A mother photographs her pay stub on a cracked phone screen at 11 PM. The image is slightly blurry, slightly tilted, and partially in shadow. Legacy systems reject it. A caseworker requests a clearer copy — adding two weeks to the process. Commonwealth's AI reads the document, extracts the data, cross-references it against the IRS income database, and verifies the information — automatically, in seconds, without human intervention.
Document verification is the single biggest bottleneck in benefits processing. A caseworker receives a blurry photo of a pay stub. She squints at the screen. She types the employer name, the gross amount, the pay period, and the pay frequency into separate fields. She checks it against the applicant's self-reported income. She flags a discrepancy — the applicant reported $1,800 but the stub shows $1,842. She sends a request for clarification. Two weeks pass. The applicant, who needed food assistance last week, waits. The caseworker, who has 399 other cases, moves on. And the system, which was designed to help, creates a two-week delay because a human had to read a photograph and type numbers into boxes.
Commonwealth's document engine automates 72% of this work. AI reads the document. OCR extracts the data. The system cross-references against IRS income verification, SSA benefit verification, DHS immigration status, and state vital records — automatically. When the data matches, the case moves forward without a caseworker touching it. When it doesn't, the system flags the specific discrepancy for human review — not the entire document.
Each hub provides authoritative data that eliminates the need for manual document review in most cases.
From AI classification through federal data hub verification — every step automated, every exception intelligently routed.
When an applicant photographs a document and uploads it through the citizen portal, the first thing legacy systems do is ask: "What type of document is this?" The applicant must select from a dropdown of 15-20 options — many of which they don't understand. Is their income verification a "pay stub," an "earnings statement," or a "wage verification"? Commonwealth eliminates this step entirely. AI classification models trained on millions of government benefit documents identify the document type from the image itself — recognizing pay stubs from 200+ employers, lease agreements in dozens of formats, utility bills from major and regional providers, and government-issued identification from all 50 states. The applicant simply points the camera and taps. The system knows what the document is.
Benefits applicants don't have scanners. They have phones — often older phones with lower-resolution cameras. The documents they photograph are crumpled pay stubs pulled from a pocket, lease agreements with coffee stains, and utility bills with faded ink. Legacy OCR systems, designed for flatbed scanner output, reject 30-40% of these images. Commonwealth's OCR engine is trained specifically on mobile-captured documents in real-world conditions: rotation correction, perspective normalization, shadow removal, blur compensation, and adaptive contrast enhancement process the image before text recognition begins. The system extracts not just raw text but structured data: from a pay stub, it extracts employer name, employee name, gross pay, net pay, pay period start/end, pay frequency, and year-to-date earnings — mapping each value to the correct field for eligibility calculation.
The most powerful verification is the one that doesn't require a document at all. When an applicant reports $1,842 in monthly wages from Walmart, Commonwealth queries the IRS Federal Data Services Hub — which returns wage data reported by Walmart on the applicant's W-2 or quarterly wage report. If the IRS data confirms the self-reported income, the case proceeds without the applicant ever submitting a pay stub. The same principle applies across six federal and state data sources: SSA confirms Social Security benefits, DHS SAVE confirms immigration status, state vital records confirm identity and household composition, employer new hire registries detect employment changes, and financial verification services confirm bank account balances for programs with asset tests. When data hub verification confirms the applicant's reported information, no document is requested. The 72% auto-verification rate means that nearly three-quarters of all verification requirements are satisfied without a single document upload.
Income verification is the most common verification requirement across all benefits programs — and the most common source of processing delays. Legacy processes require the applicant to submit four consecutive pay stubs, the caseworker to manually calculate average monthly income, and a supervisor to review any variance. Commonwealth automates this entirely: the system first checks IRS data for the applicant's most recent reported wages. If IRS data is available and within tolerance of the self-reported income, verification is complete — no pay stubs needed. If IRS data is unavailable or insufficient (new employment, recent job change), the system falls back to pay stub extraction, verifying the OCR-extracted data against the applicant's self-reported amount. Discrepancies within configurable tolerance bands (typically ±5%) are resolved automatically with a conservative income assumption. Only discrepancies outside tolerance require caseworker review — and even then, the system presents the specific data points that don't match, not a raw document.
Identity verification is the gateway to benefits access — and it is the step that most often excludes vulnerable populations. People experiencing homelessness may have expired IDs. Immigrants may have foreign-issued identification that caseworkers don't recognize. Domestic violence survivors may have IDs with an address they've fled. Commonwealth's identity validation engine reads and validates identification documents from all 50 US states (driver's licenses and state IDs), US passports, permanent resident cards, employment authorization documents, and 40+ foreign government identification documents. The system extracts biographical data (name, date of birth, address, ID number), checks expiration dates, validates document structure against known templates, and cross-references against state vital records when available. When digital validation cannot confirm identity, the system offers alternative pathways — including in-person verification at community partner locations — rather than denying access.
The most frustrating experience in digital benefits applications is submitting a document, waiting days or weeks, and then receiving a request to resubmit because the image was unreadable. By that time, the applicant may have discarded the document, lost access to it, or simply given up. Commonwealth eliminates this by checking image quality at the moment of capture — before the applicant moves on to the next step. The quality check evaluates resolution (is the text readable?), focus (is the image sharp enough for OCR?), completeness (is the entire document in frame?), lighting (is glare or shadow obscuring text?), and orientation (is the document upside down or sideways?). If any quality issue would prevent successful extraction, the system immediately requests a re-capture with specific guidance: "The bottom of the document is cut off — please include the full page" or "Glare is obscuring the dollar amount — try tilting the document slightly."
