The AI tool for matching foster children with potential adoptive families is based on the concept of online dating, but it has not been successful in meeting the unique needs of vulnerable foster kids.

There are some children who are without parents, while others have been taken away from their parents. A large number of these children are older and have significant needs or disabilities. Many of them carry the effects of past trauma, having been moved between foster homes, separated from their siblings, or subjected to sexual and physical abuse.

For many years, child protective services organizations have struggled with the task of finding permanent homes for at-risk children and teenagers. This challenge is so immense that social workers cannot always ensure a perfect match.

Thea Ramirez entered the chaotic situation with her proposed technological fix – a tool that uses artificial intelligence to predict which adoptive families will have lasting relationships. Ramirez asserted that this algorithm, created by previous employees of an online dating platform, has the potential to increase successful adoptions nationwide and improve efficiency for underfunded child welfare organizations.

In an April 2021 YouTube video, Ramirez discussed her goals to revolutionize the way America matches children and families by utilizing the Family-Match algorithm, which relies on scientific data rather than personal preferences to determine a score that can accurately predict long-term success.

A recent inquiry by the Associated Press revealed that the AI software, which is one of the few adoption algorithms available, has yielded minimal success in the states where it has been implemented. This information was obtained through public records requests made by the AP to state and local agencies, based on self-reported data from Family-Match.

Ramirez has exaggerated the abilities of the exclusive algorithm to government authorities in her efforts to broaden its scope. Despite receiving feedback from social workers that the tool was not practical and frequently led them to uncooperative families, she continued to promote its use.

According to a review of numerous documents by the Associated Press, Virginia and Georgia have stopped using the algorithm due to its failure to generate successful adoptions during trial runs. However, both states have since resumed working with Adoption-Share, a nonprofit organization run by Ramirez.

The program in Tennessee was abandoned before being implemented due to compatibility issues with their internal system. Despite state officials dedicating over two years to setting it up, social workers in Florida, where the program is expanding, have reported varying experiences with Family-Match.

According to state officials, the CEO of the organization, Ramirez, possesses some of the confidential information gathered by Family-Match. They also pointed out that the nonprofit has not been transparent about the functioning of the algorithm.

The AP discovered that these experiences offer insight for social service organizations looking to use predictive analytics but may not fully understand the limitations of the technology. This is especially important when attempting to tackle long-standing human issues, such as finding homes for children who have been deemed the “least adoptable” by judges.

Bonni Goodwin, an expert in child welfare data at the University of Oklahoma, stated that it is impossible to have a foolproof method for predicting human behavior. She also noted that adolescence is particularly unpredictable.

Ramirez, from Brunswick, Georgia, where her organization is located, declined to disclose information about the algorithm’s internal processes and rejected requests for interviews. In an email, she stated that the tool serves as a starting point for social workers and does not make decisions on whether a child will be adopted. She also challenged the testimonies of child welfare leaders regarding the effectiveness of Family-Match.

According to Ramirez, surveys and communication with our agency’s clients show that Family-Match is a useful tool for those who actively use it to aid in their recruitment and matching efforts.

According to public statements, newsletters, and a blog post, Ramirez, who used to work as a social worker and is now married to a pastor in Georgia, has consistently advocated for adoption as a means of decreasing the number of abortions.

Over ten years ago, she created a website with the purpose of linking expecting mothers with prospective adoptive families. She advertised it as the sole online community specifically for connecting crisis pregnancy centers and promised to donate 10% of membership fees to these centers, which aim to convince women to continue their pregnancies. Ramirez clarified in an email that Family-Match is not affiliated with these centers.

Her attention then shifted towards aiding children in foster care who lack guardians to care for them. According to federal data, the majority of the 50,000 children adopted in the United States in 2021 were placed with relatives, while approximately 5,000 were placed with unfamiliar individuals. Social workers claim that adoptions based on recruitment are the most challenging to execute.

Ramirez requested the assistance of Gian Gonzaga, a research scientist who oversaw the algorithms at eharmony, a dating site that emphasizes Christian values and guarantees “genuine love” for those looking for marriage. She proposed collaborating with Gonzaga to develop a tool for matching potential adoptive parents with children.

Gonzaga and his wife Heather Setrakian collaborated at eharmony and later on the Family-Match algorithm. When asked, Gonzaga deferred to Ramirez. Setrakian expressed pride in the time she spent developing the Family-Match model.

The dating site eharmony’s spokesperson, Kristen Berry, clarified that they are not associated with Family-Match. Berry referred to Gonzaga and Setrakian as previous staff members.

