Meridian Institute Analytics

Meridian is small, research-heavy, and split almost evenly between undergrad and grad (about 4,200 and 4,160 in Fall 2024). That split matters more than it sounds: any "total" quietly hides two different stories. Four domains, four owners, four dashboards that never spoke to each other.

One question, four files

Leadership thinks across domains. The tools didn't. A provost asks "is enrollment softening, is research strong enough to carry us, is headcount outrunning revenue?" in a single breath, and the old setup needed three separate sittings to answer it.

Every dashboard spoke its own dialect, too. The graduate school is international-heavy while undergrad is overwhelmingly domestic. Blend them and the number means nothing. So I split by level, by default, everywhere.

Two ways of reading, one product

People read this in two modes. Some glance: ninety seconds, give me the headline and anything on fire. Others drill: pull a pool apart by gender, country, test policy. The legacy tools dumped it all on screen at once, so the glance crowd burned their ninety seconds just figuring out where to look.

Who I designed for

I designed around six readers, from a provost who needs the headline in ninety seconds to an IR analyst who has to defend every number to the state. My bet: win the analysts who own the data, and everyone downstream inherits that trust.

Four Power BI reports. Same maroon, but no shared layout, no shared filters, no shared chart language. Nobody ever said the data was wrong. They said it took too long to find what they came for, about 90 seconds of orienting every single visit.

So I audited all four against Nielsen's heuristics on a 0 to 4 severity scale, and tied every finding to a real person and a real task.

Four reports, four mental models

Undergraduate Admissions, Graduate Admissions, Research, HR. Four domains, four mental models, four visual languages. A provost who just wanted a Monday-morning read had to open four files and stitch the story together in their head.

What was actually broken

Everything showed at once, and nothing told your eye where to start.

The biggest miss was treating the whole institution as one population. Enrollment is near 50/50 (about 4,200 undergrad, 4,160 grad). "Percent international" is around 48% for grad but only 4% for undergrad, so a blended number is a lie with a decimal point. I segmented by level everywhere.

Why it hit this school harder

The story leaders most needed to watch was a tense one: applications fell from roughly 14,170 to 10,670, the admit rate climbed (43% to 51%), yield ticked up (17% to 21%). The old layout could not tell that three-variable story. The numbers lived on different tiles, in different styles, with no shared timeline. Some people had quietly stopped looking. The whole redesign was about ending the orienting and leading with the answer.

Five rules did the deciding. If a screen broke one, it went back to grayscale.

What we were really fixing

The Institute had a reading problem, not a data problem. The numbers were right; they were scattered across four dashboards. Deans burned the first few minutes of every meeting hunting for the page, the filter, the current version.

The brief was not "prettier charts." It was: make these numbers fast to read and safe to trust, for people who open this between meetings and need an answer in under a minute.

What success looked like

I made success behavioral, not decorative. A first-time user names any screen's headline in about five seconds. A returning user answers a routine question with one filter change at most. One definition per metric. A forecast always reads as a forecast. And the analysts sign off that every number on screen means what their records say. That sign-off was the real acceptance test.

How I worked

Six passes, looping back often. The discipline I held myself to: agree with what Power BI can show before drawing anything pretty.

I audited the four legacy dashboards, mapped every idea to native behavior (drill-through, no custom workarounds), then collapsed them into one spine, grayscale first.

AI came last, on purpose. Ask Meridian AI is optional, audited for bias, human in the loop. A guide, not the decision-maker.

Enterprise work, so the box was real before I drew anything. These shaped every call.

One product, one navigation model, one set of filters that behave the same everywhere. Shallow on purpose: an Overview cockpit up top, then four owner-aligned areas, each opening on a summary before you drill into named sub-views. Filters carry over, so the drill-down continues the question you were already asking.

Two calls shaped it. Undergrad and Grad stay split (not one "Admissions" toggle) because the funnels are genuinely different: international mixes worlds apart, volumes 2x apart. One layout would always look wrong for one of them.

And one rule runs through the whole thing: segment by level, always. No lone "total enrollment," no blended "percent international." Both would describe a student who does not exist.

Four old dashboards, four layouts, one shade of maroon. I rebuilt them as one component set with one set of rules. Learn one page and you can read them all.

The vibe is calm, modern SaaS. Quiet chrome, so the numbers get the contrast. Foundations and the chart kit sit underneath; Ask Meridian AI is the one floating affordance, on top, never baked in.

These two are the heart of the platform, and they both break the same rule: one number lies. Grad students are about 48% international, undergrads about 4%. A single Institute-wide "percent international" describes nobody.

So the rule came first: segment by level, everywhere. Each gets its own overview, geography, and funnel. Shared chrome, never shared data.

Undergraduate admissions

The undergrad story is a tension, not a headline. Over two years the funnel both tightened and loosened at once.

