WA Agency Amazon Spend (wa-amazon-spend)

Pull Washington state agencies' Amazon Business purchases - 944,699 line items / $81M across 9 fiscal years (FY18-FY26), from the state's free public open data.

This is the long tail of government procurement: every individual order an agency places through Amazon - the laptop, the LEGO set, the lab beaker, the printer toner - that nobody else tracks. It is the counterpart to WA Master Contracts: master contracts show the big negotiated vendors; this shows the day-to-day retail spend underneath.

Built for AI agents, developers, and market-intelligence / audit users who want to know what WA agencies are actually buying, from whom, at what price, under which UNSPSC commodity code.

What this data is

Each record is one line item - an agency's purchase of one product:

Field	Meaning
agencyName	The buying agency (e.g. SOCIAL AND HEALTH SERVICES DEPT)
orderDate / paymentDate	Order / payment date
productCategory / productTitle	Amazon's category + the product's full title
brand / manufacturer / sellerName	Who makes / sells it
unspsc	8-digit UNSPSC commodity code (FY19+; FY18 predates the column)
quantity / itemSubtotal / shippingHandling / tax / netTotal	What it cost
fiscalYear	Which FY dataset it came from

UNSPSC is the unique angle. Master contracts and agency contracts carry no product codes - Amazon spend does. So you can query "what agencies bought under commodity class 44 (office equipment)" directly.

Typical questions it answers

• What is an agency buying from Amazon? agencyName=NATURAL RESOURCES -> every line item.
• Who spends the most? Aggregate by agency -> total spend + order count.
• What categories dominate? Aggregate by productCategory -> the tail of office/safety/IT spend.
• Surge / anomaly monitoring - which agency's Amazon spend spiked this quarter? Aggregate by agency + dateFrom, compare against last quarter. Audit signals.
• Specific products / vendors - keywords=["LEGO"], brand=HP, or an 8-digit unspsc=60141100.
• Big-ticket orders - minNetTotal=1000 for orders over $1k.

Aggregate mode

Turn on aggregate and it rolls up spend over the whole 944k-row corpus instantly, no matter how broad your filters:

Output field	Meaning
key	The agency / product category / brand / fiscal year
totalNet	Total net spend for that group
orders	Number of line items

Group by agency (who spends most), productCategory (what dominates), brand, or fiscalYear.

Inputs (all optional)

Input	What it does
keywords	Match any keyword against product title / brand / seller
agencyName	Substring match on the buying agency
productCategory	Substring match on Amazon's category
brand	Substring match on brand
unspsc	8-digit UNSPSC code - accept 1/2/4/6/8 digits (exact at 8). FY18 auto-skipped
dateFrom / dateTo	Order date range (YYYY-MM-DD)
minNetTotal	Only orders at or above a dollar amount
fiscalYear	Restrict to one FY dataset (FY18-FY26), or all
aggregate + groupBy	Roll up by agency / productCategory / brand / fiscalYear -> totalNet + orders
maxResults	Cap the number of records (default 50)

Example inputs

What one agency buys - agencyName is a substring match.

{ "agencyName": "NATURAL RESOURCES", "maxResults": 10 }

Search the product text - keywords matches the title, brand or seller.

{ "keywords": ["LEGO"], "maxResults": 10 }

Everything under one commodity code - unspsc takes 1/2/4/6/8 digits; 8 digits is exact.

{ "unspsc": "60141100", "maxResults": 10 }

Who spends the most - roll the whole corpus up by agency.

{ "aggregate": true, "groupBy": "agency", "maxResults": 10 }

Low cost

From $0.0003 per record - and each record is metered individually.

This is nine fiscal years of Washington agencies' Amazon purchases - the state publishes each FY as a separate dataset, and the schemas drift across years: FY18 has no commodity-code column at all, and the same column that is numeric in FY19 is text in FY22. We merge all nine onto one query surface that knows each year's shape, so a filter over the whole 2018-2026 span works even where a naive query would hit a type mismatch. You query all nine fiscal years at once over the full ~944k-line-item span; the per-year schema drift stays on our side. An unfiltered run returns the most recent line items as a live monitoring feed.

Example output

One line item - productCategory="eBooks" returns records like this one:

{
  "platform": "wa-amazon-spend",
  "fiscalYear": "FY26",
  "agencyName": "SOCIAL AND HEALTH SERVICES DEPARTMENT OF (DSHS)",
  "orderDate": "2026-04-07",
  "paymentDate": "2026-04-07",
  "productCategory": "eBooks",
  "productTitle": "The Art of Awareness: How Observation Can Transform Your Teaching",
  "asin": "B0999RM88N",
  "brand": "0",
  "manufacturer": "Redleaf Press",
  "sellerName": "Amazon.com Services LLC",
  "productCondition": "New",
  "unspsc": "55111500",
  "quantity": 1,
  "itemSubtotal": 19.99,
  "shippingHandling": 0,
  "tax": 1.7,
  "netTotal": 21.69,
  "paymentAmount": 21.69
}

aggregate=true, groupBy=agency - one summary row per agency (key, totalNet, orders) instead of individual line items.

Data source & freshness

• Source: data.wa.gov - State Agency Amazon Spend (Socrata SODA API, free public, no login) - 9 fiscal-year datasets.
• Update cadence: near-daily as orders are reported - the freshest state spend data on WA open data.
• Coverage: agency purchases through Amazon Business only. FY18 has no UNSPSC column (schema predates it).