Afterwards, Ramirez traveled across the nation to promote Family-Match to government officials. Her advocacy and religious beliefs garnered support mainly from those on the conservative side, including first lady Melania Trump. Trump highlighted Ramirez’s work at a foster care event held in the White House Situation Room. Ramirez has collaborated on research papers and delivered a significant presentation at the American Enterprise Institute. She has also utilized eye-catching fundraisers and leveraged her connections to persuade state officials to implement her tool.

According to social workers, Family-Match operates in the following manner: Adults looking to adopt complete a survey using the algorithm’s online platform, while foster parents or social workers enter information about each child.

Once the algorithm calculates a score to assess the compatibility of relationships, the Family-Match platform presents a roster of the most suitable potential guardians for each child. Subsequently, social workers review and evaluate the applicants.

If everything goes well, a child is selected and taken to a home to test out living together; afterwards, the parents complete the necessary paperwork to make the adoption official.

In 2018, Family-Match began pairing families in Florida and Virginia. The governor of Virginia at the time, Terry McAuliffe, a member of the Democratic party, initiated a pilot program in response to the recommendation of a campaign contributor who was appointed as the state’s “adoption champion.” In Florida, where the child welfare system is privatized, regional care organizations quickly adopted the algorithm at no cost, thanks to a grant from a foundation established by the former CEO of the company that produces Patrón tequila and his wife.

After charitable funds became scarce in Florida, the state government took on the responsibility and granted a $350,000 agreement to Adoption-Share last month for their services.

Following the initial attempts in Tennessee and Georgia, further progress was made.

According to its tax returns, Adoption-Share has earned a total of $4.2 million in revenue since 2016, with $1.2 million reported in 2022.

Officials reported that during Virginia’s two-year trial of Family-Match, the algorithm resulted in just one documented adoption.

The Virginia Department of Social Services stated that the local staff did not find the tool to be useful. They also mentioned that Family-Match has not been effective in the state.

Traci Jones, an assistant director at Virginia’s social services agency, stated that social workers were confused by the algorithm’s apparent pairing of all children with the same set of parents.

Jones stated that despite requesting it, we were unable to obtain the algorithm.

In 2022, Adoption-Share received a larger contract from Virginia for a foster care program that utilizes their Family-Match application.

The initial pilot in Georgia was terminated in October 2022 due to the tool not functioning as intended. In the end, only two adoptions were made during the year-long experiment.

Social workers expressed that the tool’s matching suggestions frequently resulted in hesitant parents, causing them to question if the algorithm was accurately evaluating the adults’ ability to adopt the children.

Ramirez held a meeting with the governor’s office and advocated for a direct allocation from a statehouse committee, praising the tool as “an impressive accomplishment.” In July, the Georgia Department of Human Services entered into a new contract with Adoption-Share to once again utilize Family-Match at no cost, according to agency spokesperson Kylie Winton.

The child welfare system in Florida is run by multiple regional agencies that handle foster care and adoption services. After requesting public records about their use of the Family-Match tool, some of these agencies had mixed opinions and were unable to clarify the accuracy of the tool’s self-reported data, making it challenging to evaluate its claimed success rate.

According to a report from Adoption-Share, a child welfare organization in Pensacola, Florida, Family-Match is responsible for 603 placements and 431 adoptions over a period of five years in the entire state. This information was obtained by the Associated Press (AP) from the third-quarter report of the 2023 fiscal year.

Attorney Scott Stevens, who represents FamiliesFirst Network, informed AP in June that out of all the trial placements suggested by Family-Match since 2019, only three have not been successful. However, Adoption-Share’s documents, which were given to AP by Stevens, show that his agency has made 76 other placements through Family-Match that do not indicate formal adoption of the children. When asked for an explanation, Stevens was unable to provide details and directed further inquiries to Family-Match.

Ramirez refused to comment on the inconsistency but did admit via email that not all connections are successful.

Ramirez stated in an email that the process of adoption can be time-consuming, and it is ultimately the responsibility of agencies to make the final decision, taking into consideration the input of both the children and judges. Adoption-Share announced on their Facebook page on Sunday that they had achieved 500 adoptions in Florida.

Jenn Petion, the president and CEO of the organization that handles adoptions in Jacksonville, said she likes how the algorithm lets her team tap into a statewide pool of potential parents. Petion has also endorsed Family-Match for helping her find her adoptive daughter, whom she described as a “100% match” in an Adoption-Share annual report.

