Skip to main content

Southern New Hampshire University • 6 min read

AI-Powered Prior Learning Assessment

Designing a machine learning tool with Google, and taking multi-course matching from under 30% to over 90% task success.

90%+

Multi-course match success

Up from under 30% after I redesigned the matching flow

81.8%

Of testers saw clear value

Said it would improve on their current process

200+

Colleagues briefed

I presented the research and design at the company retreat

AI-Powered Prior Learning Assessment
Role
Lead Product Designer
Timeline
Aug 2022 – Aug 2023
Team
6 people
Partners
Google and the Bill & Melinda Gates Foundation

Adult learners bring credit-worthy experience from work, the military, and certifications. At SNHU, evaluating it meant emails and comparing every document to the course catalog by hand. With Google and the Bill & Melinda Gates Foundation, we set out to build a machine that does the matching. I led the design from the first interviews to the alpha launch.

Challenge

A slow, manual process

  • Requests arrived by email, with no central record of history or notes
  • Analysts checked every document against the catalog by hand
  • Learners weren’t getting all the credit they had earned

Solution

One AI-assisted workspace

  • Every data source and the approval process in one tool
  • Google’s machine learning parses documents and suggests matching courses
  • Staff compare suggestions side by side and choose the closest match

Build for staff, design for students

I recommended building for staff first, so student features could grow into the same product.

SNHU’s mantra was to think of the student first, and I wrote a full feature spec for the student experience. Then I recommended we build for staff first. A service blueprint showed that staff who know the tool become the best support learners can get, and a staff-only release gave us a clear audience to test with. Student and employer features could grow into the same product later, instead of separate apps.

MVP flow diagram organized into three numbered epics (Capture, Analysis, Admin), with screens connected by user flows and data transfers.
The MVP flow across three sub-epics: Capture, Analysis, and Admin.

Flow first, then the look

Complex interactions had more than one possible solution, so I started on paper. The first sketches explored a stepper, which the final version dropped. Low-fidelity wireframes then broke each complex interaction into its own parts. With our junior designer, I prototyped the high-fidelity wireframes as we designed them, so testing could start sooner.

Two pages of pencil sketches: PLA Management laid out as a single page, then as a stepped flow, with notes beside each screen.
Sketches comparing a single page with a stepped flow.
Low-fidelity grayscale wireframes of the submit-review steps, connected as a flow, with sticky-note comments.
Low-fidelity wireframes, laid out as a flow.
High-fidelity grayscale wireframes of the Start a new review screen at several upload steps.
High-fidelity wireframes for each upload step.

The design system team set the fonts, colors, plus card designs. Their components dropped straight in, so I could test several high-fidelity ideas quickly. Where the system fell short, I came up with new pieces, then designed them with the team: a card subline for an experience’s type or completion date, a filter slide-out, plus the experience drawer. I presented each round to the departments for feedback.

An early dashboard design: a welcome message, cards for recent reviews, then a list of all active reviews with status badges.
An early dashboard idea, with recent plus active reviews.
An early Find & Match design: a search field, results, then a bar of two selected experiences with an option to add another.
An early take on Find & Match, with selected experiences pinned below the results.
An early illustrated dashboard design: a large welcome banner with an illustration of two people talking, above a searchable list of reviews.
Another dashboard idea, illustrated by a designer on the team.

The machine suggests, people decide

Credit decisions affect people’s degrees, so the machine doesn’t award credit on its own.

It suggests SNHU courses and shows its reasoning. Staff make the call. For the comparison, I chose a table like the ones people use to weigh software or pricing plans, so staff needed no new training. It also lets them scan several courses at a glance. The design system team designed its look.

How staff can trust a suggestion

  • Every suggestion carries a high, medium, or low strength of fit
  • Matching skills, competencies, and outcomes are counted side by side
  • Staff choose the closest match themselves
  • If nothing fits, they flag the experience instead of forcing a match
A single document comes in, breaks into parts, then each part gets sorted and matched. The machine’s parts are interchangeable, so each run is random but sized to the wait. My concept, designed and animated by our motion designer.
Strength of fit table comparing an outside course with three SNHU courses by fit rating, level, credits, skills, competencies, and outcomes.
Strength of fit: suggested courses side by side, each with a rating and the evidence behind it.

