Harsh DesaiDesign and research, forever wondering
← Galaxy

Product Design · 2020–Present · Current role

GHX

Helping hospitals decide which medical devices to buy, and whether it paid off.

GHX’s value analysis products help hospital teams make supply decisions with evidence: whether a new device should come in, which products in a category to standardize on, why some cases cost more than others, and whether a contract’s rebates were ever paid. I’m a senior product designer there. This page plays a few flows, redrawn on our current design system with sample data; the rest of the work stays private.

Role
Senior Product Designer
Year
2020–Present
Domain
Hospital supply chain
Status
Current role

Problem

Which devices a hospital buys is decided by committees of clinicians and supply chain staff, with evidence, physician preference, spend and contract terms spread across different places.

Approach

Put the evidence, the spend and the people involved in front of whoever is deciding, at the moment they decide.

Result

GHX customers report results such as Riverside Health’s 28% cut in joint costs in one three-month project, and $520,000 saved in a year at ECU Health.

01 · Context

Value analysis, for people who aren’t analysts.

Value analysis is how a hospital decides which medical devices and supplies to use. A new product is requested, often by a physician, reviewed by a committee of clinicians and supply chain staff, and approved or declined. Afterwards the same teams look for variation in cost and use, and check that contracts deliver what they promised.

I design across those products, and across the wider GHX platform: an AI orchestration platform, the Registration Center and the Vendormate kiosk. Most of the value-analysis flows below are redrawn on the Lumere Design System we use today, with a fictional health system and its people; the device makers are real. Category optimization is shown as designed.

Product introduction
New product requests, with research content beside the decision.
Category optimization
Comparable products side by side, with spend and evidence.
Physician-level analytics
Where cost varies by procedure and physician, and why.
Rebate intelligence
Tiers, earnings and payments, tracked against each contract.

02 · Design: Product introduction

The request is where the decision happens.

A new product request carries its own context: the group it joins, what it replaces and what research says.

New product request, step 1: PMX02 typed into the product field, with PriMatrix Dermal Repair Scaffold from Integra LifeSciences suggested. Beside the form, how the committee reviews requests and two similar requests elsewhere in the system, one approved and one declined.
A requester starts from a name or a catalog number, and the form finds the product.

A request starts with one product and one requester, but a committee decides it. So the request gathers what they need: the product group it belongs to, the products it would replace and what those cost today, any financial interest the requester has declared, and who still has to sign off.

Research content sits inside the form instead of in a separate library. If the evidence shows little difference in outcomes across a group, the requester sees that while filling it in, and the reviewers see it before they vote.

03 · Design: Crosswalks

What would we buy instead, and what would it save?

A crosswalk maps every line a hospital buys to one supplier’s equivalent, then prices the switch.

The Crosswalks tab of a spine implant standardization project: four crosswalks, each with who fills it, its status and how many of the currently purchased lines have an equivalent.
Each crosswalk maps what the hospital buys today to one supplier’s equivalents, and the supplier’s rep can fill it in.

Part of the same value analysis tool. When a project considers moving a category to another supplier, supply chain creates a crosswalk from what the hospital buys today, and the supplier’s rep fills in their equivalent for each line, with a price. Both sides can talk it through on the crosswalk itself.

The hospital sees coverage, how confident each match is and what each line would save, then generates a scenario at the conversion rate it believes, beside the project’s other options.

04 · Design: Catalog decisions

One decision per catalog, not per product.

Decisions are recorded catalog by catalog, at the facilities they apply to, each with its reason.

Record decisions for a pacemaker project: three products and seven catalogs, each approved, denied or left undecided at a number of the 14 facilities, with a reason; two products show a mixed decision across their catalogs.
A product can be approved at some facilities and denied at others, catalog by catalog, each with its reason.

A project ends in decisions, and a product is rarely one decision. One size or configuration is approved everywhere, another only at the facilities with the volume, a third left for the physicians to weigh in. So decisions are recorded at the catalog level, across the project’s facilities, with a reason and a clinical conclusion, and many catalogs can change at once.

For a catalog the project moves away from, the committee also approves substitutes, ranked, and they go into the substitute library for when an item is backordered. Those decisions are what reports and measurement read after the project is archived.

05 · Design: Physician-level analytics

Why some cases cost more than others.

Every case for a procedure, sorted by cost, then what the expensive ones have in common.

