Revolutionizing Social Benefit Delivery: AI with Human Touch
Abstract
US social agencies have witnessed more operational
challenges during and after the COVID-19 pandemic. Generally, a social agency
needs to be prepared to serve its changing population every month; however, its
resources are often limited. This paper covers two key measurements defined by
Food and Nutrition Services (FNS) and how emerging technologies using AI can improve
operational efficiencies.
Keywords: Application Processing Timeliness (APT), Case and Procedural
Errors Rate (CAPER), Intelligent Document Processing (IDP), Robotic Process
Automation (RPA), AI Co-Pilot, Health and Human Services Automation, Food and
Nutrition Services (FNS).
1. Introduction
Social agencies face
constant challenges in
efficiently and effectively serving the needy population. Regulatory
authorities in the US,
like Food and Nutrition Services (FNS), monitor these efficiencies through
various measures, such as
the Application
Processing
Timeliness rate, Case And
Procedural
Error Rate (CAPER), and Payment
Error Rate. Limited staff and resources make it more challenging to manage the
delivery for a fluctuating population.
This paper will primarily
cover two aspects of efficient benefit delivery: Application Processing Timeliness
and accuracy of actions and procedures (as measured by CAPER).
Post and pre-pandemic impact
is visible; most states are struggling to bring back their performance or make further
improvements.
Figure
1:
Application processing timeliness rate – pre and post-pandemic.
Figure 2: Case and procedural error rates
by fiscal year-national rates.
3. Emerging Technology to the Rescue
Automation while keeping the
human touch at the core will help improve accuracy and timeliness.
3.1. Application processing timeliness
The key obstacles to timely
application processing are:
Agency-Caused Delays: Agency-caused
delays are when the application is with the agency and no action is taken or no
communication is sent to the client. One of the main reasons for this is the
backlog, i.e., the agency did not have enough resources to process these
applications on time. Another reason could be an inefficient system of record,
i.e., caseworkers are often receiving system errors while processing the
application, which delays the processing till the issue is resolved.
Client-Caused Delays: Client-caused delays occur when the agency prompts the client for action but the client fails to provide information. This can occur if the communication is unclear or not received on time by the client, e.g., the notice is lengthy and confusing, there is returned mail, or the client cannot respond on time.
The agency should aim to
avoid agency-caused delays and minimize client-caused delays. The following
technologies can help the Health
and Human Services
industry meet
this goal.
1)Intelligent Document Processing
(IDP): As the name suggests, IDP
leverages AI to understand scanned documents intelligently. The key attributes
of IDP and use cases in Health and Human Services industry are:
a)Structured and Unstructured Data
Extraction: It can extract data from the
application form, driving license, Pay stub etc.
b)Document Classification: It can help sort the document into various categories for instance
client submits the full package which may include SNAP application, Medicaid
recertification, and residency and income proofs. IDP can sort these documents
into respective categories
c)Accuracy & Exception Handling: It can score the accuracy of data extraction and the agency can
configure when human review is required based on the accuracy rate of extracted
data.
d)Mapping The Data To The System of
Record: Intelligently mapping the
extracted data to the eligibility systems can save merit caseworker time.
e)Continuous Learning and Improvement: Machine learning is used for self–learning and improvement. It learns
based on feedback provided by merit caseworker reviewers while managing
exceptions.
IDP can support the caseworker in document processing and data
mapping/entry in the system of record, which can save 60-100 minutes per
application or more. (Note: participating households in the SNAP program
in the month of April 2024 ranged from 10608 to 3146853. Approx per state
average 418885. Source:
https://www.fns.usda.gov/pd/supplemental-nutrition-assistance-program-snap)
2)Robotic Process Automation (RPA): Robotics can help intelligently automate repetitive tasks. Here are
some example use cases of RPA that can be implemented in Health & Human Service
agencies:
a)Synchronizing Task and Activity Status: One of the caseworker's common challenges is updating the task status
after working on the related activities to match the activity status with the
task status. Many times, agencies need to bulk delete the tasks as they are
redundant or may not be required as the associated activity is already
complete. RPA can save the caseworker and agency’s effort to reconcile these
statuses by automatically running this process of validating and updating the
task status based on the associated activity status.
b)Enhance IDP: RPA can further enhance document
processing by creating tasks for the merit staff to review the mapped document
for eligibility
c)Verification with internal and external systems: RPA can help verify household details from trusted sources.
d)Returned mail processing: RPA with
OCR/IDP can extract the forwarding address details from the returned mail
envelope, update the address in the record system, and regenerate the notice or
create the task for the caseworker.
Procedural
Errors: Some common procedure errors are incorrect processing
of wages and salaries, Shelter expenses, and household composition.
Incorrect
Notices: Incorrect notices can be triggered based on incorrect
actions performed on the case.
AI
and machine learning technology can aid the caseworker in case processing to
avoid or minimize these errors. Some of the examples are:
O-piloting: An AI assistant can guide the caseworker based on case data and common
errors that have occurred in similar cases. AI assistants can be trained using
Job aids, Quality audit recommendations, and Policies.
A real-time caseworker assistant or co-pilot will help provide a safe
place for the caseworker to ask questions, catch errors proactively, and guide
the caseworker in processing the application.
Process Mining: Business process mining tools can
help the supervisor manager track an end-to-end process at the desktop level.
This can provide opportunities to compare the standard process and diversion
from it. It can further provide insights into the cause of delays and errors. The
outcome can be used to train the AI assistant to provide real-time assistance
on standard processes to manage error-prone processing.
4. Futuristic View
AI with Human Touch in Health and Human Services will soon
involve AI doing the repetitive work and caseworkers handling more complex or
exceptional cases.
Figure
3: Futuristic
view - common and repetitive actions are performed by AI, and focused complex
actions are performed by humans.
4. References
1. Snap Application Processing Timeliness (2022) https://www.fns.usda.gov/snap/qc/timeliness
2.SNAP Case And Procedural Error Rates (2023) https://www.fns.usda.gov/snap/qc/caper
3.SNAP data tables – Sates level Participation
and Benefits (2024) https://www.fns.usda.gov/pd/supplemental-nutrition-assistance-program-snap.