendment-stage records in its public procurement feed. Construction work represented 1,442 of 8,337 CPV records, or 17.30%. Manufacturing represented 2,751 of 4,371 NAICS records, or 62.94%.

The source mix explains the split. TED supplied 8,278 of the 8,337 CPV records, while every NAICS record came from SAM.gov. The percentages describe visible notice activity inside one dated, normalized release. They do not measure total procurement market size, unique procedures, contract value, or every tender that existed in the covered jurisdictions.

This report uses the hash-verified August 8 public release, its manifest, the seven-day change records, and the observation-matched seven-day history. The edition uses schema 2.0, was generated on August 9, and was analyzed for this article on August 12.

What the August 2–8 category snapshot shows

Five findings are defensible from the published evidence:

1. CPV construction was the largest visible category. CPV division 45 accounted for 17.30% of the CPV panel. Division 71, covering architectural, construction, engineering, and inspection services, added 10.30%. Because each record was assigned to one primary division, those two rows together represented 2,301 of 8,337 CPV records, or 27.60%.
2. The U.S. federal panel was more concentrated. NAICS manufacturing sectors 31–33 represented 62.94% of the NAICS panel. Construction added 13.93%. The five leading NAICS sectors accounted for 90.92% of the 4,371-record denominator.
3. Taxonomy is also a source boundary. The CPV panel was 99.29% TED and 0.71% Contracts Finder. The NAICS panel was entirely SAM.gov. Comparing their percentages as though they were competing categories in one market would mostly compare publication systems and classification rules.
4. Newly observed does not mean newly published. The new delta means WebTruffle first saw the stable record during this observation window. A backfill, delayed source response, or first successful collection can also create a new observation.
5. Counts are the useful measure in this release. Estimated value existed on only 3,618 of 12,880 selected records, or 28.09%, and those amounts spanned 15 currencies. A pooled spend chart would be less informative and more misleading than the count distribution.

The result is best used as a category-screening signal. It can tell a team where visible notice flow accumulated during a documented week and which categories deserve a narrower buyer, geography, deadline, or code-level review. It cannot tell a supplier the total addressable market without deeper history and a purpose-built market definition.

What counts as newly observed demand

“Demand” needs a written operational definition. In this report it means a stable normalized record that met all three conditions:

• its observation date fell from August 2 through August 8, 2026;
• its published delta type was new; and
• its normalized record stage was opportunity or amendment.

The result was 12,880 records: 12,879 opportunity-stage records and one amendment. The opportunity stage is broader than “bid-ready tender.” It includes solicitations, competition notices, presolicitations and, in this release, five TED pre-market consultation notices because of the schema mapping. A supplier still needs the source notice, status, deadline, procurement method, and documents before treating a row as an actionable bid.

The record grain matters too. One row is one stable normalized source record at the selected observation date. It is not necessarily one unique procurement procedure, lot, requirement, or future contract. TED can publish multiple notice versions and lifecycle forms around one procedure. U.S. notices can be related by notice or solicitation identifiers. This report does not collapse records by similar title or buyer name, because fuzzy consolidation would make the result less reproducible.

The selection deliberately excludes 1,171 newly observed planning records. Planning is a valuable forward signal, but it is not the same event as a competition notice. It is analyzed separately below.

Finally, new is a pipeline-relative state. The change product describes when WebTruffle first observed a stable record, not a legal assertion about when a contracting authority first published it. For a source-publication study, use comparable source timestamps and quantify collection latency separately.

Source mix comes before category rank

The 12,880-record denominator was composed of:

• TED: 8,278 records, or 64.27%;
• SAM.gov: 4,543, or 35.27%;
• Contracts Finder: 59, or 0.46%; and
• Find a Tender: zero in this selected schema 2.0 slice.

Find a Tender's zero is not evidence that the UK portal published no opportunities. The August data-quality benchmark found that the public schema 2.0 Find a Tender population was contract-heavy and contained no normalized opportunity-stage rows. The source itself supports tender-stage notices; the Find a Tender OCDS API guide explains how those releases are retrieved and classified. The zero therefore belongs in the article as a population and mapping limitation, not as a market conclusion.

The primary classification split was similarly source-conditioned:

• CPV: 8,337 records, or 64.73% of the selected flow;
• NAICS: 4,371, or 33.94%;
• PSC as the primary scheme: 143, or 1.11%; and
• source-native fallback: 29, or 0.23%.

