Refolk
PlaybookMarket and talent intelligence

The Public-Sector Award Share Playbook: Award Records to a Dated Share Table

Go from one product category and one jurisdiction to a vendor share table by obligated value and agency count that holds up line by line against the award IDs.

21 min readLast reviewed November 17, 2026Read as Markdown

Key takeaways

  • Sum obligations grouped by contract_award_unique_key and never sum base_and_all_options, because a ceiling is not spend: four $200M DoD AI agreements had $45.5M obligated while USAspending showed $0.
  • End every award-share window at least 90 days before the pull date, because Department of Defense and Army Corps of Engineers contract data reach USAspending 90 days late.
  • Federal IT product and service codes classify how software was bought, not what it is: DA01 is labor, DA10 is as-a-service and 7A21 is perpetual license, so a category needs written description rules as well as codes.
  • In Refolk's index of 778 US profiles titled Federal, Public Sector or SLED Account Executive, the top employers in a 25-profile sample are product vendors and no reseller appears in the top 10, which is why the award recipient is often not the vendor whose product was sold.
  • Every obligated dollar in the table belongs either to a named product vendor or to an explicit unattributed row, never to a reseller like Carahsoft acting as seller of record.
  • Share by buying-agency count swings more than share by value: one third-party analysis reports awards under $15,000 are 0.7% of reported dollars but 83.7% of award count (method unverified).

This playbook turns public contract award records into a vendor share table for one product category in one jurisdiction. It is for strategy and research teams, talent-intelligence analysts, and operators who need competitor market share from government contracts but cannot rely on vendor claims or pay for analyst reports. You end with a dated table that shows share by obligated value and by buying-agency count, includes an unattributed row, states its coverage gap, and can be defended line by line against the award IDs.

What does a defensible award-share number measure?

An award-share number measures each vendor's portion of obligated dollars, and of distinct buying agencies, across the reported award records for one category, in one jurisdiction, over a fixed window with a stated as-of date. It does not measure total government spend, contract ceilings, or a vendor's commercial market share.

The narrow scope is deliberate. An award record is public, countable and tied to an identifier, so every cell in your table traces back to a row someone else can pull and check. Case studies, install counts and review volumes cannot be audited that way.

Terms used throughout, each defined once:

  • Obligation: money the government has legally committed to pay on an award. This is the numerator of every share.
  • Ceiling: the most an award could be worth if every option is exercised. It is not spend.
  • IDV (indefinite delivery vehicle): a contract under which agencies place separate orders. Schedule contracts and blanket purchase agreements (BPAs) are examples.
  • Task or delivery order: an order placed under an IDV, carrying its own obligations.
  • Modification: a change to an existing award, reported as its own transaction.
  • PSC and NAICS: the product and service code and the industry classification code attached to a federal award.
  • UEI (unique entity identifier): the recipient's identifier in the federal system.

What it proves: a named agency committed a stated amount to a named recipient, for something the record describes. When it lies: the recipient is a reseller, the ceiling was read as spend, recent defense data had not yet been published, or the purchase went on a card and never reached the award database. What it cannot tell you: whether public-sector share tracks commercial share. I have found no public evidence on that relationship, so do not scale one into the other.

From award records to a share table

  1. Fix scope
    One jurisdiction, one window ending 90+ days before the pull, one as-of date
  2. Define category
    Codes pull the candidates; written description rules decide what counts
  3. Pull and collapse
    Transactions grouped by award key, obligations summed, ceilings dropped
  4. Attribute
    Reseller-held dollars traced to the product vendor or left unattributed
  5. Publish with gaps
    Share by value and agency count, plus coverage lines
Every share figure in the final table traces back through attribution and collapsing to a set of raw transactions pulled on a recorded date.

Which award records should I pull, and how fresh are they?

In the US, pull federal contract transactions from USAspending, which publishes the data contracting officials submit to FPDS. In the EU and UK, use the award notices published on TED and Find a Tender. Each system has its own reporting floor and publication lag, and both limit what your table can see.

