02State Tax & Revenue
Refund-fraud defense and revenue-system modernization that pay for themselves.
State tax agencies are the largest revenue-side IT systems most state IT shops operate. They also sit at the largest fraud surface — refund-fraud rings, false-filing schemes, and identity-theft refund attacks have cost states billions over the last decade. The detection layer is the one that pays for the modernization.
- Annual state tax collections
- $1.1T+
- Refund fraud — annual loss
- $1–3B
- Identity-theft refund attempts
- millions / yr
01What we keep seeing
Three failure modes that pre-date the AI conversation.
01
Refund fraud detected after the refund leaves
Most state tax revenue agencies run identity-theft detection as a batch process on returns already accepted. The IRS sees the same return through different lenses; the state sees it once, post-pay.
02
Single-return scoring misses ring activity
Fraud rings refile across multiple TINs at the same address, the same bank account, or the same prep service. Single-return models can't see the ring. Graph signals over claimant + bank + preparer can.
03
Adverse-action notices are unreadable
When a refund is held, the notice generated rarely tells the filer what to do. Hold queues fill with legitimate filers who got bad letters; fraudsters meanwhile re-attempt with adjusted submissions.
02How we work the seam
Specific practices. Specific outcomes. No platitudes.
Practice 01
Outcome
Move detection to before the refund check is cut
Pre-issuance refund-fraud scoring
Move the fraud signal from post-issuance audit to pre-issuance scoring at the moment the return is e-filed. Graph features over claimant ↔ bank account ↔ preparer ↔ address surface ring patterns single-return scoring misses.
Practice 02
Outcome
Block synthetic-identity returns at submission
Identity-theft refund-attempt detection
Signals on filer behavior, device fingerprint, return composition, and credit interactions detect the synthetic-identity refund attempt before a hold is needed. Pairs with IRS data-sharing where available.
Practice 03
Outcome
Reduce legitimate-filer churn in the held-refund queue
Notice readability and appeal pathway
Rewrite the held-refund notice with counsel review, plain language, and a clear path back to release. Pulls down false-positive friction without weakening the detection layer.
Practice 04
Outcome
Per-determination reconstructibility
Provenance for OIG defense
Every refund-hold decision must carry the features, model artifact, and policy text that produced it. State OIG and the AG's office routinely audit these — the trail decides whether the program survives review.
03Market scale
The size, growth, and obtainable share of this market.
Every benefits-decisioning segment shares the same pattern: program dollars in the hundreds of billions, services spend in the single-digit billions, modernization and detection investments compound across the program spend.
TAM
$1.1T
Program dollars in scope
SAM
$2.4B
Annual services spend
5-yr CAGR
7.2%
Rolling SAM growth
3-yr SOM
$65M
Vardr-obtainable
Vardr captures a small but compounding share of the addressable services spend; the underlying program dollars dwarf the modernization budget.
Federation of Tax Administrators state collections totals. SAM = state DOR modernization + refund-fraud defense services. SOM = 3-yr Vardr-obtainable at current scale.
5-year SAM growth — State Tax & Revenue
$2.4B → $3.4B
04Programs we focus on
State tax and revenue agency modernization.
State Department of Revenue
Personal income tax refund processing
Returns, refund issuance, identity-theft holds, appeals workflow.
State Department of Revenue
Earned Income Tax Credit (state-level)
State EITC refund-fraud exposure, cross-referenced against federal EITC and SNAP eligibility data.
State Department of Revenue
Sales and use tax modernization
Marketplace facilitator collection, audit workflow, remittance-reconciliation tooling.
State and county Department of Revenue / Assessor
Property tax assessment review
Anomaly detection on assessment patterns, appeal queue triage, valuation drift signals.
05Questions worth asking
Open these with us — or with anyone else.
We bring these to every program-office conversation. Use them whether or not we end up working together.
- 01
Of refund-fraud losses last fiscal year, what share was detected pre-issuance vs post-issuance?
- 02
Can your fraud signal join claimant identity to bank account to preparer to address, or only to one of those?
- 03
What does the held-refund notice say, and has counsel reviewed it this fiscal year?
- 04
For a refund hold flagged eight months ago, can you reconstruct the exact features and model artifact that triggered it?
06Related insights
Working in state tax & revenue and want our take?
45-minute principal-level briefing. Bring the program, the constraint, the deadline.