All work

Case study 02 HHS / Administration for Children and Families

Shepherd Case Management System

Redesigning the system that connects survivors of human trafficking to federal benefits.

Client
HHS / ACF, Office on Trafficking in Persons
Role
UX lead, case management redesign
Platform
Cloud, three microservices
Timeline
2020 – 2021
50% faster access to benefits for survivors
applications processed, same staffing
2 national awards, 2021

Overview

Certification is the step that unlocks federal assistance. It ran on email.

The Office on Trafficking in Persons certifies survivors so they can reach housing, legal aid, benefits, and case management. In 2018 the office received three times more service requests than in any previous year, and the process behind certification had been built for a smaller caseload.

Process before and after

From email attachments to a single access point

The old path crossed four handoffs before a case existed.

Before

  1. Password-protected document by email
  2. Manual printing
  3. Digital signing and scanning
  4. Manual entry into a tracking database

Slow, hard to keep secure, and burdensome for the office and the survivor alike.

After

One portal. Submit, track, review, and issue in one place.

  1. Request service
  2. Review service
  3. Case service

Three separately deployed services behind one consistent interface.

The path a request took before the redesign, and after.

The problem

A document workflow with people attached to it.

Before the redesign a request arrived as a password-protected document attached to an email. Case specialists printed it, signed it, scanned it, and entered the details by hand into a tracking database. Correspondence, consultations, and referrals were managed separately through more email. The process was slow, difficult to keep secure, and burdensome for the office and for the person waiting on the other end of it.

The burden fell hardest on the person the system existed to help. Because each agency ran its own application process, survivors had to recount their victimization repeatedly, to separate offices, to qualify for services that all rested on the same underlying determination. Proving what happened to you, again, to another stranger, is not an administrative inconvenience.

Journey, before and after

The same request, through two different systems

Three parties. In the old process the survivor’s account was told again at every boundary.

Before

Survivor and requester

  1. Email a password-protected document
  2. Call the office to ask about status

OTIP

  1. Print
  2. Sign
  3. Scan
  4. Re-key into tracking database
  5. Manage by email thread

Other federal agencies

  1. Each agency runs its own process, and the account is told again at every one

Slow, hard to keep secure, and the burden of repetition falls on the person least able to carry it.

After

Survivor and requester

  1. Submit once, in six saved sections
  2. Track status in real time

OTIP

  1. Review in queue
  2. Expedite or consult
  3. Refer to grantee
  4. Submit for approval
  5. Issue letter

Other federal agencies

  1. Verify letter status online, no phone call required

One account, told once. A common certification the participating agencies accept.

Reconstructed from the system’s entry points and from ACF and REI published descriptions of the process.

That is the operational problem. The design problem underneath it was different.

In most software a mistyped field is an annoyance. Here it was a delay for someone who had already waited too long.

And it sat inside a case file that four federal agencies would later need to reconcile, under policy requirements that touched nearly every screen.

Constraints

What could not change.

Policy fields could not be removed
Every required field existed because a statute or regulation put it there. The work was not simplification by deletion. It was making sure the requirements did not bury the actual job.
Three separately deployed services
Request, Review, and Case each shipped on their own schedule. Interaction patterns and role-based permissions had to stay consistent even when the pieces behind them moved independently.
Four agencies downstream
Homeland Security, Justice, Housing and Urban Development, and Labor all touch these cases. A record that made sense only inside ACF was not good enough.
Two audiences, one system
A member of the public submitting on someone else’s behalf, and a trained case specialist working a queue. The same data, entirely different needs.
Confidentiality under load
The old process kept victim information secure by passing password-protected files between inboxes, which was burdensome to sustain and fragile at volume.

Whatever replaced it had to be safer without being slower.

Discovery

Personas first, then the journeys they take.

I built personas for the people the system actually serves: a member of the public submitting on a child’s behalf, a case specialist working a queue, a funded grantee receiving referrals, a downstream agency needing to verify a letter. Then I mapped each of their journeys through the existing process.

Laying those journeys side by side is what produced the reframing everything else follows from. Every one of them was organized around a document changing hands. None of them were organized around a decision being made. The requester was assembling a package. The case specialist was processing a package. The agencies downstream were re-opening a package that had already been assembled twice.

That is why the redesign starts with the decisions rather than the forms.

Approach

Reorganized around decisions, not documents.

