Tanu Tyagi Portfolio

Introduction

Overview

Outcomes

User Persona

Solution

Key Decisions

Reflection

case study · ENCORA Client engagement · pharma supply chain

Predicting the U.S. drug shortage crisis 30 days before it hits patients.

Built at Encora for a Tier-1 U.S. pharmaceutical supply-chain operator (NDA-anonymised). RxRadar is an AI-powered procurement intelligence platform that turns reactive crisis response into proactive supply-chain decision-making for hospital procurement directors, pharmacists, and distributor buyers.

Built at Encora for a Tier-1 U.S. pharmaceutical supply-chain operator (NDA-anonymised). RxRadar is an AI-powered procurement intelligence platform that turns reactive crisis response into proactive supply-chain decision-making for hospital procurement directors, pharmacists, and distributor buyers.

Reading Guide

Shipped

What this is: a client-facing POC showing how procurement teams can act 30 days earlier on drug shortages.
What you’ll learn: the bet (outcomes), the primary user, and the one-screen decision loop (Drug Detail).
For quick summary : read 02 (Outcomes) and 04 (Deep dive).

30-days

Forecast horizon, surfaced on every drug page

8

End-to end screen shipped to clickable prototype

Duration

10 Days

💰The cost of doing nothing

The U.S. Durg supply chain is failing in real time and the people closest to patients are the last to know.

📌 Takeaway: This isn’t a “data problem” — it’s a time-to-decision problem for hospital procurement.

In 2024, U.S. active drug shortages reached 323 — the highest level in 23 years. The impact concentrated in the drugs hospitals can least afford to lose: chemotherapy, anaesthesia, ICU, and emergency medications. The five numbers below are not statistics about inconvenience. They are statistics about cancer patients whose chemotherapy gets postponed, ER patients whose epinephrine isn’t on the cart, and hospitals whose ratings drop because they could not deliver scheduled care.

In 2024, U.S. active drug shortages reached 323 — the highest level in 23 years. The impact concentrated in the drugs hospitals can least afford to lose: chemotherapy, anaesthesia, ICU, and emergency medications. The five numbers below are not statistics about inconvenience. They are statistics about cancer patients whose chemotherapy gets postponed, ER patients whose epinephrine isn’t on the cart, and hospitals whose ratings drop because they could not deliver scheduled care.

💊

323

Active drug shortages in 2024 — highest in 23 years.

💊

Audit Design Assets

Active drug shortages in 2024 — highest in 23 years.

💊

323

Active drug shortages in 2024 — highest in 23 years.

💉

50%

Affect sterile injectables — chemo, anaesthesia, ICU.

💉

50%

Affect sterile injectables — chemo, anaesthesia, ICU.

🏥

95%

of U.S. hospitals report shortages.

🏥

95%

of U.S. hospitals report shortages.

🩸

50+

Life-saving drugs at critical levels nationwide.

🩸

50+

Life-saving drugs at critical levels nationwide.

📦

80%

of U.S. drug ingredients imported, mostly from China & India.

📦

80%

of U.S. drug ingredients imported, mostly from China & India.

“The Procurement Director carries the weight of every shortage — and learns about it from a phone tree of ‘Out of Stock’ responses, often after the shelf is already empty.”

— STAKEHOLDER MAPPING SYNTHESIS, RXRADAR DISCOVERY

🧱What we build

The outcome, first.

📌 Takeaway: We shipped a single surface that turns forecast → action into one continuous flow.

I’m leading with outcomes because the story of a procurement product is the story of what becomes possible. Five mechanisms, each tied to a behaviour change for the Procurement Director.

01

ML model trained on historical supply, sales velocity, regulatory events, and recall feeds — surfaced as a forecast chart, not buried in a report. v1 model target: >80% precision on a 30-day window for Tier-1 sterile injectables.

30-day shortage forecast on every drug page

02

Forecast → regulatory alerts → ranked alternates → live supplier quotes, all collapsed into the Drug Detail page.

One screen replaces five tabs

03

Substitutes ranked on potency, dosage equivalence, cost, and clinical compatibility — not just chemical class. Match score model targeted >0.75 weighted F1 against pharmacist-validated pairings.

9 / 10 match score for every alternate

04

Procurement Directors compare suppliers and request a quote without leaving the drug they are investigating.

FDA / EMA / WHO alerts as a primary surface

05

Recall, counterfeit and dosage-update alerts live in the same frame as the procurement decision — not a footnote.

In-context Quote action

TRANSLATED TO THE USER’S DAY

"A decision that used to start with “Why are we out of this?” — now starts with “We have 30 days. Here are three substitutes ranked, five suppliers ranked, the recall update logged. Let’s act.”

👥 The user we built around

Wide stakeholder map. One primary user.

📌 Takeaway: One primary user (Procurement Director) keeps the product coherent under a wide stakeholder map.

I mapped both sides of the supply chain — manufacturer-side roles (Revenue, Supply Chain Ops, Demand Planner, Pricing) and dispenser-side roles (Procurement Director, Pharmacy Ops, Buyer, Operations Analyst). One user carries the weight of every shortage: the Procurement Director. Designing for them — and serving everyone else through role-based settings — was the most consequential framing decision in the project.