Total knee arthroplasty at Northbrook Health: 3,101 cases by 64 physicians, $5,492 median cost per case. A field of cases from low cost to high with percentile benchmarks, and below it five product groups by how often each turns up in low- and high-cost cases.
Every case for the procedure on one line, from low cost to high, with benchmarks that can be dragged.

Cost per case for a procedure varies, sometimes widely. The analysis puts every case on one line from low cost to high, then compares the two ends: which product groups turn up more often in the high-cost cases, which products inside a group, and which physicians.

GHX describes the point of this as supporting discussions with clinical teams. The design follows that: it shows patterns a supply chain team can bring to physicians, with the cases behind every number.

06 · Design: Rebate intelligence

Money a contract promised, and whether it arrived.

What each contract has earned, what was paid, and which tiers are still within reach.

The Rebates dashboard for Northbrook Health: $6.32M calculated across 5 tracked contracts, $2.35M received, $3.97M outstanding and 10 payments to log; a Start here card to log 3 payments for Edwards TAVR, with quick actions; calculated against received by quarter; tracked contracts, each with its signals and a next step; a rebate calendar; strategies; and tier opportunities sorted by what can still be reached this period.
The dashboard starts with what to do next: payments to log, money below what was calculated, and tiers within reach.

Device contracts often pay rebates in tiers, by volume, spend or market share. Rebate intelligence tracks those commitments, compares what a hospital should have earned with what was paid, and finds rebate dollars that were missed.

The dashboard starts from what someone can act on today: the payments still to log, the money that came in short of what was calculated, and the tiers still within reach this period. From there, a contract breaks down by rebate and period, and each rebate shows the tier it is on, what the next one pays, and the purchases that count toward it.

07 · Design: Category optimization

Comparable products, side by side.

A product group compared on spend, contracts, attributes and what the evidence says about each product.

By product group

The implantable cardioverter defibrillator group: 52 products from 6 brands, criteria, evidence insights and research links, then the products purchased across 14 facilities ($5,162,757 in all), filterable by contract, pacing, type, MRI compatibility and heart failure monitoring, each with its contract, key differentiating attributes, eaches, most recent price with benchmark percentile, and total spend.
A product group lists what the system buys, with spend, contract, benchmarks and the attributes that tell products apart.

By vendor

Vendor insights for Northbrook Health, partnered with 82 vendors: $41.6M of spend, 67 vendors on contract, $6.1M off contract; a sourcing calendar of upcoming renewals with the spend at stake; and every vendor with its contracts, total and off-contract spend, share of spend and a price strength percentile, expandable by category and product group.
Every vendor a hospital buys from, with its spend off contract and how its prices compare.

Category optimization looks at a whole group of comparable devices at once, using the hospital’s own purchasing data. Each product shows its contract, the attributes that tell it apart, and what the system paid against a benchmark. Picking a few opens them side by side over the list, so it is easier to see where a system pays more for a product the evidence rates the same.

The same question can start from a vendor instead of a product group. Vendor insights lists every vendor a hospital buys from, with its off-contract spend and how its prices compare, and each vendor opens on what the hospital buys from it, priced strategies (consolidate with that vendor, or renegotiate what is above market) and how well other vendors cover the same products.

08 · Design: AI orchestration

Agents that work the orders, in the open.

AI agents work the orders between hospitals and suppliers, and hand people only what needs a decision.

Provider · Backorders

Substitution Management: $8.1M of backordered spend and 14K backordered lines; cards for 14 urgent items flagged on inventory and clinical priority, 52 items found with other vendors, 40 equivalent products matched on specifications and safety class, active evaluations, compliance checks and contract checks; and the six AI agents working on them, with their confidence and what each is doing.
Backordered items, ranked by clinical priority, with the agents already looking for substitutes.

Provider · Price exceptions

A provider’s dashboard on the GHX Orchestration Platform: automation savings, expiring contracts, rebates, content and compliance; questions to ask ResiliencyAI; and cards for outstanding price exceptions, utilization, master data, vendor performance and expiring items, each with a recommended action and the agents’ performance on it, such as 47 POs flagged, a 61% auto-resolved rate and 1,842 exceptions closed.
Every card pairs a recommended action with what the agents already did on it: flagged, auto-resolved, closed.

Supplier · Shortages

Recommendation Management for a supplier: $8.1M of backordered revenue and a 68% rate of items with recommendations; cards for 52 critical requests, items at risk of backorder with an equivalent match, expiring contracts and new pricing requests; and the agents working on them.
A supplier sees which customers are waiting on backordered items, and how many already have a substitute.

