Benefits Data Trust wind-down

Benefits Data Trust was a Philadelphia nonprofit whose navigators, its call-center staff, helped low-income people apply for public benefits. They screened each caller, working out which of many programs the caller likely qualified for, with custom tools such as Community Prism. It was not an algorithmic decision or risk-scoring system, and the failures here came from its closure.

How it worked

Benefits Data Trust was founded in 2005. Under data-sharing agreements with state and city agencies, it identified people likely to qualify for benefits and contacted them. This is called targeted outreach.

It ran a live call center. It held the City of Philadelphia's benefits contract, called BenePhilly, from 2008, plus state contracts. Its navigators screened each caller for many programs at once. They stayed on the line until the application was submitted.

In 2023 it reported helping more than 120,000 people get about $182 million in benefits. It reported more than $10 billion over its lifetime. Philanthropy paid for roughly two-thirds of its work, and government contracts for about one-third.

How it closed

In June 2024, days after parting with its chief executive, the board voted unanimously to wind the organization down within 60 days. The Philadelphia Inquirer reported the chief executive's departure about two weeks before the announcement. A Technical.ly post-mortem put it at roughly a month.

The board set the 60-day window itself, by its own vote. The board's only official explanation was "a perfect storm of circumstances." At the end of 2023 the organization held about $12 million in reserves free of restrictions. It projected about $32 million in 2024 revenue, and the board had been shown an expectation of breaking even.

A search for a partner to take over the work failed. In words the sources quote, "we had exhausted all avenues." The Philadelphia call center took its last call on July 25, 2024. The organization closed on August 24, 2024.

Its layoff notice to the state, which employers must file before mass layoffs, covered 273 employees. The lowest-paid call-center workers were the most affected.

What each side says

A Technical.ly post-mortem, sourced from staff, blamed a built-in gap between costs and income. Each successful application reportedly cost a few hundred dollars in outreach and call-center time, against roughly $2 for online self-service tools. One insider put monthly expenses at $2.5 million against $1 million in revenue. These figures rest on anonymous staff sources, so they are staff claims, not audited figures.

About 120 current and former employees signed an open letter demanding an explanation. The founder said there was "no malfeasance, misappropriation of funds, or fraud." Former employees contend the 60-day timeline was avoidable. That claim is contested.

The Pennsylvania Attorney General's office said it reviews charity failures for breach of fiduciary duty, the legal duty of a charity's leaders to act in its interest. As one former executive put it, the high-tech nonprofit "died the old fashioned way."

Who was left without a successor

The closure left active government partnerships with no designated successor. For the Pennsylvania Department of Aging, Benefits Data Trust had processed nearly 48,000 applications from 27,018 households in its final year. They included nearly 9,000 applications for PACE, a prescription-subsidy program, and 23,000 Medicare drug-subsidy assists. The department leaned on a subcontractor.

Benefits Data Trust was exceeding the BenePhilly contract's performance requirements through the first half of 2024. That work was redistributed across six partner agencies. A Tennessee SNAP food-assistance contract was reported ending with no replacement.

Help with applications moved to harder-to-use channels. One nonprofit took a state SNAP-enrollment contract the following January and hired former staff. Other providers brought the work in-house. Referral waits were reported at several months.

Kristin Romens of The Pew Charitable Trusts called this a "cascading effect." Pew, a funder of Benefits Data Trust, put the benefits going unclaimed by eligible Philadelphians at more than $450 million a year.

What happened to its data and tools

The fate of Benefits Data Trust's custom screening applications, toolkits, and chatbots was reported unresolved at closure. Pennsylvania's own Compass system had already replaced its Community Prism tool. The sources do not describe Compass.

The case file says the closure orphaned the agency data-sharing links. The sources read for this case do not say what became of the data itself.

What this network is drawn from

This network follows the pattern the case file describes. It does not reconstruct the organization, its call center, or Community Prism. It shows the screening tools, the navigators, the agency data and records, the board, and the partners who relied on the work. The households served are outside the network.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. Closing a pathway means mistakes stop passing along it. The work along it goes on.

This case has a budget of 11 units. Each tool costs the same at every target level.

Explore (No Targets) sets no targets. Service Targets Only asks for two things. The network must be self-correcting, meaning its mistakes are corrected rather than building on each other. The screening tools must also be helping the navigators' work. Before any tool is used, the screening tools are helping, but the network is at a tipping point, not self-correcting.

Five tools meet Service Targets Only on their own at their standard setting: Peer sharing rules, Escalate checks, Mark AI-written records, Store less data, and Gate record entries. The cheapest cost 2 units. Escalate checks raises checking when trouble is flagged. Mark AI-written records labels what the screening tools produced in the records. Gate record entries requires sign-off before anything enters the records. Review on schedule does it at its stronger setting, for 4 units. Lingering effects is a Lab setting in which mistakes and reliance on the system stay after their cause is gone. The Lab starts with it on. Upgrade model does it at its stronger setting, for 5 units, while lingering effects are on. With lingering effects off, Vet connections also does it alone, for 3 units, and Upgrade model does not. More than 400 different sets of tools within the budget meet this level.

Under Service and Safety Targets and All Governance Targets, the targets can be met within the budget. Both levels ask you to close every failure pathway, among other targets. Four are open before any tool is used. They are Screening results to navigators, Agency data and rules used in screening, Navigators record applications and outcomes, and Partners send people to navigators.

The cheapest way costs 9 units. Peer sharing rules closes Partners send people to navigators. Escalate checks closes Screening results to navigators. Mark AI-written records closes Agency data and rules used in screening. Gate record entries or Store less data closes Navigators record applications and outcomes.

Every way to meet either level includes Peer sharing rules and Escalate checks. Service and Safety Targets can be met with 10 different sets of tools. There, Vet connections can close the agency data pathway in place of Mark AI-written records. All Governance Targets can be met with 5 sets, and each includes Mark AI-written records.

Upgrade model is in no set that meets either level. Review on schedule is in none that meets All Governance Targets.

More is not better here. Every tool at its strongest setting costs 36 units, more than three times the budget. It closes every failure pathway, but the screening tools then no longer help the work, so it meets no level's targets.