I reorganized the journey around the decisions a case specialist actually makes, rather than around the documents the agency needed filled in.

The old process was a document workflow with people attached to it. The new one is a decision workflow with documents attached to it.

It also set the goal for everything downstream: one account, told once. Consolidating services and support from multiple federal agencies into a single access point meant a common certification that participating agencies would accept, so a survivor no longer had to prove their victimization separately to each one. That is a policy achievement as much as a design one, and it took the joint team to get there. What design owned was making the single account complete enough, and legible enough, that four agencies could act on it without sending the person back to the beginning.

Rebuilding intake

The front door, and the place a person in crisis meets the federal government.

I broke one long form into six sections, sequenced so a requester is asked what they already know before they are asked what they may have to go and find out.

Intake structure

One long form became six answerable sections

Sequenced so a requester is asked what they know first, and the hardest questions last.

  1. Requester Who is submitting, and on whose behalf
  2. Child Identity and location of the young person
  3. Concerns What indicators have been observed
  4. Narrative The account, in the requester’s own words
  5. Experience Structured detail on the trafficking
  6. Consent Information sharing and permissions
Progress is saved between sections, so a requester who does not yet have an answer is not blocked from starting.

Requester information comes first because anyone submitting knows it. The child’s details second. Then the concerns, what has been observed. Then the narrative, in the requester’s own words, because a structured field cannot hold an account of what happened to someone. Then the structured detail. Consent last, once the person understands what they are consenting to.

Designing for the cost of a mistake

Three safeguards, where an error carried the most weight.

Error prevention

Three patterns, applied where a mistake costs the most

A mistyped field here is a delay for someone who has already waited too long.

  • Safe defaults

    The low-risk option is the one that takes no effort. Nothing consequential happens because a field was left alone.

  • Confirmation

    Anything that cannot be undone asks first, and asks in language that names what is about to happen.

  • Audit trail

    Every state change is reconstructable later, so a case can be reviewed across agencies without guesswork.

Caseworkers estimated the patterns cut data-entry errors at critical decision points by around 30 percent. That figure comes from caseworker feedback rather than instrumentation.

Caseworkers estimated these patterns cut data-entry errors at critical decision points by around 30 percent. That figure came from their feedback rather than from instrumentation, and it is stated that way everywhere it appears.

Across three services

A user moving between them should not be able to tell.

Request, Review, and Case were separately deployed. That meant interaction patterns and role-based permissions had to be defined once and hold everywhere: the same confirmation behavior, the same status language, the same rules about who can see and do what, regardless of which service was rendering the screen and regardless of which of them had shipped most recently.

Roles and permissions

Five roles, one record, different views of it

Defined once so behavior stayed identical across the Request, Review, and Case services.

Which of the five roles can perform each of the seven system capabilities
Role Submit a requestTrack own statusReview a caseRefer to granteeExpedite in emergencyApprove and issue letterVerify letter status
Requester public, or on someone’s behalf Full access Full access No access No access No access No access No access
OTIP case specialist works the queue No access Full access Full access Full access Full access No access Full access
OTIP leadership approval authority No access Full access Full access No access Full access Full access Full access
Grantee organization funded service provider No access No access Scoped to their own referrals No access No access No access No access
Benefit-issuing agency downstream federal partner No access No access No access No access No access No access Full access
  • Full access
  • Scoped to their own referrals
  • No access
The hard part was not the grid. It was keeping it identical across three services that shipped on different schedules.

What came of it

Half the wait, twice the throughput, two national awards.

Program outcomes

What the program reported after launch

Published by ACF and REI Systems. These belong to the joint team across design, engineering, cloud, and delivery.

  • 50% faster access to benefits for survivors
  • benefit applications processed, same staffing
  • 10 hrs of staff time saved for each victim processed
  • 75% reduction in pending caseload by FY2020

Recognition

  • ACT-IAC Igniting Innovation May 2021 · selected from 169 nominations
  • FedHealthIT Innovation Award July 2021

The 30% figure is estimated from caseworker feedback, not instrumented measurement.

The joint REI Systems and ACF team delivered a system that cut the time for survivors to reach benefits by half and let the office process twice as many applications with the same staffing. ACF reported roughly ten staff hours saved for every victim the system processes, more than 800 hours in fiscal 2019 alone, and a 75 percent reduction in pending caseload by the end of fiscal 2020.