Orders between hospitals and suppliers go wrong in small, expensive ways: an item goes on backorder, a price doesn’t match the contract, a customer is left waiting. GHX describes its orchestration platform as helping supply chain teams sense, decide and act faster, across providers and suppliers.

The three flows here are three of those jobs. A hospital finds a substitute for a backordered item, and sees how the agent found it. A hospital approves a price correction the agent could not settle alone. A supplier recommends substitutes to the customers waiting on a shortage, and sends them a bulletin the AI drafted.

The design work is about trust. Each agent shows its confidence and what it is doing now, and can show its work. What an agent resolved on its own is kept apart from what a person approved, and nothing goes to the other side of the trade until a person sends it.

09 · Design: Registration Center

Connecting trading partners, without the ticket.

Every request shows who has it, what’s next and when it escalates, on both sides of the trade.

Provider · Approve a request

My work for a provider’s registration admin: 12 items need them, 38 are waiting on suppliers, 9 go-lives are scheduled and the median time to enable is 6.2 days. Each item has a reason, an age and one next action; suppliers that have waited show when they escalate; exceptions, scheduled go-lives and 41 requests handled automatically this week sit beside them.
Home starts from what needs this person, and how long everything waiting on the other side has waited.

Provider · Change in bulk

Hierarchy and locations for Kaiser Permanente: the regions and facilities as a tree, each location with its type, account count and flags for possible duplicates, a new warehouse and Tukwila Warehouse closing on March 1, and an offer to register the new warehouse with all 37 suppliers in one change set.
Locations live in one hierarchy that flags likely duplicates and a warehouse about to close.

Supplier · Respond

My work for Medline’s registration team: 214 requests need approval and 86 already match its customer master, ready to approve together; the ones that need a look, with a mismatch, a missing bill-to or a question; scheduled go-lives, customers it is waiting on, and the changes GHX made this week.
A supplier’s home starts from the requests that already match its customer master, ready to approve together.

GHX support · Admin mode

The GHX queue in admin mode, under a banner saying every change is logged as by GHX: 408 steps GHX owns across all organizations, with tabs for unassigned, mine and needs C+, bulk Assign to me and Complete as GHX, and a reason on every row.
GHX’s teams get a queue of only the steps GHX owns, across every organization, and every change is logged under their name.

Before a hospital can trade with a supplier through GHX, the two set up and connect their accounts in the Registration Center. Much of that is managed by hand: trading partners are added and kept current one at a time, and when a request stalls, it becomes a support ticket that GHX’s own teams chase between the two sides. GHX calls this kind of manual coordination workflow debt. Research with customers and GHX’s teams found the Registration Center works for one task at a time, and breaks down at scale and at the handoff, where each side waits on a step the other controls.

The rebuild is drawn as user flows for the three roles that share it, each starting from its own home. A hospital’s registration admin approves what a supplier changed, registers accounts, and moves locations in change sets that preview every account they touch, apply on a date and can be undone. A supplier’s team approves in bulk the requests that already match its customer master, and sends back what does not as a diff for the hospital to accept. GHX’s support teams work a queue of only the steps GHX owns, in an admin mode where every change is logged under their name and GHX never approves for a partner. Around all three, routine requests are handled by rules, feeds and matching against master data, and every automated decision is logged. People step in for what is unusual, not for what is routine.

10 · Design: Vendormate kiosk

Checking in at the front desk, without the front desk.

Vendor reps check themselves in at a lobby kiosk, from scanning in to a printed badge.

The kiosk home screen at a health system’s medical center: Welcome visitor, with two large choices, a personal visit or a business visit, and a language switch.
One kiosk serves every visitor; a supplier rep starts with Business Visit.

Vendormate is how hospitals credential the supplier reps who come on site. The kiosk is where a visit starts, for personal visitors and supplier reps alike: a rep picks Business Visit, scans in or checks in by hand, says which department and why when no visit is planned, answers the wellness questions, and leaves with a printed badge.

It stands in a lobby and is used by strangers, standing up, often in a hurry. So the screens are large and plain, there is a language switch, and Start over is always in reach.

11 · Result

What the platform has done for hospitals.

I can’t share internal metrics. These are the numbers GHX publishes about its value analysis offering and what customers have done with it. They are the product’s results, built by many people, not mine alone.

Riverside Health
28% reduction in joint costs, through a single three-month project.
ECU Health
$520,000 saved in a year.
Evidence library
48K+ products, 2,800+ groups of comparable products and 350+ curated evidence insights.