I mapped both sides of the supply chain — manufacturer-side roles (Revenue, Supply Chain Ops, Demand Planner, Pricing) and dispenser-side roles (Procurement Director, Pharmacy Ops, Buyer, Operations Analyst). One user carries the weight of every shortage: the Procurement Director. Designing for them — and serving everyone else through role-based settings — was the most consequential framing decision in the project.

🪏 Deep Dive

The decision loop, on one screen.

📌 Takeaway: The Drug Detail page is the spine — it answers “will it shortage, what do we do, and from whom?”

Procurement decisions used to be five tabs of context-switching. RxRadar’s Drug Detail page collapses them into one continuous reading rhythm — walked through here with the worked example of Morphine Sulfate Injection. Top: framing. Middle: signals. Bottom: action.

Procurement decisions used to be five tabs of context-switching. RxRadar’s Drug Detail page collapses them into one continuous reading rhythm — walked through here with the worked example of Morphine Sulfate Injection. Top: framing. Middle: signals. Bottom: action.

01

A line chart projecting shortage probability over the next 30 days, anchored to a Last 30 days reference. The forecast is the hook.

Why now? — 30-day forecast

02

Orders 342 vs Supplies 312 — the user sees where the gap is concentrated, not just that one exists.

Where? — geographic heatmap

03

Live, time-stamped, sourced: Supply Disruption · Counterfeit Alert · Dosage Update. The difference between a defensible decision and a regulatory incident.

What’s changing? — regulatory alerts

04

Hydromorphone (Hikma) — $9–11 · 45 days · 9/10 match. Ranked on potency, dosage, cost, clinical compatibility — not just chemical class.

Substitute? — alternate drugs

05

Pfizer (Current, 4.5) · Hikma 4.3 · Baxter · Novartis · GSK. One-click Quote action, in context.

From whom? — supplier quote

"The procurement decision is one decision. Splitting it across screens forces the user to hold the picture in working memory — folding it into one screen turns it into a single sweep of the eye."

““The procurement decision is one decision. Splitting it across screens forces the user to hold the picture in working memory — folding it into one screen turns it into a single sweep of the eye."

📈 five decisions

The trade-offs I had to defend.

📌 Takeaway: The value is in trade-offs we can defend, not in feature count.

Each row surfaces one decision — the alternative I considered, what I chose, and what it cost. This is the artifact most quoted in design reviews because it shows defensible thinking, not just outcomes.

Each row surfaces one decision — the alternative I considered, what I chose, and what it cost. This is the artifact most quoted in design reviews because it shows defensible thinking, not just outcomes.

01

Predict, don’t just report.

02

Procurement Director as primary, everyone else through roles.

03

Forecast → Alternate → Quote, on one surface.

04

Trust as a UI primitive, not a tooltip.

05

Start with the screen that does the most work.

🪞 WHAT’S NEXT & REFLECTION

What’s next — and what I’d do differently.

📌 Takeaway: The next phase is measurement + pilot validation, not more UI.

TRANSLATED TO THE USER’S DAY

“v1 build for client leadership review. Next: a two-week shadow study with a Tier-1 hospital procurement team — instrumenting forecast click-through, alternate-acceptance rate, and time-from-alert-to-quote-request against a baseline week of phone-tree procurement.”

What I would do differently.

Validate the 9/10 match score with a clinical-pharmacist sample early. The score is a strong UI affordance — the model behind it deserves clinical review at higher sample size.
Test density on 13” laptops, not just full-screen. The Drug Detail page is information-rich on 1440px+ — a hospital corridor laptop view needs explicit responsive treatment.
Bring engineering into the alert taxonomy earlier. Naming Supply Disruption vs Recall vs Dosage Update consistently across model + UI + audit log was the most expensive coordination cost.
Push the v1 measurement plan into engineering hand-off. Forecast click-through, alternate-drug acceptance rate, and time-from-alert-to-quote-request are already drafted as v1 success metrics; the next step is wiring them into the client’s analytics pipeline before the shadow study.tt

Validate the match score earlier with pharmacists. The score is a strong UI affordance — the model behind it deserves clinical review at higher sample size.
Test density on 13” laptops, not just full-screen. The Drug Detail page is information-rich on 1440px+ — a hospital corridor laptop view needs explicit responsive treatment.
Bring engineering into the alert taxonomy earlier. Naming Supply Disruption vs Recall vs Dosage Update consistently across model + UI + audit log was the most expensive coordination cost.
Push the v1 measurement plan into engineering hand-off. Forecast click-through, alternate-drug acceptance rate, and time-from-alert-to-quote-request are already drafted as v1 success metrics; the next step is wiring them into the client’s analytics pipeline before the shadow study.tt

What this project shaped in me

“How to anchor a wide stakeholder map in a single, defensible primary user.”

“How to make trust a first-class component, not a copy decision.”

“How to fold a multi-step decision into one screen without losing the rigor of each step.”

“How to design inside a system where a missing UI affordance has a real clinical consequence.”