CPV and NAICS answer different questions. The European Commission describes the Common Procurement Vocabulary as a system for standardizing references used to describe procurement contracts. The U.S. Census Bureau defines NAICS as a production-oriented classification of business establishments. NAICS therefore indicates the industry associated with a U.S. requirement; it is not a product taxonomy interchangeable with CPV.

CPV construction led the CPV-coded notice mix

Construction work was the largest CPV division with 1,442 records. Architectural, construction, engineering, and inspection services followed with 859. Medical equipment, pharmaceuticals, and personal care products ranked third with 762.

The next five visible divisions broadened the mix:

• sewage, refuse, cleaning, and environmental services: 569;
• transport equipment and auxiliary products: 479;
• business services including law, marketing, consulting, recruitment, printing, and security: 409;
• IT services: 385; and
• repair and maintenance services: 369.

The leading five divisions represented 4,111 records, or 49.31% of the CPV denominator. That leaves just over half of the panel distributed across the remaining divisions. The headline is therefore not “CPV demand was only construction.” It is that construction and engineering were the strongest visible cluster inside a still-diverse week of CPV-coded notices.

This panel should not be described as a balanced Europe-and-UK estimate. TED supplied 8,278 of its 8,337 records. Contracts Finder supplied 59. A source-weighted distribution can be perfectly calculated and still be unrepresentative of every jurisdiction. For source coverage, thresholds, and publication routes, use the SAM.gov, TED, and UK source comparison.

For the connector behind a source-native TED analysis, use the TED Search API v3 guide. It covers complete iteration, versions, multilingual fields, raw XML, OJ S packages, and the validation boundary between notice projections and lot-level evidence.

The analysis uses the first normalized primary CPV division per record. Additional CPV codes remain useful tags for discovery, but counting every code would let one notice contribute to several categories and make shares exceed 100%. A mutually exclusive primary division is the correct denominator for this ranking.

NAICS manufacturing led the U.S. federal mix

Manufacturing represented 2,751 of 4,371 NAICS records, or 62.94%. This report combines codes 31, 32, and 33 into the official Manufacturing sector family. Treating those prefixes as three unrelated sectors would mechanically split one Census sector and distort the ranking.

Construction followed with 609 records, or 13.93%. Professional, Scientific, and Technical Services contributed 225; Administrative and Support and Waste Management and Remediation Services contributed 221; and Other Services except Public Administration contributed 168.

Together, Manufacturing and Construction accounted for 3,360 records, or 76.87% of the NAICS panel. The first five sectors accounted for 3,974, or 90.92%. That concentration is analytically useful, but it must be interpreted through the source: all 4,371 records came from SAM.gov.

The Manufacturing share does not mean that manufactured goods represented 62.94% of all public procurement value, nor that 62.94% of the requirements were completed purchases. NAICS groups establishments by production process. A notice's assigned industry is an important competition and supplier-qualification signal, but the exact product or service may be clearer in its PSC, statement of work, title, or attachments.

For monitoring, the two-digit sector is only the first cut. A supplier should continue from Manufacturing into the applicable three- through six-digit NAICS hierarchy, then add agency, set-aside, geography, notice type, and deadline rules. The broad sector finds the territory; the detailed code defines the practical watchlist.

PSC adds a product and service lens

PSC is the federal product-and-service vocabulary, and its official manual library should be retained with the code version. The analyzed release uses PSC April 2025 group labels. PSC complements NAICS: one describes the predominant product, service, or research-and-development category, while the other describes the industry associated with the requirement.

Only 143 selected records used PSC as the primary normalized scheme, so a primary-PSC league table would have a fragile denominator. Ten of those records also lacked a mapped division label in the public schema. This report therefore does not place a tiny PSC ranking beside the much larger CPV and NAICS panels as though the evidence were equivalent.

PSC still adds a useful companion view. Among the 4,543 SAM.gov opportunity/amendment-stage records, 3,152, or 69.38%, carried at least one normalized PSC code. The companion aggregation uses the first normalized PSC code per record, mapping alphabetic service codes to their letter-family group and numeric product codes to their first two-digit group. The release mapped 3,051 of those records to a group label; 101 remained unlabeled at that level. Among the mapped groups, the most common were maintenance, repair, or alteration of real property (322); maintenance, repair, and rebuilding of equipment (321); construction of structures and facilities (171); instruments and laboratory equipment (164); and utilities and housekeeping services (133).

That companion view overlaps the NAICS panel by design. Do not add its counts to the NAICS counts. Use NAICS to organize the industry universe and PSC to refine what the federal buyer is seeking. Preserve both original codes and the manual versions rather than forcing a lossy one-to-one crosswalk.