Petion stated that Family-Match helps social workers make improved decisions and matches. However, Family Support Services, the agency she represents, did not share any data on the effectiveness of Family-Match.

Over the last five years, the Children’s Network of Southwest Florida, located in Fort Myers, has reported 22 successful matches and eight adoptions through the use of the Family-Match tool. This is in contrast to the hundreds of matches and adoptions that their social workers were able to achieve without the tool.

Bree Bofill, the manager of the adoption program at Citrus Family Care Network in Miami, stated that social workers discovered the tool to be ineffective as it frequently recommended families who were not suitable matches.

Bofill expressed frustration with the algorithm stating that although it shows a match between children and families, in reality the families are not truly interested in adopting the children.

Bofill expressed that evaluating the effectiveness of the tool was challenging because social workers were occasionally instructed by Family-Match representatives to tell potential parents to sign up for the tool, even if it had no influence on the adoption. This allowed the algorithm to take credit for the match.

Winton, the spokesperson for the Georgia agency, informed AP about a related problem. Family-Match had the ability to take credit for matches if the child and parent were already in the system, even if the program didn’t create the match. According to a “confidential” user guide from April 2023 that was posted online, Family-Match instructed social workers not to remove cases that were matched outside of the tool. Instead, they were instructed to record the match in the system so that Adoption-Share could improve its algorithm and connect with the families involved.

Ramirez did not respond to Bofill’s statement, but stated via email that Family-Match’s reports accurately reflect the information entered by social workers into the system.

Authorities in Virginia, Georgia, and Florida expressed uncertainty about the methodology used by the tool to evaluate families, given the delicate nature of the factors influencing the algorithm.

Family-Match in Georgia is currently collecting information on potential sexual abuse experienced by foster youth, including the gender of the perpetrator and any criminal history they may have. This data also includes whether the youth identifies as LGBTQIA. Usually, this type of sensitive information is only accessible through highly protected child protective services case files.

The state of Tennessee used a modified version of the algorithm’s survey for potential parents, which included questions about their household income and their perceived level of creativity. They were also asked to indicate their agreement or disagreement with a statement about seeking guidance from a higher power, according to documents obtained by AP.

The Tennessee Department of Children’s Services raised concerns about the information that Family-Match wanted to gather during their review of the proposed assessment. State officials questioned the necessity of certain sensitive data points and how they affected the match score, as documented in an internal report where they shared their questions and feedback on the algorithm. Ramirez clarified that the agency did not dispute the accuracy of the survey, and stated that these discussions were a necessary part of the streamlining process.

According to Virginia officials, after entering families’ data into the tool, the data was then owned by “Adoption Share.”

In Florida, two organizations admitted to using Family-Match in an unofficial capacity without a contract, but declined to disclose how they safeguarded the data of children.

Ramirez did not disclose whether Family-Match has removed pilot data from their servers, but stated that her organization conducts compliance audits and adheres to contractual agreements.

Social justice activists and experts in safeguarding sensitive information have expressed concern about the growing use of predictive analytics by government agencies. These individuals argue that such methods can worsen existing racial inequalities and unfairly target families for factors outside of their control.

Adoption-Share is a member of a select group of organizations claiming that their algorithms can assist social workers in finding suitable foster or adoptive families for children.

According to Suresh Venkatasubramanian, a former assistant director at the Office of Science and Technology Policy under the Biden administration and current faculty member at Brown University, children are being used as experimental subjects for these tools. This is a major issue.

Adoption-Share is actively working to broaden its reach, targeting areas such as New York City, Delaware, and Missouri. These locations are currently evaluating the company’s proposal. Ramirez also recognized a chance last year to showcase Family-Match to the U.S. Health and Human Services Department, highlighting the value of their tool as a useful resource.

This year, Adoption-Share reached an agreement with the Florida Department of Health to use Family-Match in developing an algorithm with the goal of expanding the number of families who are open to fostering and/or adopting children with complex medical needs, as stated in state contracts. Despite multiple requests for comment, representatives from the health department did not respond.

Connie Going, a longtime Florida social worker whose own viral adoption story Ramirez has described as her inspiration for Family-Match, said she didn’t believe the tool would help such vulnerable children. Going said the algorithm gives false hope to waiting parents by failing to deliver successful matches, and ultimately makes her job harder.

According to Going, we have placed our confidence in something that is not completely beneficial, resulting in wasted time for social workers and negative emotional impacts on children.


For any inquiries, please reach out to AP’s worldwide investigative team at [email protected] or through the link