Multi-course matching: from under 30% to over 90%

Validation with 13 staff, 10 tasks, and 5 scenarios
<30% → 90%+
Multi-course match success
From the first round of testing to the next, after I moved matching behind the results view

What changed

  • Matching moved after the results view, out of the upload flow
  • Instant feedback on every uploaded document
  • Every required action reviewed in one place
  • Harder matches handled at each person’s own pace
  • Complex reviews saved and finished later
The Find and match screen: a drawer listing two added experiences, plus a card to upload or add more.
Each upload shows up right away in the experiences drawer.
Close-up of the drawer: the added card slides in ahead of the others, then its green outline fades.
Session report showing 6 experiences found, 3 matched, 12 to 15 potential credits, and a list of four required actions.
The session report collects every required action in one place.
Saved sessions list: a search field, session cards with status labels, plus sort and filter controls.
Saved sessions let staff finish a complex review later.
Where the first round struggledShare of participants for each of the 10 tasks. Tasks 8 and 9 asked them to match one course, then several.
  • Matching tasks
  • Other tasks
Show the data as a table
Usability test results by task, as a share of participants
TaskRated difficultNeitherRated easyCompleted
Task 110%10%80%80%
Task 210%10%80%90%
Task 310%20%70%90%
Task 410%10%80%100%
Task 510%10%80%90%
Task 610%10%80%100%
Task 710%20%70%100%
Task 8 · match one course40%0%60%80%
Task 9 · match several50%0%50%30%
Task 1030%0%70%90%

Rounded to the nearest 10%, as charted in the original research readout.

Staff from Admissions, the Registrar’s office, and other departments found the first prototype intuitive. Matching was the exception: 80% completed a single match, but fewer than 30% could match several courses at once, and 40–50% rated matching difficult.

“I know if I had anything like this when I was in ESS, I would have been ecstatic.”

— Usability test participant

Prototype for each audience

I built two prototypes for two audiences: an alpha that scoped the MVP for engineering, and a beta that showed stakeholders the full V1. My front-end background let me work through problems with our tech leads in their terms. I split the files by sub-epic, redlined every component state, and joined grooming twice a week. With Google, I worked out how matches are scored and what document parsing needed.

The beta prototype’s flow map: dozens of connected screens, with red regions marking the flows the alpha prototype left out.
The beta prototype’s flow map. Red marks what the alpha left out.
Redline specs for the upload card’s states (default, hover, focus, active), with spacing and type notes.
Upload card specs for every state. The enlarged view shows the full sheet, including the experience drawer.

Outcomes

Before launch, I presented the research and design to 200+ colleagues at the company retreat, including senior leadership. We also demoed the tool for every department. The alpha went to Admissions and Registrar staff in August 2023. It’s an internal tool and I left SNHU after the alpha, so I don’t have production numbers.

How we planned to measure it

  • How often staff agreed with the machine’s matches
  • New matches staff wouldn’t have found on their own
  • Where the match staff chose ranked among the suggestions
  • Time to evaluate each request

This tool is an incredible value-add for my role as a Prior Learning Analyst. It will impact my day-to-day in such a positive way.

— Prior Learning Analyst, alpha user

What I learned

  • Utilitarian tools still need moments of joy. Course matching can take a minute or two, and staff see that wait often. My concept: a single document comes in, breaks into parts, then each part gets sorted and matched. The parts of the machine are interchangeable, so every run plays out at random, yet each is right-sized to fill however long matching takes. I fleshed it out with our motion designer, who designed and animated it as a Rube Goldberg machine.
  • Illustrations matched to each task, drawn by another designer on the team, did more than add visual appeal. They made a utilitarian product feel light and friendly.

Team

I led design for the whole project, starting alone with the product owner on the research, problem statements, feature spec, and first flows. A junior designer joined after the low-fidelity stage, working with me on research and high-fidelity design. The design system team owned the visual language, while teammates created the illustrations plus motion. I was promoted to Lead Product Designer in June 2023, once the designs were complete and the tool was in development.

My Role
Lead Product Designer: research, strategy, UX/UI design, prototyping, plus the handoff to engineering and Google
The Team

Nancy D.

Product Owner

Nicole C.

Product Manager

Joshua H.

Data Analyst

Angelika K.

Researcher

Krista S.

Junior Designer

Next case study

Butte College • Mobile-First UX & Front-End • 2 min read

Search for Classes

A mobile-first redesign with no rebuild budget: 400% more mobile class searches and 88% fewer support tickets.

As Butte College’s one-person web team, I shipped it in under two months, with CSS and targeted JavaScript on top of the existing tool.

View Case Study