The IRS paid Palantir $1.8M to build SNAP, an AI platform that selects audit targets, now in pilot phase.
The IRS contracted Palantir for $1.8 million to build the 'Selection and Analytic Platform' (SNAP), a tool designed to identify high-value audit, tax collection, and criminal investigation cases. The agency cited fragmentation across 100+ legacy systems as the core problem SNAP addresses. The tool is currently in a pilot program and is limited to existing IRS data. Palantir has held IRS contracts since 2014, totaling over $200 million.
SNAP is a case-selection engine built on top of decades of fragmented IRS data systems — a classic enterprise data unification problem solved with Palantir's ontology-based architecture. The technical challenge here isn't the ML model; it's ingesting, normalizing, and ranking signals across 700+ legacy methods. Developers building compliance or fraud detection tools can study this as a blueprint for legacy data modernization at scale.
If you're building a fraud detection or case prioritization feature, benchmark your current scoring pipeline against Palantir's public ontology documentation — specifically how they handle multi-source entity resolution without a unified schema.
Suggested exercise · not independently tested. Access requirements and costs may vary.
Go to developers.palantir.com and navigate to the Foundry or Ontology SDK section
Search for 'entity resolution' or 'case management pipeline' in the docs
Compare the data unification approach to your current fraud scoring architecture — note where your pipeline has equivalent fragmentation to the IRS's 700-method problem
Tags