·

Product Design · Fintech · Lead Designer

Workplace benefit enrollment redesign to increase auto-verification

Fidelity lost its loan verification partner, so I reframed enrollment as a read-back: show us your loan, we read it back instantly. The result: 67% of loans auto-verified, up from 0, and no 5-day wait for most participants.

Jump to the design
67%
of loans auto-verified, up from 0
30%
more enrolled participants
5 day
manual loan review removed for most users
Benefit enrollment redesign preview: get started, the recovery screen when an account number is missing, and the enrolled success screen

TL;DR

THE PROBLEM

Fidelity lost its automated loan verification partner. Failed documents went to a 5-day manual review, and enrollees dropped off.

WHAT I DID

Reframed enrollment as a read-back: you show us your loan, we read it back instantly.

WHAT CHANGED

More loans verified automatically, more people finishing enrollment, and the 5-day manual review gone for most.

01 · CONTEXT

What broke, and why it mattered

Fidelity lost its automated loan verification partner due to legislative changes, disrupting the student debt benefit enrollment process. When uploaded documents failed verification, they were sent to a manual review team with a 5-day response time via chat.

As the market leader in student debt workplace benefits, Fidelity faced mounting pressure as enrollment rates dropped and operational costs threatened profitability. The business model — priced per enrolled participant — made every abandonment costly.

Every document that fails auto-verification goes to a manual review, and that cost grows as the product gains users, so the goal is an estimated 85% auto-verification rate once all proposed solutions are live.

Lost data Ops cost Users can’tfind Enrollees lost aggregationpartner goes up the rightdocuments to a 5-day review
  1. Lost dataaggregation partner
  2. Ops costgoes up
  3. Users can’t findthe right documents
  4. Enrollees lostto a 5-day review

02 · HOW I FRAMED IT

From submit-and-wait to a read-back

Participants treated enrollment like mailing paperwork: upload something, then wait days to learn if it worked. That model hid every mistake until it was too late to fix. I reframed it as a conversation: you show us your loan, we read it back to you instantly, and you fix anything that's off before you leave. Guidance, servicer selection, and PDF extraction all serve that one idea. We ruled out adding more instructions up front, because users skipped text they couldn't act on yet.

I aligned partners on three goals (raise auto-verification, lift completion, cut drop-off) and a phased plan that shipped value every few weeks: guidance first, then servicer selection, then the feedback loop.

FEEDBACK LOOP 5-day manual review Beginenrollment Upload adocument Correct it in theflow See what’smissing Enrolled! The old path when adocument failed Extracted data,read back No 5 daywait anddrop off FEEDBACK LOOP 5-day manual review Begin enrollment Upload a document See what’s missing Correct it in the flow Enrolled! The old path when a document failed Extracted data, read back No 5 day wait and drop off

03 · KEY SOLUTIONS

One enrollment, four moments

Follow one participant through enrollment in the prototype below. Each moment shows what broke, what I tried, and what it changed. You can choose the happy path or the feedback loop path.

Dynamic benefit guidance up front

What was wrong

Our previous guidance page had large text blocks users likely skipped.

What shipped

Working with our enterprise patterns team, we designed an interactive component to encourage engagement with key information: benefit fit and document requirements.

In-market learning

Users dropped off when we linked out to plan details mid-enrollment. We moved the link to the guidance page, where people learn about the benefit, so enrolling never takes them out of the flow.

What I triedFrom where we started to what shipped · select one to preview it

NEGOTIATION OF THE SCOPE

More scope means more personalized guidance and higher automation

Trading off the vision for what we can accomplish at first

With a vision that covered every use case, I led the negotiation across engineering, product and business partners to decide what could go into the first release. Engineering weighed in on what was feasible, product set its requirements, and I worked to find a middle ground that still balanced what users needed.

First release

5 scenarios
  • Not a PDF or unreadable

    Return to upload with a message, or send the document for manual review.

    First release
  • Servicer not found

    Return to upload with a message, or continue with the servicer they selected and the account number we read.

    First release
  • Account number missing

    Return to upload with a message, or enter just the account number.

    First release
  • Balance missing

    Return to upload with a message, or enter just the balance.

    First release
  • Participant name missing

    Return to upload with a message, or review the extracted data.

    First release

Tradeoff: in the vision, cut from the first release

3 scenarios · slotted for a future release
  • Servicer doesn’t match

    Update the servicer to match the document, or upload a different one.

    To be incrementally released
  • Loan paid off

    Upload another document, or add a different loan with a balance.

    To be incrementally released
  • More than one issue

    Upload another document, or add the missing balance and account number.

    To be incrementally released

WHAT WE LEARNED AFTER LAUNCH

Name mismatches were the biggest reason documents failed

A participant’s name on their document didn’t match the one we had in our system, and our feedback page was generic in that case. So we used in-market data to choose the next release: a feedback page built for name mismatch, which helps people either correct the name or confirm the name on the document is theirs.

We expect this page to raise the auto-verification rate by about 10 points, to 77%.

  • Last name mismatch

    New: confirm the loan is yours, or send it for review.

    Next release
  • All other name mismatches

    The existing page, which stays as is.

    Unchanged

04 · OUTCOMES

How the results built

Each release raised auto-verification on its own. Together they got 30% more people through enrollment and removed the 5-day manual review for most of them.

Auto-verification rate 0%20%40%60%80%100% 85% estimate 12%22%67%77% Start 01Before youstart 02Add yourloan 03Fix it inthe flow NextName mismatchpage LaterMorescenarios Auto-verification rate 0%20%40%60%80%100% 85% estimate 12%22%67%77% Start 01Beforeyou start 02Add yourloan 03Fix it inthe flow NextNamemismatch LaterMorescenarios

A PATTERN THAT CARRIES INTO PAYMENTS

The same loop will verify payments

After enrollment, participants upload payment documents to get credit for the benefit, the same way they upload loan documents.

05 · KEY LEARNINGS

Three things I'd take into every project from here.

01

Test in Market

In fast-moving markets, testing solutions in-market can be advantageous if you have development capacity to iterate quickly.

02

Goals Over Requirements

Ground work in specific goals over lengthy requirement documents. For every decision, ask if it will move the needle to meet your goal.

03

Vision First

Starting a project with technical restraints, even in a time crunch, can be limiting. Start with the best solution for the user, then scale back to what is achievable now.

Next project

Task Usage Predictor

View next case study