A child welfare algorithm reduced bias by 83%. AI for small businesses needs this

The results are solid. Trusting them is not

Several independent evaluations were carried out. These evaluations produced unusually clear findings. Researcher Katherine Rittenhouse and her colleagues conducted a study on this tool. As a direct result, they reported that reduced the racial disparity in selection rates for higher risk referral types by 83%. The point difference in selection rates between white and black children fell from 10.6 to 1.8. Additionally, this tool reduced the black-white gap in removal rates by 73%, from 4.3% to 1.2%. As in the previous study, the study indicated that there would likely be a reduction in disparity, although it again indicated that rater processing time decreased by approximately 5%. The rate of decrease in filter processing time was relatively minor compared to the significant degree of increased fairness in these cases.

The system remains somewhat controversial. Caseworkers reported that they did not understand what inputs the algorithm used in its assessment. Parents and advocates expressed opposition to scoring based on their children’s performance (although the data indicated a positive outcome), as well as the general use of a “scoring” process. Allegheny County has reduced the algorithm’s role to that of an advisory tool. The final decision is made by the human selection staff. This approach provides for a human presence during the selection process; However, a tool can reduce bias while making people question both the methodology and the results.

A newer tool tries to act sooner

He Trust FundBattle The model is another attempt at the same thing and much younger. It will test an AI-powered educational software product in Chicago as part of a pilot project in 2026. The Allegheny tool works after a complaint has already been filed. On the other hand, TrustFundBattle seeks to operate much sooner. Their goal is to reach young people before they even encounter the justice system. The program uses an app and handheld unit to teach decision making and money management. Any rewards awarded through this program are deposited into savings accounts that can be used to continue your education or employment.

Mark Barron, founder and CEO of TrustFundBattle, wrote in a statement: “We have spent decades investing enormous amounts of money in response to the consequences of youth incarceration, while investing comparatively little in preventing first contact with the justice system. AI gives us the opportunity to shift resources upstream, when behavior, trust, responsibility and decision-making patterns are still developing.” “Young people don’t change because they are given another sermon,” he said. “They change when learning becomes interesting, progress becomes visible, and positive decisions create meaningful rewards. By combining AI, gaming, behavioral science, and milestone-based financial incentives, we can make life skills education something that young people actively participate in rather than passively receive.” “Prevention creates a different type of return on investment,” Barron said. “When a youth avoids incarceration, develops stronger life skills, completes educational milestones, and begins to build a financial foundation, the value extends beyond an individual. Families, communities, schools, employers, and public systems benefit.”

Actual cost rarely appears on a dashboard

The cost data illustrates why the initial move can be worth testing before a company has any testing. There is a significant amount of money at stake and, as the Bureau of Justice Statistics in its 2018 A report on state prisoners released in 2005 in 30 states showed that 83% of those prisoners were arrested at least once in 9 years. According to the Vera Institute of Justice, local governments pay annually 25 billion dollars to finance prisons. On average, $47,057 per person is spent to keep someone incarcerated. Many states have much higher average youth incarceration costs. In fact, the national average annual cost to safely confine juveniles has increased from approximately $149,000 per year in 2014 (Justice Policy Institute) to more than $214,620 per year.

Hedwig Lee, then a professor of sociology at Washington University in St. Louis, has studied how incarceration affects families. “Criminal justice policy is not just policy about criminal justice,” Lee said. “It’s also a health policy decision, a family policy decision, and an economic policy decision. That kind of data is much harder to collect than a simple dashboard metric. A similar mistake can be found among many small businesses that measure AI ROI. Most focus on immediate results: time saved or tickets closed. However, costs avoided through AI often remain hidden from reporting. It would take years for researchers to demonstrate the Allegheny results. Those results were never measured by a single quarter’s performance or a short test.

Both examples point to a different type of return on investment: risk prevention, not just automation of a task. During their review of vendors providing artificial intelligence, a small business owner should ask themselves the same question. Is the system waiting until a failure occurs, at which point it takes action or adjusts its options before they become unchangeable costs? TO tool that waits for a problem to occur and then interventions can still add value. An example would be a chatbot that responds to complaints about a product or service. Another example would be a fraud detection model that identifies fraudulent transactions once they are completed. These are examples of intervention tools. Tools designed to prevent problems from occurring are proactive. Examples include identifying invoices where payment is likely to be late so you can make changes to payment terms or using natural language processing to identify a customer’s potential frustration in their initial communication with a support representative rather than the fifth communication.

Artificial intelligence tools wait for failure or prevent it

The conclusion is not clear to either of them. The Allegheny tool has years of data to back it up. However, the Allegheny tool also has the same problem as many other tools: its users don’t trust what it tells them. TrustFundBattle has a great layout for an idea. However, as with all ideas in development, it has no measurable outcome. Ultimately, both examples demonstrate that preventative measures against AI-related risks are very different in terms of return on investment (ROI) than standard automation processes. This new type of ROI may be more difficult to quantify. A person simply asking about your automation process “what does this tool automate?” He is looking at the small picture. The big picture is how far will the automation tool go?