Compare the seven-day flow with the rolling baseline

A one-week result can be a spike. To test whether the leaders were peculiar to August 2–8, we also calculated the same primary-division distribution over the release's rolling latest-record population, limited to opportunity and amendment stages.

The rolling opportunity/amendment-stage population contained 49,362 records observed from July 14 through August 8. Its CPV denominator was 31,641 and its NAICS denominator was 17,160.

The leading CPV shares were close across the two cuts:

• Construction work: 17.30% in the seven-day flow versus 17.93% in the rolling baseline;
• architectural, construction, engineering, and inspection services: 10.30% versus 10.45%;
• medical equipment and pharmaceuticals: 9.14% versus 8.87%;
• sewage, refuse, cleaning, and environmental services: 6.83% versus 6.79%; and
• transport equipment: 5.75% versus 5.24%.

The leading NAICS sectors were also directionally consistent, although the seven-day U.S. flow was more manufacturing-heavy. Manufacturing was 62.94% for the week and 60.06% in the rolling baseline. Construction was 13.93% versus 15.37%. Professional, Scientific, and Technical Services was 5.15% versus 5.41%.

This comparison is a stability check, not a trend. The retained history spans only 26 observed days, and the rolling population is a latest-record snapshot rather than a complete multi-year event series. A trend claim needs repeated comparable windows, stable mappings, and an explicit test for changing source composition.

Planning notices are a separate forward signal

Adding planning stages to the selection produces 14,051 newly observed pre-award records. The 1,171 planning records were split across SAM.gov (886), TED (281), and Contracts Finder (4). They represented 8.33% of the broader pre-award flow.

Planning notices can reveal future direction before a solicitation exists. They can include market engagement, prior information, sources-sought, or other early-stage signals depending on the source. They are useful for account planning and category research, but they should not be presented as open competitions.

A practical tracker should maintain at least two lanes:

1. Forward signal: planning and market-engagement records that may become a requirement.
2. Competition signal: opportunity and amendment records that merit notice-level qualification.

Then link the lanes through official identifiers when the source provides them. Do not assume a similar title proves that a planning notice became a particular solicitation. The government contract tracker guide explains how to keep state, versions, deadlines, and lifecycle links explicit.

Record counts are not procurement spending

Only 3,618 of the 12,880 selected records had a normalized estimated value. TED supplied 3,574 of those values, Contracts Finder 44, and SAM.gov none. All populated amounts used the tender estimate basis, but they appeared in 15 currencies.

Three problems prevent a defensible pooled spend ranking:

1. Coverage differs by source. Value coverage was 43.18% for TED, 74.58% for the small Contracts Finder slice, and zero for SAM.gov. A value ranking would systematically exclude the entire NAICS panel.
2. Currency is not normalized economic value. Adding EUR, GBP, CZK, RON, PLN, and other currencies without dated exchange rates creates a meaningless total. Converting them adds exchange-rate timing and policy choices that this snapshot does not make.
3. A notice estimate is not spend. Estimated procedure value, lot value, maximum framework value, awarded value, contract value, obligation, and payment are different measures at different grains.

Record counts have limits too. A high count can reflect many small requirements, multiple lots or forms, active publication practice, or several records around one procedure. But unlike a sparsely populated cross-currency total, the count measure has a complete and auditable denominator for this selected flow.

For a market-sizing project, start with the government contract analysis framework: define the buyer universe, geography, category rule, lifecycle stage, event-date window, source coverage, and retrieval cutoff before selecting a measure.

Turn the snapshot into a market monitor

The broad distribution is a starting point. A useful category monitor narrows it in five steps.

1. Choose the source-native vocabulary

Use CPV for the TED and UK path, NAICS plus PSC for U.S. federal opportunities, and retain any additional source-native scheme. Do not begin with a universal category label that hides the source code.

2. Move below the broad division

Construction work and Manufacturing are too broad for most sales or research teams. Select the applicable detailed codes, publish the inclusion list, and version it. If a portfolio uses several codes, keep the mapping many-to-many and report unmapped records.

3. Add buyer, geography, and notice-stage rules

A category can contain very different buying motions across agencies and places. Segment by authoritative buyer identity where available, jurisdiction, procedure or notice type, set-aside, and stage. Keep planning separate from competition and award outcomes.

4. Add qualification evidence

Attach the official source link, deadline type, documents, procurement method, description, and last-observed state. In this selected flow, a normalized opportunity stage is not itself proof that a bidder can still respond.

5. Measure recurring flow

Track newly observed, materially changed, cancelled, closed, awarded, and still-actionable records as separate measures. Compare equivalent weeks only after checking source health and schema version. Publish both the numerator and the population that could have qualified.

