Solutions to address today’s most common requirements
Accenture researched how current AI technologies could be used by the U.S. federal government, documenting nearly one hundred discrete use cases across various agencies in the civilian, national security, law enforcement and public safety, and healthcare sectors.
Five common categories with wide applicability emerged where AI can most readily drive performance improvements today. These include case management, customer service, human capital management, fraud and risk management, and inspections and maintenance.
Challenges – High claim or case volumes, manual data entry, lack of integrated data, and ad hoc decision-making result in long processing times, inaccurate reporting and inconsistent outcomes.
AI Solution - Machine learning with robotic process automation can speed up data entry and minimize errors. Semantic technologies can automate a broader array of decisions, and intelligently route complex cases to a specialized case manager.
AI in Action – A federal agency needed to help reviewers more quickly determine the verification information required from claimants. Accenture quickly created a prototype solution that used machine learning and predictive analytics to streamline and shorten the evidence gathering process for adjudicating claims.
Using the agency’s existing data warehouse, it employed models to “learn” what evidence is typically required or most effective based on historical claims. With this insight, the solution was able to generate a recommended evidence list at the onset of each claims review process, eliminating time-consuming and frustrating back and forth communication with the claimant.
Challenges – Extended call wait times and case backlogs create a poor customer experience that can result in non-compliance, loss of revenue, and poor public reputation.
AI Solution - Virtual agents with natural language processing can enable customer self-service and route citizens to the right information or representatives, freeing up call centers and employee time for more complex cases.
AI in Action –Service centers for the U.S. Citizen and Immigration Service (USCIS) receive approximately 14 million calls annually. To mitigate call volume, and provide a better customer experience, USCIS added self-service options to better support online audiences. A virtual assistant chat service – “Emma” – helps visitors access information more easily across the USCIS website.
According to USCIS, the solution has dramatically reduced the amount of time visitors spend looking for information. Emma now answers over one million questions per month, providing the agency with a significant amount of data which can be used to improve the service in the future.
Human capital management
Challenges – The inability to identify the right talent pool and execute targeted recruiting in a timely and effective manner as well as difficulty assessing the competencies of potential hires can compound existing workload/backlog issues.
AI Solution - Natural language processing and machine learning can identify applicants with the right skills and behavioral attributes for future job success.
AI in Action – A large international consulting firm wanted to streamline the new hire process and improve the validity and predictive accuracy of prospective employee assessments. Since many candidate interviews were done remotely, developing more accurate initial screens around this constraint was a priority. Using machine learning, they deployed an assessment solution that can extract and understand behavioral attribute patterns from a candidate’s digital interview through audio, video, and text analysis.
The benefits include gaining insight into skills and attributes that matter, building teams with top performers, improving retention rates, and reducing the recruiting process cycle time for global recruiting efforts. Results include: saving 9 hours average in the interview process, improving the remote candidate experience due to location and time flexibility, and improving the interview-to-hire ratio with smarter assessment tools.
Fraud and risk management
Challenges - Disparate data and siloed organization structures result in difficulty identifying and proactively mitigating vulnerabilities across the enterprise.
AI Solution - Robotic process automation can validate and integrate data from multiple sources. Machine Learning can analyze behaviors to identify emerging trends in fraud and abuse so that agencies can act before they cause significant damage.
AI in Action – A government agency responsible for social welfare benefits needed a better way to combat fraudulent claims. Through the use of machine learning, Accenture was able to analyze case files and interview transcripts to identify specific attributes associated with high-risk groups. A solution was then implemented to monitor for suspicious transactions using these criteria.
The initial machine learning models resulted in a 20 percent improvement in fraud detection compared to the current methods in place. The models continue to learn and refine, improving accuracy as more data is analyzed over time.
Inspections and maintenance
Challenges - Manual, infrequent inspections and maintenance processes result in high cost of repairs and safety concerns.
AI Solution - Remote sensors, video analytics, and machine learning can be combined to model and predict risks, such as mechanical failures, contaminated food, or public safety threats to save the investigator time by pinpointing higher likelihood cases, and in that way, the investigator can take preemptive action.
AI in Action – An international oil and gas production company sought to implement predictive maintenance to reduce repair costs, improve uptime and minimize lost production. Using a human-centered approach to problem identification, Accenture worked with the client’s engineers to codify required maintenance, potential breakdowns and proposed process controls. Ultimately, more than 3,000 fit-for-purpose models were developed to predict equipment failure and pinpoint necessary maintenance intervals.
These predictive asset models currently monitor more than 200,000 sensors spanning more than thirty equipment types. By using these models to analyze operating data and inspection reports, the company was able to predict high risk-areas for maintenance and potential failure, resulting in safer working conditions, higher production levels, lower operating costs and more effective incident investigations.