Domain Atlas / Child welfare & family services
Eckerd Rapid Safety Feedback: origin and spread
Work with this case in the PAN Lab ↗
Eckerd's Rapid Safety Feedback spread from Hillsborough County, Florida to child-welfare agencies in several states — promoted on the vendor's own reported gains and highlighted as “innovative” in a 2016 federal commission report — years before an independent 2022 peer-reviewed evaluation found the process did not lower repeat high-severity maltreatment among children identified as high risk (a joint odds ratio of about 1.05).[4]
What happened
Eckerd (then Eckerd Youth Alternatives / Eckerd Kids) became Hillsborough County's lead child-welfare agency in 2012, after at least eight children had died under the prior provider's supervision, and in January 2013 launched Rapid Safety Feedback — a predictive tool built with the analytics firm Mindshare Technology. The tool mined a jurisdiction's own case records to flag open cases whose features resembled past serious-injury and fatality cases, routing them to quality-assurance reviewers who coached frontline workers rather than to any automated action. Eckerd reported that Hillsborough had no maltreatment fatalities over a roughly three-year period and that case-practice measures improved on average about 22 percent; those figures were the company's own, reported through the press, and Eckerd itself said it did not claim the analytics caused them, while critics argued no evidence tied the gains to the tool. On the strength of that reported success, the tool was adopted by agencies in several states; a 2016 report by the federal Commission to Eliminate Child Abuse and Neglect Fatalities highlighted the approach as "innovative" with a Hillsborough case study, by which point it was already in use across those states. A 2022 peer-reviewed evaluation later found the process did not lower repeat high-severity maltreatment among children identified as high risk, reporting a joint odds ratio of about 1.05.
The sociotechnical reading
Most cases in this Atlas were scrutinized after they scaled; this one scaled on an endorsement and a claim no one had independently tested. In system-map terms the load-bearing failure was not a wide error pathway or a broken override but an empty one: the independent-validation check never fired. A single proprietary profile, whose weights no caseworker or family could see, was copied and tailored jurisdiction to jurisdiction, so its blind spots travelled with it instead of being re-tested on arrival — a monoculture spread across agencies. The legitimacy that drove that spread came from the vendor's own before-and-after numbers and a federal commission's stamp, not from evidence the tool changed outcomes; when an independent evaluation finally ran, it found no effect. The governable moments were upstream of any dashboard: procurement that demands documented, checkable performance claims; a sign-off that asks "on what evidence?" before the next agency adopts; and a re-evaluation cadence that treats a launch endorsement as a question rather than an answer. It is the Atlas's clearest case that the most dangerous number in a deployment can be the one everyone trusts because someone respectable repeated it.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.