FPDS is the system where federal contracting officials submit transaction-level data for contracts and IDVs. Agencies submit within about three business days of the award transaction, and USAspending publishes about two days later. Defense data is the exception: Department of Defense and U.S. Army Corps of Engineers submissions are delayed by 90 days. The FPDS.gov public website and its ezSearch tool have been shut down, and SAM.gov is now the official home for federal contract award data. If an internal process still points at FPDS.gov, update it.

SystemReporting floorPublication timing
USAspending / FPDS (civilian)$15,000 micro-purchase threshold~3 business days + ~2 days
USAspending / FPDS (DoD, USACE)$15,000+90 days
EU TED€140k central / €216k sub-central≤30 days after conclusion
UK Find a Tender (below-threshold)£12k central / £30k other30 / 90 / 120 days

In the EU, a contract award notice must be sent for publication no later than 30 days after the contract or framework agreement is concluded. In the UK, under the Procurement Act 2023, a Contract Details Notice is due within 30 days of the contract being entered into, or 120 days for a light touch contract. The full set of deadlines is 30, 90 or 120 days depending on the buyer and contract type. All of these floors are revised from time to time, so re-check them before each pull (see the last section).

For a US pull, these are the fields that matter. The USAspending Data Dictionary gives the official definitions.

FieldUse it forNever use it for
total_obligationThe spend numeratorMixing with ceiling values
base_and_all_optionsContext on deal sizeAny sum or share
piid, parent IDV PIIDLinking orders to their vehicleDeduplicating on its own
Recipient UEIIdentifying the seller of recordAssuming it is the product vendor
PSC, NAICSFirst-pass category filterDefining the category alone
DescriptionProduct and manufacturer namesTrusting without a sample check

The documentation describes base_and_all_options as "the ceiling and amount of the award", which explains how it ends up in sums by mistake. The federal data standard also includes an Ultimate Parent Unique Identifier, which lets you roll subsidiaries up to their parent company before you rank vendors.

Known gap: I have not checked the field lists for TED, Find a Tender or any state or regional portal, including their value fields, CPV codes and supplier IDs. Before you scope a non-US table, open a handful of award notices and confirm three things: a value field, a classification code, and a stable supplier identifier. If there is no stable supplier ID, you will need an extra entity-resolution step before step 4.

The procedure, step by step

Eight steps take roughly four to eight analyst days for one category in one jurisdiction. These durations are my desk estimates, not measured benchmarks. Almost all of the spread comes from steps 5 and 6.

Award records to a dated vendor share table

  1. Fix jurisdiction, window and as-of date
    Pick one jurisdiction and a fiscal-year window that ends at least 90 days before the pull date, then record the as-of date. About an hour.
  2. Write the category definition
    Choose the PSC set, the NAICS set, the description terms and the product names to include and exclude, with legacy codes mapped. Done when the include and exclude list is written down; half a day to a day.
  3. Pull the transactions
    Pull the code set for the window from the USAspending bulk download or the transaction API, where each modification arrives as its own row. Done when the raw file is saved with its as-of date; about half a day.
  4. Collapse to one row per award key
    Group by contract_award_unique_key, sum the obligations, and keep ceiling fields out of every calculation. Done when there is one row per key and the totals reconcile; about half a day.
  5. Hand-label a sample and state precision
    Label a random sample of matched awards as in or out, then compare the code-only filter with the code-plus-description filter. Done when both precision figures are written down; one to two days.
  6. Attribute reseller-held awards
    Match each reseller's UEI and parent schedule PIID against its published manufacturer and dealer lists, then read the descriptions. Done when every dollar is attributed or sits in the unattributed row; one to three days.
  7. Build the share table
    Compute share by obligated value and by count of distinct buying agencies, keeping the unattributed row visible. About half a day.
  8. State the coverage gaps
    List card spend, below-threshold buys and the DoD lag as separate lines under the table, with the as-of date. About an hour.

How do I define the category with codes and description terms?

