Pounce Ai
Veteran Owned
Washington State Certified Veteran-Owned Small Business
The Application of new 10x technology should be triaged to an area of the most governmental good.
DSHS should consider a phased approach to the problem of fraud rather than an omnibus strategy focused upon EBT fraud writ large.
EBT Fraudsters and Their Legacy.
It is far more difficult to address fraud once an individual is already in receipt of ongoing benefits. This is because rules for terminating benefits, the friction associated with discovery of proof of fraud, and rules by other agencies, such as the Social Security administration, specifically has standards that make it very difficult to address fraud absent a notorious use or investigation which demonstrates significant fraud.
Rather than attempt to create a solution that chases the tail, lag indicators of EBT fraud, we propose addressing a key lead indicator for reducing fraud: drastic improvement in the integrity of benefits authorizations. In short, by radically reducing the data burden in the domain specific medical records evaluations before the grant of benefits, DSHS/DDS examiner/adjudicators will have the time they need to focus upon benefit validation, thereby reducing erroneous claims and identifying other lead indicators of ineligible claims. While we propose a phased approach, please note that this is a human centric approach. It takes a data burden of staff and allows that same staff to shift focus to validation.
The question of eligibility for benefits relies to a great extent on what is in the medical evidence. DDS examiners/adjudicators can use Pounce Ai's validated summaries to quickly screen out unmeritorious claims, identify improper online benefit applications and submittals/use of stolen identities at the outset.Conversely, a one-size-fits-all product is unlikely to be satisfactory either in the short or long term for the following reasons:
1. The current rate of technological change is exponential. Any effort to create an overarching systemic fraud detection capability will be obsolete by the time of the contract ends.
2. The complexity of EBT monitoring in the context of its regulatory scheme (banking, and wire transfer channels) will require (at the very least) different subject matter experts than merely those involved with fraud detection at the level of the application. The costs considerations of accomplishing both simultaneously are daunting. Monitoring costs alone would likely exceed the contract amount of $250K in one year. Moreover, addressing EBT channels without materially improving the application benefits process will result in "chasing the tail" rather than a systemic solution.
3. Large scale efforts to create a one-size-fits-all application at this stage of generative AI is a significant strategic mistake. The offerings for a single comprehensive approach given the changes of technology are not mature enough to track fraud in the application and the various disbursements and clearing houses used in the various public commercial spheres where EBT is accepted. This approach will not work nearly as well as appropriately human-supervised agentic models and interfaces tightly bound to a narrow use case initially.
I
In short, DSHS would be wise to narrow the scope of the challenge in order to first assess meeting goals in a single domain channel from which they can expand outward in a systematized way in order prevent gaps that bad actors may exploit.
Given the protentional size of the monies at issue, a gap exploited by significant crime organizations or hostile state actors may result is significant losses akin to those lost during the Washington State Unemployment Fraud. Attempting to solve a protentional 1 Billion dollar loss by relying upon a $250,000 award in Gen Ai is unwise.
During rapidly changing environments such as Gen Ai, we know that phased approaches work. A phased approach leverages the new technology while carefully assessing the newer capabilities on the horizon that may further revolutionize the mission, while checking for gaps and data sinks, that may expose critical vulnerabilities.
The following phased approach avoids or minimizes all the problems listed above:
Phase 1: Use Gen Ai to create a legally relevant, condensed, validated summary of the medical evidence at the application stage with a chat UI (user interface) to query the data. This will create multiple, positive, robust results for the DSHS examiner/adjudicator. However from an anti-fraud point of view, the use of Gen Ai in this way would drastically reduce the time these evaluators would have to spend on record evaluation thereby increasing the accuracy of claim determination. The more accurate the claims determination, the less the fraud enters the entitlement stream.
At DSHS, fraud detection is part of everyone's duties. The reality is that many of the DSHS employees on the front line of services are overburdened with data and paper-pushing related duties which creates a myopic condition, fixated on tasks that thereby misses when fraud enters the stream. The first phase, Phase 1; should be narrowly tailored and validated before moving on the Phase 2.
Phase 2: Pounce Ai will produce suggestions/recommendations to the examiner/adjudicators. These recommendations indicate whether the individual applicant meets the requirements for disability or not and why; or whether more evidence is needed.
Phase 3: Pounce Ai provides Consultative Examiners with the most relevant evidence and summaries in the case, so the examiners can conduct their review of the record and the exam of the claimant with far more specificity and improved outcomes - the right evidence, at the right time, for the right functional assessment. This too allows the DSHS/DDS examiner/adjudicators with an opinion built upon an accurate summary of the record from which to develop the proposed medical consultant review. Again, all of which is validated and free of bias.
It is well known that the number one problem that affects users of DSHS services, including all disadvantaged groups, is an substantial delay in receiving benefits.
Pounce Ai will allow disability adjudicators to use their expertise to facilitate a rapid decision. The decisions will be far more accurate and swift than can be accomplished in the present system.
The ramifications on fraud prevention will be far-reaching. Direct fraud (persons intending to defraud in the application) can be detected through the shifting of the DSHS/DDS adjudicator's focus from massive amount of medical data, to critical analysis of the legal standards at issue. In short, examiners and adjudicators can do what they were hired to do -- problem solve for the citizens of Washington.
Finally, a phased approach will drastically reduce another type of fraud, which is persons receiving an erroneous grant of benefits (or persons who have gotten well and no longer need the benefits, but keep mum). Currently, these grants are not stopped because of an overwhelmed system. However, with Pounce Ai, a renewed emphasis can be applied to the application of Continuing Disability Reviews, CDRs which are currently underemployed due to time issues and other agency priorities.
Pounce Ai's solution to the DSHS Business Integrity Challenge is to employ Gen Ai in the application channel and to allow DSHS to expand out to the various inbound channels for benefits, while the DSHS continues using existing methods of combating EBT fraud. In the next 12-24 months Ai systems will replace the existing tools currently in use for monitoring wire fraud, vender fraud and data systems connecting vendors.
However, new systems will not outclass Pounce Ai's Gen Ai for use in medical summaries validation because our system is not subject to supercedure. Pounce Ai is SME trained to a narrow use case - to alleviate the profound impact of delay upon the truly disabled.