• Applications fell ~14,170 to ~10,670.
• Admit rate rose ~43% to ~51%.
• Yield ticked up ~17% to ~21%.

Fewer applications, easier admit, slightly better yield. Every KPI carries a year-over-year delta, because "10,670 applications" tells a dean nothing and "down ~25%" starts a conversation. That single-cycle blindness is exactly what the legacy dashboards got wrong.

Graduate admissions

Graduate runs on a different engine: smaller pool, dramatic international skew, heavy loss between admitted and enrolled. So I built it around stage conversion, not multi-year trends.

The funnel makes the real problem legible: more than two thirds of admitted applicants decline. That gap, not the application count, is where the yield work lives.

With source markets concentrated (India by far the largest), that concentration is a risk as much as a strength. When one country supplies a big share of a class, a visa-policy change becomes an enrollment event.

Why I let the two diverge

One "admissions" template would have been easier. I chose not to. Undergrad is built for multi-year tension; grad for stage conversion and source-market concentration. The chrome, navigation, and the Ask AI affordance are shared so it still feels like one tool. The analysis splits because the two populations genuinely are different. That is the one call the legacy dashboards never made.

Undergrad and grad are one pipeline. Research and HR are different animals. Research is a money-and-output story told over years; HR is a workforce story where the key numbers are also the most sensitive. Both legacy dashboards had the same flaws as admissions (maroon chrome, pie charts, tiny type), but the fix was different because the questions were different.

I kept the overview-and-drill-down spine, then changed the top-line KPIs, the time horizon, and how I handled demographic data.

Research Intelligence

This office does not ask "how many proposals this month." It asks "is the funding pipeline healthier than it was three years ago." So I led with money, over time.

The KPI band pulls apart three numbers people kept confusing: funding (new) ~$21.4M (did we win it), awards (active) ~$52M (what we manage), expenditures ~$35.5M (are we spending it). The legacy version split these across separate pies, so people compared slices and drew the wrong conclusion. Putting them side by side, one definition each, was the single highest-value fix.

The decade trend anchors the page: one year tells you almost nothing; the ten-year slope tells you growing, flat, or sliding. Output (publications, patents, H-index ~184) sits off to the side as the lagging proxy it is.

HR Workforce

HR needed the most restraint: do less, more clearly, and treat people's data with care.

The KPI band is deliberately plain: headcount ~1,363, split right away into faculty ~708 and staff ~655, two populations run by different rules. New hires and turnover sit next to it so headcount reads as a flow, not a still.

The body answers the three questions an HR leader actually asks: who we have now, how the workforce is changing, and who we have by demographics, faculty versus staff.

The demographic panels are where I designed most carefully:

• Race/ethnicity and gender are composition, not performance. Nothing ranks or targets.
• Small headcounts are suppressed or grouped. "100% of 3 people" is noise and a privacy risk.
• No AI scoring here. Ask Meridian AI summarizes; it never predicts, ranks, or flags.

Why these two share a pattern

Both are longitudinal stories the legacy dashboards flattened into single-year pies. Both have a headline number that misleads on its own and only helps once you split it. Admissions taught me to lead with the funnel; these taught me to lead with the multi-year arc, and to know when not to compute at all.

The funnel is the whole job for enrollment leaders, so it sits right up top.

Five stages, plus a sixth most dashboards skip: summer melt, where yield quietly leaks out.

Why funnel and yield share a screen

One is meaningless without the other. Year one: applications fell 25%, but admit rate rose 43% to 51% and yield 17% to 21%. Narrower at the top, more efficient at the bottom.

The melt band

Real money that rarely gets a home. Showing deposited-versus-enrolled as its own delta gives leaders a number they can actually plan against.

Geography is the second lens. "Percent international" means nothing until you split it by level, so each level gets its own map.

Undergraduate Geo

At 4% international, the domestic story is the story. I lead with a US map, because the enrollment cliff is regional and Meridian sits right in the Northeast.

Graduate Geo

International-heavy as it is, the source-market world map leads. The pipeline is concentrated, and that is the risk: one visa shift can move a whole class.

The fastest proof: the old pain on the left, the new answer on the right.

Tab by tab

Undergraduate Admissions. Applications fell 14,170 to 10,670, admit rate climbed 43% to 51%. A Northeast school feels the demographic decline harder, so geography earned its own screen.

Graduate Admissions. The worst of the pie problem. International share now splits by level instead of pretending one number covers both.

Research. A dense grid of fiscal-year tables became a proposals-to-awards pipeline with flagged trends.

HR. Headcount 1,363, 124 new hires. Glance up top, detail below.

What carried across all four

One platform, not four prettier reports. The same patterns everywhere: KPIs with deltas, trends by default, charts chosen for the question, one predictable drill path.

No formal usability lab on this one, and I won't pretend otherwise. Here is what I actually leaned on, and the outcomes the design is built to produce.

What came of it, what I would redo, and where Meridian goes next.