Define the category in two layers. A PSC and NAICS set pulls the candidate awards, and written include and exclude rules (product names and description phrases) decide which candidates count. Codes alone fail because federal IT PSCs classify how something was bought, not what product it is.

GSA's guide to finding PSC codes maps "Application" to three codes:

  • DA01: labor
  • DA10: as-a-service
  • 7A21 (Business): perpetual license

This has two consequences. A product sold both as SaaS and as a perpetual license lands in two codes, so a single-code pull undercounts it. And 7A21 holds every business application, so the code alone cannot separate one kind of business software from another. Description matching is therefore unavoidable.

Code drift. The October 2020 PSC revision end-dated 40 IT service codes and 30 IT product codes and activated 23 and 17 new ones. 7A21, for example, replaced legacy code 7030. If your window crosses that revision, map the legacy codes to the current ones before comparing periods. Otherwise the series shows a collapse that is really a renaming.

Description quality. USAspending publishes an "Award Description Data Quality" disclosure, which tells you descriptions are not uniform. No published precision or recall figures exist for code-based or keyword-based category filters, so there is no benchmark to borrow and you have to measure your own. Hand-label a random sample of matched awards, sized so you can read every row in a day. Record how many the code-only filter got right and how many the code-plus-description filter got right. A filter that is precise but too narrow is a recall problem. To find what it misses, search your product names across awards outside the code set and log the misses.

Category definition sheet
Category:
Jurisdiction:
Window: (ends at least 90 days before the pull date)
As-of date:
PSC include: DA01, DA10, 7A21
PSC legacy mapping: 7030 -> 7A21 (records before the October 2020 revision)
NAICS include:
Description include terms:
Product names, by vendor:
Description exclude terms:
Known resellers to attribute (name, UEI, schedule PIID):
Precision, code filter only: __% on n = __ labeled awards
Precision, code + description rules: __% on n = __ labeled awards
Owner:
Last changed (date, reason):

Fill every line before the pull. Change a line only with a dated note explaining why.

How do I collapse modifications and task orders so nothing is counted twice?

Group the transactions by contract_award_unique_key and sum the obligations within each group. Never key on PIID alone, and never sum base_and_all_options. You are done when there is one row per award key and the totals reconcile to the transaction file.

The transaction API returns one row per modification. A modification number of zero ("0") marks the initial award, and every later number is a change to that award. A task or delivery order carries a parent award ID that points to the contract it was placed under. USAspending's prime award summary rolls up all transactions that share the identifiers making up the unique award key, which has this format:

CONT_AWD_{piid}_{agency_toptier_code}_{parent_piid}_{parent_agency_code}

PIID alone is not unique. The same PIID can appear at different agencies and in different fiscal years, so deduplicating on PIID silently merges unrelated awards.

IDV-level rows. An IDV can contain many contract award summaries and even other IDVs. Excluding IDV-level rows when their child orders are already counted is good practice, but no official rule requires it. Check your own pull instead of assuming. If the IDV rows carry obligations that repeat their children's, exclude them. If they carry none, they are harmless. Either way, write down what you did.

Three reconciliation checks before moving on:

  • The summed obligations after collapsing equal the summed obligations before.
  • The row count after collapsing equals the count of distinct award keys.
  • A dedupe on PIID alone gives fewer rows than a dedupe on the full key, and the difference is the number of collisions you avoided merging. Log it.

How do I trace reseller-held awards back to the product vendor?

Treat the award recipient as the seller of record, not the owner of the share. Attribute each reseller-held dollar using the manufacturer or product name in the description, the parent schedule PIID, and the reseller's published manufacturer and dealer lists. Anything that does not resolve goes in an unattributed row and is never credited to the reseller.

