Niche is a two-sided marketplace. Schools pay to reach prospective students. Students use Niche to discover and research colleges. The business works when those two sides connect meaningfully.

Higher ed was facing a demographic cliff. Niche was already great at the moment a student first showed interest in a school: a follow, a save, an info request. The problem was everything after. Students were signaling intent, then stalling out before applying, before responding to outreach, before accepting a Direct Admissions offer. Schools were watching qualified interest leak out of their funnels and had no way to act on it. The initial brief leaned toward a school-facing messaging tool. My first move was to reframe it.

I ran discovery research with 10–12 schools and pulled CRM integration data showing exactly where interested students were going dark: between the follow and the application, between school outreach and a response, between a Direct Admissions offer and an accept. Affinity mapping across all three sources made the pattern obvious, students weren't dropping off because they'd lost interest. They were dropping off because the process felt opaque and overwhelming.

Students described the window between "I like this school" and "I've taken a real next step" as a black box, no map of remaining steps, no sense of what was meaningful versus marketing, no way to tell whether silence meant "you're fine" or "you missed something." Multiply that by the 5–10 schools each student was following, and the anxiety was paralyzing.

The reframe didn't land in a single meeting. I made the case in pieces across three review sessions: the CRM data showing where interest died, then the affinity map of student anxiety quotes, then the auto-complete risk memo. By the third session, the question had changed from "how do we build the messaging tool?" to "how do we build the student journey?"

A task-based experience that meets students at the moment they show interest in a school, then walks them forward step by step. Schools define what "next" looks like for them. Niche surfaces it in a format students recognize: a clear list of meaningful next actions, the trusted Inbox channel for school messages tied to each one, and progress the student owns.

The product's value to schools wasn't uniform. Some needed basic ways to reach interested students. Others needed customizable journeys and a portal to manage them. Others wanted full personalization and expanded reach. PM led the commercial structure and pricing case; my contribution was the feature mapping, translating the school-maturity patterns I'd seen in research into which capabilities sat in which tier, and where upgrade pressure would naturally land. The product became a new revenue category that didn't exist before the project started.

PM owned the commercial structure and school-side stakeholders. I owned the strategic reframe, ran the school discovery interviews, synthesized across CRM data and existing Direct Admissions student research, and designed the experience end-to-end. Our data lead built the funnel instrumentation that let us measure influence honestly: not "did Yield touch this enrollment," but "did the student act because of it."

The harder fight wasn't the reframe. Leadership was leaning hard into Inbox as Yield's primary delivery channel because schools loved Inbox. But Inbox's value to students was its signal-to-noise ratio. Pumping more partner-driven content through it was the fastest way to recreate the email problem Inbox had just solved. I argued Yield should use Inbox surgically, only for messages tied to a specific student task, and keep the rest of the experience on the Yield surface. That position took two months of repeated framing to land. The Inbox channel stayed clean.

Half the value of designing this product was choosing what it wouldn't do. Three calls shaped v1 more than anything we shipped.

The cycle metrics matter, but the more durable outcome was what Yield seeded for the rest of the org. Two patterns from this project are now in use beyond it.

The Inbox-channel tension was the fight I should have anticipated, not reacted to. I framed Yield's reframe well, then walked into the channel debate without a position prepared. Two months of repeated framing got it landed, but if I'd written the channel-integrity memo on day one, it would have landed in two weeks. Strategic foresight is the staff move I keep practicing.

The commercial tier structure was a late addition. Tier construction wasn't my deliverable, but PM was figuring out the structure in parallel without the school-maturity patterns from research in hand. Next time I'd hand that synthesis to PM in week two, not week ten, so the tier story emerges with the design instead of bolted on.

Niche had the data showing student interest. Schools had real updates to send: direct admissions offers, application deadlines, campus visits and events. But the communication layer between them was fractured, emails going to spam, messages sent at the wrong moment, no centralized place for students to manage outreach from schools they actually cared about.

The initial brief was to improve message deliverability. I argued it was a bigger structural problem than that.

Students applying to college receive enormous amounts of outreach. The problem wasn't a lack of messages. It was that students couldn't distinguish the ones that mattered. We interviewed 13 students and pulled behavioral data on what they actually opened. The pattern was sharp: students glanced at messages but only clicked in when they recognized the school name. Everything else got filtered as noise, and once a channel earned that filter, it didn't get unfiltered.

We landed on a two-tier inbox. The primary view stayed scoped to schools the student had already engaged with, protecting the signal. A secondary filter let students opt into messages from unengaged schools, on their own terms. Sales got the partner value they needed. Students kept the inbox they trusted.

Before locking the architecture, I prototyped the two-tier flow and tested it with students. The question was whether the secondary filter would feel like punishment ("schools you don't care about") or autonomy ("messages I'm choosing to see"). We iterated on the label and entry point until students described it the second way without prompting.

PM partnered on strategy and pushed the partner-value lens. I anchored every decision in the student experience. By the end, students had learned Inbox was high-signal: every message they opened reinforced the trust. That trust was the channel Yield would later need to convert.

Saying no to a bad proposal is easy. Saying yes to a better one is harder, and more useful. Sales had a real revenue ask. The Director of Product had a real constraint. The student need wasn't negotiable. The work wasn't designing the inbox; it was finding the third option that didn't betray any of them.

Most product disagreements aren't user-vs-business in any clean way. They're framing problems. The job is to find the design that makes both sides true.

Niche operates as a two-sided marketplace. Schools pay to connect with prospective students. For that to work, students need to engage and convert into leads schools care about.