The free government tenders and RFP dataset exposes the daily, current-candidate, change, history, and governance products needed for that workflow. A tailored feed becomes useful when the category definition, sources, history, enrichment, or delivery rules need to differ from the public edition.

Reproduce the analysis

For a tested manifest-verification and DuckDB workflow over the same public release, begin with the government tender data with Python guide. The steps below retain the exact category-selection logic used for this report.

The exact release pattern is:

https://github.com/webtruffle/government-tenders-rfps/releases/download/2026-08-08/{artifact}

The three core artifacts independently matched these SHA-256 hashes:

manifest.json    cbbe78fe20b38cb94ebbf21144638207ba796f2b546eee2f4785b51565a9d2b1
changes.csv      e3ea6a768c55f9ff904b8d8c6eb52a72339aaee56ed0d81158377cd5056587ae
history-7d.csv   565eadab9a7ec7bb877c2c6d6ac671f85586d4ea6afea8f0244cab40337f676e

The manifest is 33,479 bytes, changes.csv is 17,912,455 bytes, and history-7d.csv is 59,342,190 bytes. The manifest declares the artifact hashes and grains; verify it before calculating.

This standard-library Python reproduces the selection and primary category counts after the two CSV files are downloaded:

import csv
from collections import Counter

selected = set()
with open("changes.csv", encoding="utf-8-sig", newline="") as source:
    for row in csv.DictReader(source):
        if (
            row["delta_type"] == "new"
            and row["record_stage"] in {"opportunity", "amendment"}
        ):
            selected.add((row["id"], row["observation_date"]))

naics_families = {
    "31": "31-33",
    "32": "31-33",
    "33": "31-33",
    "44": "44-45",
    "45": "44-45",
    "48": "48-49",
    "49": "48-49",
}

counts = Counter()
matched = 0
with open("history-7d.csv", encoding="utf-8-sig", newline="") as source:
    for row in csv.DictReader(source):
        key = (row["id"], row["observation_date"])
        if key not in selected:
            continue

        matched += 1
        scheme = row["classification_primary_scheme"]
        division = row["classification_division"]
        label = row["classification_division_label"]
        if scheme == "NAICS":
            division = naics_families.get(division, division)
            if division == "31-33":
                label = "Manufacturing"
            elif division == "44-45":
                label = "Retail Trade"
            elif division == "48-49":
                label = "Transportation and Warehousing"

        counts[(scheme, division, label)] += 1

assert matched == len(selected) == 12_880
for key, records in counts.most_common():
    print(*key, records, sep="\t")

The observation-date join is important. Joining only by stable ID to a later current-candidate file could attach a later state or omit a record that closed before the cutoff. All 12,880 selected (id, observation_date) keys matched exactly one enriched history row.

Suggested citation:

WebTruffle. “Public Procurement Demand by Category: August 2026 Snapshot.” Published August 12, 2026. Analysis of 12,880 newly observed opportunity/amendment-stage records in the government-tenders-rfps schema 2.0 release dated August 8, 2026.

Include the release date, schema version, filter, and category denominator when citing a percentage. “Construction was 17.30%” is incomplete without “of 8,337 CPV-coded selected records.”

Limitations and what not to infer

This report has explicit boundaries:

• It measures records observed by one normalized pipeline, not recall against an official universe of everything that should have been published.
• The seven-day window is August 2–8, but new means first observed, not necessarily first published by the source.
• The selected stage can contain notices that are not bid-ready. Source status, notice type, deadline, and documents still govern actionability.
• A source record is not automatically a unique procedure, lot, purchase, or future contract.
• CPV, NAICS, and PSC are kept separate. Their distributions are also proxies for source and geography in this edition.
• Find a Tender contributed no normalized opportunity/amendment-stage records under public schema 2.0. That is a known population and mapping boundary, not a claim of no UK market activity.
• The ranking uses one primary classification division per record. Additional codes are not counted in the mutually exclusive share.
• Classification presence does not prove that the publisher selected the most analytically useful code.
• The public retained history began July 14, so the rolling comparison is a short baseline, not a seasonal or annual trend.
• Value coverage, basis, currencies, and object grains do not support a pooled spending estimate.
• A later schema, source correction, or backfill can change the result. Dated reproducibility is the protection against silently moving claims.

The companion data-quality benchmark documents the release's source health, stage mix, field presence, and schema 2.0 mapping gaps. Read that evidence before generalizing this category snapshot.

Frequently asked questions