The documented case is Carahsoft. It describes itself as the Master Government Aggregator for many of its vendors, managing their public sector reseller networks. It publishes a list of manufacturers on its schedule, including Microsoft, Salesforce, ServiceNow and Splunk, and a separate list of authorized dealers. In one DHS justification, it was named the principal distributor for much of the software DHS already buys. DHS also signed a five-year, $2.7B BPA on Carahsoft's GSA schedule, and that BPA admits manufacturers only where Carahsoft is their exclusive schedule distributor. The contract vehicles themselves concentrate many vendors' sales under one UEI. Without attribution, that UEI takes their share.

No one has published what portion of government software spend runs through resellers, so there is no correction factor to apply. You attribute award by award.

Attribution evidence, outermost first

  1. Recipient of record
    The reseller's UEI on the award, which says who invoiced, not whose product shipped
  2. Contract vehicle
    The parent schedule or BPA PIID, which narrows the possible manufacturers
  3. Reseller's published lists
    Manufacturers on the schedule and authorized dealers
  4. Description text
    A manufacturer or product name that settles the attribution
  5. Product vendor
    Credited in the table, or the dollars stay unattributed
The award record names the outer layer, and each inner layer either confirms the product vendor or sends the dollars to the unattributed row.
The award recipient is the seller of record, not the owner of the share.

There is a people-side signal that explains why this matters. In Refolk's index of professional profiles, product vendors staff government selling, not resellers.

778
US profiles in Refolk's index titled Federal, Public Sector or SLED Account Executive
In a 25-profile sample from this set, the top employers were product vendors (Abnormal AI 4, Wiz 3, then Anthropic, Google, Okta, Nutanix and others at 1 each), and no reseller appeared in the top 10.
Title (sample of 25 from 778)CountShare
Federal AE1768%
Public Sector AE624%
SLED AE14%
Public Sector Major AE14%

Five of the 25 sampled profiles (20%) are in the Washington DC area. The likely reading is that vendors run their own federal sales teams while resellers hold the contract, which is why the award recipient is so often not the vendor. Use this as a cross-check, not as evidence of share. If a vendor has a visible federal sales team but little award value under its own UEI, search the reseller-held awards for its product names before calling its share small. The limits are real: this is a 25-profile, title-based sample, not a count of headcount per vendor.

When descriptions are vague and a schedule covers dozens of manufacturers, the people who know which product an order covered are the reseller's public-sector sales directors and the vendor's federal reps. Finding the right ones is usually the slow part, and Refolk handles it from a plain-English request. Phrase the request by title and employer, not headline keywords. In Refolk's index, a search for US capture managers and procurement analysts with "government federal public sector" in the headline returned zero, and the UK version did too. That result reflects how the query was phrased, not the size of the market.

A conversation corroborates an attribution but does not replace the record. The table still cites award IDs. Mark any attribution that rests on a conversation so a reviewer can tell it apart from one the description proves.

How does award-share analysis go wrong?

It goes wrong in eight recurring ways, and each one produces a number that looks precise and is wrong. Four are counting errors and four are coverage errors. Run every check before you publish.

Counting errors

  • Ceiling counted as spend. Looks like a $200M agreement credited as $200M of share when only part of it is obligated. Check: no base_and_all_options column appears in any sum.
  • Modification double count. Looks like a vendor whose busy contract (many modifications) inflates its value several times over. Check: the row count equals the count of distinct award keys after collapsing.
  • Keying on PIID alone. Looks like one vendor absorbing an unrelated award from another agency or year. Check: dedupe on the full key and compare row counts with a PIID-only dedupe.
  • Reseller credited with share. Looks like Carahsoft appearing as a top "vendor" when it is selling Microsoft or Salesforce. Check: compare each reseller-held award against the reseller's published manufacturer list. Anything unmatched goes in the unattributed row.

Coverage errors

  • Lag read as decline. Looks like recent DoD quarters at or near zero. Check: the window ends at least 90 days before the pull date.
  • PSC drift. Looks like a category dropping off a cliff when legacy codes such as 7030 stop being used. Check: legacy codes are mapped to current ones for records before the October 2020 revision.
  • Bad descriptions. Looks like a keyword filter that confidently includes the wrong products. Check: keyword matches are scored against the hand-labeled sample, and precision is stated.
  • Card spend ignored. Looks like a seat-based SaaS vendor that seems absent from government. Check: card spend is stated as its own coverage line. Small seat purchases on purchase cards never reach FPDS.
$45.5M
Obligated on four DoD AI agreements worth $200M each, while USAspending showed $0
The headline figure was ceiling, the real figure was obligation, and the public record had not caught up because of the reporting lag.

