Client: A major federal civilian agency overseeing social programs.
Overview:
Centennial has been pivotal in reducing improper payments by establishing an AI Center of Excellence (AI CoE) that operationalizes model governance, MLOps, and human-in-the-loop controls using ML risk scoring, graph/network analytics, and NLP to detect anomalies before funds are released. This strengthened data quality and auditability while safeguarding model artifacts and payment data, driving lower error rates and quicker, well-documented recoveries.
Challenges:
The agency, with over 6,000 employees and more than 1,000 vendors, faced rising Improper Payment risk driven by fragmented processes, inconsistent eligibility determinations, and siloed data that limited enterprise visibility. Manual reviews were slow and uneven across regions. Legacy systems produced data latency and incomplete linkages between payees, beneficiaries, and vendors. High volumes and seasonal spikes strained staffing and increased error rates. Governance and audit expectations under PIIA and OMB A-123 intensified pressure to detect issues earlier and document determinations with clear evidence. The agency lacked standardized, real-time risk indicators and had limited capacity to operationalize AI responsibly with traceability, explainability, and security for sensitive payment data.
Solution:
Centennial anchored the solution that is aligned to PIIA and OMB A-123 with internal controls mapped to the GAO Green Book and COSO. Security and privacy controls followed NIST SP 800-53 and the RMF with a Zero Trust posture. AI governance adopted the NIST AI Risk Management Framework with model risk management, explainability, bias testing, drift monitoring, lineage, and auditability. Delivery used Agile, DataOps, MLOps, and DevSecOps to move from pilots to production safely. Time to value was accelerated with Centennial’s pre-built Improper Payment AI toolkit that included a rules and eligibility library, a feature store for payment risk signals, anomaly and outlier models, a graph risk engine, NLP validators for documentation, a human-in-the-loop review console, and audit-ready dashboards and templates.
- Conducted a current-state assessment and control mapping to baseline processes, data flows, and systems and to align existing controls with PIIA, OMB A-123, the GAO Green Book, and COSO.
- Uplifted the data foundation by building governed pipelines, lineage, and quality checks and by creating a reusable feature store for payment risk signals.
- Deployed pre-payment and post-payment risk scoring models to identify high-risk payments before disbursement and to surface suspect transactions after disbursement.
- Used anomaly and outlier detection to catch novel patterns and emerging fraud typologies that rules alone missed.
- Applied graph and network analytics to reveal hidden relationships across vendors, beneficiaries, addresses, and bank accounts to detect collusion and circular flows.
- Used NLP to extract and validate key fields from forms, invoices, and case notes and to compare those fields against policy and program rules.
- Harmonized business rules into a single governed catalog with versioning and transparent links between rules, model outputs, and determinations.
- Enabled human-in-the-loop adjudication by routing high-risk alerts to reviewers with explanations, evidence packs, and confidence scores and by capturing feedback to improve models.
- Prioritized recovery and recapture by scoring recoverability and generating audit-ready packets that supported demand letters, appeals, and investigative referrals.
- Instituted model governance and MLOps with model inventory, approvals, change control, monitoring, drift alerts, and periodic revalidation with full lineage and artifacts preserved.
- Enforced security and privacy with least privilege, encryption in transit and at rest, data minimization, and structured access patterns for sensitive PII and payment data.
- Integrated with enterprise payment, grants, and financial systems and connected to Treasury Do Not Pay and other sanctions and eligibility sources.
- Managed performance with dashboards that tracked precision, recall, false positives, cycle time, recoveries, and Improper Payment rate trends with quarterly scorecards.
- Drove change management and training that upskilled reviewers and program staff on AI-assisted workflows, model explanations, and policy-aligned determinations.
- Stood up an AI Center of Excellence that provided repeatable playbooks, shared code repositories, and secure sandboxes to scale payment-integrity use cases across programs.
Results:
Within the first year, the AI-enabled payment-integrity program materially reduced Improper Payment exposure and sped recoveries. High-risk transactions were intercepted before funds went out, post-payment reviews became faster and more defensible, and audit posture improved under PIIA and OMB A-123. Reviewer productivity rose with human-in-the-loop workflows and explainable models, while data quality and lineage controls created a durable foundation for continuous improvement.
- Reduced the overall Improper Payment rate by 22–30% year over year, with the largest programs showing the steepest declines.
- Prevented an estimated $35–$55M in improper disbursements through pre-payment risk holds and decisioning in the first 12 months.
- Recovered an additional $12–$20M post-payment by prioritizing high-yield cases and accelerating evidence-pack generation.
- Improved model performance to a validated precision of 0.85 and recall of 0.70 on priority programs, while cutting false positives by 35%.
- Shortened adjudication cycle time from a median of 14 days to 5 days, and increased reviewer throughput by 40% with explainable AI and evidence packs.
- Expanded pre-payment risk screening coverage to over 85% of eligible payments, and increased post-payment hit rates by 3–4x on targeted audits.
- Reduced repeat audit findings by 60% and achieved on-time compliance reporting with clean evidence trails mapped to PIIA and A-123 controls.
- Lowered data quality defects tied to eligibility and identity fields by 45% through lineage, validation rules, and monitored feature pipelines.
- Stabilized operations with monitored MLOps, limiting unplanned model drift incidents to fewer than two per quarter and standardizing monthly model refreshes.
- Cut data integration latency from next-day to under two hours for high-risk feeds, enabling near-real-time holds for suspect transactions.
- Achieved a first-year program return on investment estimated at 4–6x through avoided losses, incremental recoveries, and productivity gains.
- Completed training for 100% of frontline reviewers and analysts, with 90% reporting improved clarity of determinations due to model explanations and standardized rules.
- Identified and disrupted multiple vendor and beneficiary risk networks using graph analytics, leading to referrals and strengthened preventative controls.
Conclusion:
Centennial brings a proven blend of mission understanding, AI craftsmanship, and governance rigor to payment integrity, and we pair that with a true partnership model. We co-designed the roadmap with agency leadership across finance, program, acquisition, legal, and security. We aligned decisions through joint governance boards, and measured progress with shared KPIs tied to PIIA and OMB A-123 outcomes. Our teams embedded alongside agency analysts and reviewers to stand up human-in-the-loop workflows, transfer knowledge through hands-on training, and build internal capability so improvements persist. We contributed pre-built toolkits, feature stores, and playbooks. We established clear communication cadences, accelerated ATO with NIST controls baked into the pipelines, and documented reusable patterns that the agency can scale to adjacent fraud, waste, and abuse missions. This partnership approach ensures Centennial’s technology and methods strengthens the agency’s own capacity to sustain lower improper payments, faster recoveries, and stronger audit readiness over time.
About Centennial Technologies:
Centennial Technologies is a trusted provider of tailored solutions for government entities, prioritizing outstanding client service. Our diverse range of comprehensive Financial Management Solutions includes Financial Advisory, Reporting, Performance Management, Business Process Management, Financial Support, Benchmarking, and Risk Management. Our Information Technology (IT) Services offer expertise in Program Management Office (PMO), Application Development and Support, Cloud Solutions, Business Intelligence, Cyber Security, and Emerging Technologies. Renowned for our unwavering commitment to excellence, Centennial Technologies empowers government agencies and commercial clients to fulfill their objectives efficiently and effectively. Contact us directly for more information on how we can support your goals.
Contact Us today to get started.