Government benefits programs have strict document retention requirements — SNAP records must be retained for three years, Medicaid records for five years in most states, and TANF records for the duration of the time limit plus additional years. In legacy systems, documents are stored in paper files, scanned into image repositories with minimal indexing, or scattered across program-specific document management systems. Retrieving a specific document for an audit, appeal, or fair hearing can take hours or days. Commonwealth maintains every document in a unified, encrypted, searchable repository — indexed by household, program, document type, verification status, submission date, and caseworker. Any document can be retrieved in under three seconds. Retention policies are configured per program and enforced automatically — documents are purged only when retention periods expire, with configurable legal hold capabilities for cases under appeal or investigation.
Not every verification can be automated — and the system is designed to recognize when human judgment is needed. When automated verification finds a discrepancy between the applicant's self-reported data and the data hub or extracted document data, the case is routed to a caseworker with a structured resolution package: the specific data points that don't match (not the raw document), the source of each data point, the magnitude of the discrepancy, and a recommended resolution based on policy. For example: "Self-reported income: $1,800/month. IRS data: $1,842/month (employer: Walmart). Discrepancy: $42 (2.3%). Recommendation: Use higher amount (conservative). Impact on eligibility: No change across any program." This structured presentation enables the caseworker to resolve the discrepancy in minutes rather than the hours required to manually review a document, calculate income, and determine impact.
A state HHS agency integrated Commonwealth's document engine with all six federal data hubs. In the first year, 72% of income verifications were completed through IRS data hub cross-reference alone — without the applicant submitting a single pay stub. Identity verification through state vital records eliminated manual ID review for 84% of applicants. The average verification timeline dropped from 14 days to 2 days. Document-related requests for information (RFIs) — which had been the primary cause of processing delays — decreased 68%. The caseworker team that had been dedicated to document review was reduced from 18 to 5, with the 13 reassigned workers moved to case management roles where they could focus on helping families rather than reading pay stubs.
An urban county processing 40,000 applications annually deployed Commonwealth's AI document classification and OCR extraction. Previously, each submitted document required a caseworker to open the image, identify the document type, manually key extracted data into eligibility fields, and file the document in the case record — averaging 8 minutes per document. With AI classification and extraction, the same process takes 30 seconds: the system classifies the document, extracts structured data, and maps it to eligibility fields automatically. The caseworker reviews a pre-populated summary and confirms or corrects. Document processing throughput increased from 40 documents per worker per day to 280. The quality gate — real-time image checking at the point of capture — reduced unusable document resubmissions by 78%.
During the Medicaid continuous enrollment unwinding, three states used Commonwealth's document engine to automate renewal verification at scale. Instead of mailing renewal forms and waiting for returned documents, the system proactively checked IRS income data, SSA benefit data, and state vital records for each household approaching renewal. For 64% of households, all verification requirements were satisfied through data hub cross-reference alone — enabling "ex parte" renewal without any contact with the beneficiary. For the remaining 36%, the system identified only the specific data points that needed updating and sent targeted requests through the citizen portal. Across all three states, procedural coverage loss rates averaged 5% — compared to 18-25% in states using legacy processes. An estimated 140,000 eligible people retained their Medicaid coverage because the system verified their continued eligibility without requiring them to navigate a manual renewal process.
I used to spend my entire day reading pay stubs. Eight hours a day, five days a week, reading photographs of pay stubs on a screen — squinting at blurry numbers, calculating averages, typing data into fields. I processed about 40 documents a day. The AI processes 280 per worker now, and I review the pre-extracted data in a few seconds. My job went from typing numbers I could barely read to confirming numbers the system already extracted. I am doing the same work in one-seventh the time. The other six-sevenths? I am actually talking to families. Helping them understand their benefits. Doing what I thought this job would be when I took it.
The IRS data hub integration changed everything. Sixty percent of our applicants work for large employers — Walmart, Amazon, McDonald's — who report wages quarterly to the IRS. Before Commonwealth, we required every one of those applicants to submit four consecutive pay stubs. Now the system checks IRS data first. If the IRS confirms the wages, we never ask for a pay stub. The applicant does not have to find the document, photograph it, and upload it. The caseworker does not have to review it. The verification is already done. Fifty-eight percent of our wage verifications are now completed this way — invisibly, instantly, with zero burden on the family or the worker.
During the Medicaid unwinding, we had to re-verify 280,000 people in 90 days. Other states were mailing paper forms and losing 20% of their Medicaid population to paperwork failures. We let Commonwealth check the data hubs first. For 64% of our beneficiaries, the IRS confirmed their income was still below the threshold, SSA confirmed their benefit amounts hadn't changed, and the system renewed their coverage automatically — without the beneficiary lifting a finger. They never even knew the renewal happened. They just kept their health insurance. One hundred and forty thousand people across three states kept their coverage because we let computers do what computers do — check data — instead of asking families to mail paper to prove what the government already knew.
Request a demonstration of Document Management & Verification — including AI classification, data hub integration, and automated income verification.