How do I build, read and publish the share table?

Publish two share columns, one by obligated value and one by buying-agency count, plus an unattributed row and a coverage block, all stamped with the window and the as-of date. The two columns answer different questions, and the gap between them is often the most useful finding.

Share by value is a vendor's obligated dollars divided by total obligated dollars in the category. Share by agency count is the percentage of buying agencies that bought from the vendor at least once. One agency can buy from several vendors, so this column does not sum to 100%. Say so in a footnote.

The two views diverge sharply. One third-party analysis reports that awards under $15,000 make up 0.7% of reported dollars but 83.7% of award count, because many small orders sit alongside a few very large ones. The period and method of that analysis are unverified, but the pattern explains why share by agency count swings more than share by value.

Reading a vendor's two share columns

High share by obligated valueLow share by obligated value
Anchor-deal holder
Name the one or two awards driving it and check them for ceiling errors
Broad incumbent
A defensible lead, but confirm none of its dollars came through a reseller
Marginal or misattributed
Search reseller-held awards for its product names before calling it small
Long-tail seller
Most exposed to the card-spend gap, so flag it in the coverage block
Low share by agency countHigh share by agency count
Where a vendor sits on value share and agency-count share tells you which check to run before you trust its row.

Coverage block. State the gap without estimating it away. No one has published what portion of a given category's spend falls below the reporting thresholds. Purchase card data is not reported into FPDS; GSA publishes it separately as SmartPay statistics.

Below-threshold measure (US)Value
SmartPay spend FY2025 (purchase, travel, fleet)$39.4B
Average SmartPay transaction FY2025$480
Purchase card spend FY19$22B
Electronics merchant-code share of FY19 card spend5.7%
Awards under $15k: share of reported dollars0.7% (unverified)

These figures give a sense of scale, not a correction factor. The card totals cover all categories, and the electronics merchant codes do not map to any one software category.

Vendor share table, publishable skeleton
Category: ______ | Jurisdiction: ______ | Window: ______ to ______ | As-of: ______
Method: transactions pulled from USAspending, collapsed by contract_award_unique_key, obligations only

Vendor | Obligated value | Share by value | Buying agencies | Share of agencies*
Vendor A | | | |
Vendor B | | | |
Vendor C | | | |
Unattributed (reseller-held, unresolved) | | | |
Total | | 100% | |

* Share of agencies = % of buying agencies that bought from the vendor. Does not sum to 100%.

Coverage and limits
- Purchase card spend: not in FPDS, not included
- Awards below the micro-purchase threshold: not reported, not included
- DoD / USACE: window ends __ days before pull to clear the 90-day delay
- Category precision: __% (code + description rules) on n = __ labeled awards
- Unattributed: __% of obligated value
- Award key list: attached, one row per contract_award_unique_key

Keep the unattributed row even when it is small, and never delete a line from the coverage block.

Before the table ships

  • The window ends at least 90 days before the pull date and the as-of date is printed on the table.
  • No base_and_all_options value appears in any sum.
  • Row count equals distinct contract_award_unique_key count, and collapsed obligations equal transaction obligations.
  • Legacy PSCs are mapped for records before the October 2020 revision.
  • Precision of the category filter is stated with its sample size.
  • No reseller UEI appears as a vendor row; its dollars are attributed or unattributed.
  • Attributions that rest on conversations are marked separately from those proved by descriptions.
  • Card spend, below-threshold buys and the DoD lag each have their own coverage line.
  • The award key list is attached so any cell can be traced back to its rows.

How do I keep the share table current?