Heading into 2023 planning, the graduate segment was our lowest-performing vertical. Traffic was thin. Engagement was thinner. Selling products to grad schools was hard because we couldn't show the audience.

To unlock business value, we needed to understand what grad searchers actually wanted from us, and how we could help them connect with schools in a way that felt useful.

The brief was to fix the grad segment. PM's first instinct was a redesign sprint. I pushed back. The data we had, thin traffic, low time-on-page, high exit rates, told us students were leaving, not why. Redesigning against that would just give us nicer exit screens. I made the case that a short discovery sprint would tell us more than eight weeks of redesign work built on the wrong assumptions. PM agreed. We ran discovery first.

I started with secondary research, then ran a discovery workshop with PM, engineering, and data, structured around documenting what we thought we knew, what we were guessing, and where our model of the student had never been tested. From that, we wrote a problem statement to align the team:

Five knowledge-gap themes emerged from the workshop:

Research artifactResearch questions & objectives · FigJam

I led interviews with 12 prospective grad students, people who had recently started a program, or been accepted and were waiting to begin. We dug into how they searched, what they prioritized, and what made them choose one program over another. Recent enrollees were close enough to remember what actually drove their decisions, with enough distance to articulate it. Twelve hit saturation.

Four themes came through clearly:

Research artifactInterview synthesis · affinity map

The most surprising finding changed how we thought about the entire segment:

I led a user journey mapping session with the team to document where the current experience broke down, and where we had the most opportunity to close the gap between what students wanted and what we showed them. From that work, we aligned on a product vision:

Design artifactGrad searcher journey map

Success criteria we defined together:

I facilitated a cross-functional ideation session. Each team member sketched how they'd solve the problem; we converged on what was technically feasible and most aligned with student needs. The direction broke from our standard product: each graduate program would have its own presence on Niche, surfacing cost, format, schedule, specialization, and student outcomes at the program level rather than the school level. This mirrored how grad students actually searched, and changed the data relationship with partner schools, who'd need to provide richer, program-level data.

Ideation artifactConcept exploration · program-first mobile

High-fidelity prototypeSearch → school → program

Usability testing surfaced two issues: students couldn't find their way back from a program page to the school's main page, and the rankings were ambiguous about whether they reflected the program or the institution. Both were copy and labeling fixes, an explicit back-to-school navigation link, and context beneath rankings clarifying they reflected the institution as a whole.

Product · LiveProgram-level data on school pages

Schools became more motivated to update their program information through the partner portal once they saw what richer data was unlocking. The leading indicator was student engagement: search interactions on grad content jumped 800%+ in Q3 2022. The trailing outcome was business: graduate market share moved from 8% to 17%.

We doubled market share with what we shipped. The full vision, individual program profile pages, never made it. Leadership pulled the plug on continued investment shortly after launch, and resources transitioned back to undergrad.

What I'd do differently: invest earlier in the strategic case, not just the design work. By the time the architecture call landed, leadership was weighing infrastructure lift against an unknown ceiling for the segment. A clearer written picture of what grad could be, sized, modeled, presented to the room, might have changed the conversation. That was the lever I didn't pull, and the one I'd reach for first next time.

What stays with me is the methodology. Leading with research before solutioning is what made the work land. The data did the convincing on every cross-functional decision that mattered.

Fifth Third Bank wanted to build something with AI. That was the extent of the brief. No customer problem. No success metric. No scope. Just executive signal that this was the direction.

At a staff level, an underspecified brief isn't a gap to fill. It's the actual job. I was positioned for it through my work on the online account opening team, where I'd already seen the downstream cost of customers choosing the wrong products. That context gave me a starting hypothesis worth testing.

Data showed most customers weren't in products suited to their financial situation. The bank saw lower engagement and higher churn. Customers paid avoidable fees and didn't understand why. Neither side was winning. The easy path was to build what was asked for: an AI demo. I pushed for a different frame.

Before any design work started, I ran stakeholder interviews to understand what leadership actually needed versus what they'd asked for, eleven stakeholders across digital, product, and consumer banking. One theme dominated: the bank wanted to be a relationship bank, not just a transactional one. That was the insight that unlocked the product direction.

On the customer side, we ran a user interview and co-creation session with 12 participants representing Fifth Third's five core personas. What we heard was consistent: customers wanted financial advice that was tailored, proactive, and available when they needed it, not buried in an FAQ or routed to a call center.

Before any screens existed, I ran a two-hour design exploration with the team. The prompt wasn't "design a chatbot." It was: how would a trusted friend with financial expertise ask about your money? I used out-of-industry references to break fixedness around how a traditional bank presents itself.

That session produced the conversational design principles Jeanie was built on, turn-taking, compliment patterns, tone calibration, how to handle failure states without breaking trust. It also gave the team shared vocabulary, which mattered when we hit disagreements during build.

Usability confirmed what the product had assumed: when Jeanie could resolve the issue directly, participants preferred her over a wait queue for a human agent. Cross-functional alignment with compliance and legal held the product in scope and kept the trust framing intact through launch. Fees from mismatched products and conversion trended positively for customers who completed the guided flow, though those metrics weren't formally instrumented at this stage.

The harder challenge wasn't the design work, it was keeping the product pointed at the right problem long enough to ship it. There were multiple moments where the ask drifted back toward "make it more like open AI." Each time, I had to reorient the conversation toward what we'd actually set out to solve, and why that was still right.

The conversational design principles and component patterns we documented didn't disappear when Jeanie shipped. They became a reference for how Fifth Third approached messaging design after. That's usually when you know the framing held, when it keeps running without you.

This was also the first time I worked closely with an NLU backend. The gap between what a model can do and what a product should do is a design decision, not an engineering one.