Re-pull on a fixed cadence with a rolling window that always ends at least 90 days before the pull date. Re-check every moving reference (thresholds, codes, portals and reseller lists) before each run. Keep each published version and compare it with the next, rather than overwriting it.

Diff by award key between versions. New keys are new business. A changed obligation on an existing key is a modification, which can add money or remove it. A vendor's share moving because of a single re-pulled key is a footnote, not a trend.

Re-check these before each run, because each one has changed before:

  • Thresholds. The US micro-purchase threshold rose from $10,000 to $15,000, and the simplified acquisition threshold is $350,000. Confirm both on the Acquisition.gov threshold changes page. EU thresholds are set by delegated regulation for a two-year period. UK floors come from the Procurement Act 2023.
  • Codes. Check the current PSC manual on Acquisition.gov for end-dated or newly activated codes, and update the legacy mapping line on your definition sheet.
  • Portals. The move from FPDS.gov to SAM.gov shows that access points change. Confirm your pull route still works and still returns the same fields.
  • Reseller lists. Download the manufacturer and dealer lists again. A manufacturer added to a schedule changes which awards you can attribute.
  • Method notes. The USAspending Analyst's Guide and the public usaspending-api repository on GitHub document field and endpoint changes.

If you want a second reviewer who has worked with these files, ask Refolk for government contracts analysts who list USAspending or FPDS data work. One hour of their time on your collapse and attribution logic will catch more errors than another day of rereading your own spreadsheet.

Questions practitioners ask

Can I use contract ceiling values to estimate a competitor's government market share?

No. A ceiling is the most an award could reach if every option is exercised. It is not money spent. Four DoD AI agreements worth $200M each had $45.5M obligated between them, and USAspending showed $0 because of the reporting lag. Base share on summed obligations only, and confirm that no base_and_all_options column appears in any sum before you publish.

Why does a reseller like Carahsoft show up as a top vendor in federal software spend?

Because it is often the seller of record. Carahsoft describes itself as the Master Government Aggregator for many vendors and holds schedule contracts covering manufacturers such as Microsoft, Salesforce, ServiceNow and Splunk. The award names the reseller's UEI. Attribute those dollars to the product vendor using description text, the parent schedule PIID and the reseller's published lists. Put anything you cannot resolve in an unattributed row.

How recent can my award share window be?

End it at least 90 days before your pull date. Civilian awards reach USAspending within a few business days, but Department of Defense and Army Corps of Engineers data are held back 90 days. A window that runs up to the pull date will show recent defense quarters as near zero. That reads as a decline that never happened.

Does government award share tell me a vendor's commercial market share?

No. I have found no public evidence on how public-sector share relates to commercial share, so treat the two as separate measures. Award share tells you who the government committed money to, for one category, jurisdiction and window. If you need commercial share, build it from separate evidence and do not scale the government number up.

Are PSC codes enough to define a software category in federal contract data?

No. Federal IT product and service codes describe how something was bought: DA01 is labor, DA10 is as-a-service and 7A21 is perpetual license for business applications. A single product can land in two codes, and 7A21 holds every business application. Use codes to pull candidate awards, then apply written description terms and product names, and measure precision on a hand-labeled sample.

How do I account for government purchases below the reporting threshold?

State them as a coverage gap. Do not estimate them into the shares. Purchase card data is not reported into FPDS; GSA publishes it separately on the SmartPay site, where FY2025 spend across purchase, travel and fleet cards was $39.4B. No one has published what portion of a given software category falls below the threshold, so list card spend and below-threshold buys as their own lines under the table.

Try it on the search you came here for

Stop building boolean strings. Just describe the person.

Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.

  1. 01Describe them

    One plain sentence. Role, city, stack, stage, whatever matters to you.

  2. 02I read the web live

    GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.

  3. 03You read the shortlist

    Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.

  • No boolean, no filters, no seat to buy. One box.
  • Read at search time, so a profile updated yesterday counts today.
  • Every step visible as it runs, every name with its reason.

500 free credits on sign-up. No card, no demo call. See real searches.

Read next