The Making of AMS AI: Designing Medical AI Procurement

Hospital procurement teams were still managing critical supplies through spreadsheets, emails, and manual follow-ups. Stock shortages were often discovered too late. Vendor comparisons required hours of manual work. Decisions were reactive rather than informed — and every delay had tangible operational consequences.

The AMS AI team brought strong domain expertise and a conviction that AI could meaningfully improve this process. What they needed was clarity: where AI should intervene, and how it could simplify — rather than complicate — everyday work.

Building the flow before the interface

Early on, we deliberately avoided jumping straight into screens. First, we built.

For AI-native products, flows cannot be designed in Figma alone — they need to be implemented, tested, and experienced through real interaction. We started by building early AI-driven flows to observe how procurement behaves in motion: what inputs teams actually provide, how the system responds, and where assumptions break once real data is involved.

Alongside this, we mapped procurement as it happens inside hospitals, from monitoring inventory to sourcing products, requesting quotes, and comparing offers. This helped us identify where teams lose the most time and where mistakes are most likely to occur.

Three priorities became clear:

• 𐩒Reducing reactive work caused by the late discovery of shortages
• 𐩒Reducing manual effort across sourcing, RFQs, and comparisons
• 𐩒Making hidden inefficiencies visible — surfacing signals teams rarely have time to notice, but that quietly drive cost, risk, and delay

These priorities shaped every decision that followed — and directly informed which product capabilities were designed first.

Making shortages visible before they become critical

One of the most impactful optimisations focused on early visibility. Rather than designing dashboards that merely report problems after the fact, we centred AMS AI on shortage alerts up to 14 days in advance. This feature was designed to shift procurement from constant firefighting to planned action, giving teams time to explore alternatives, engage vendors, and make decisions without urgency dictating quality.

From a design perspective, alerts were treated not as passive notifications, but as decision triggers: clear, contextual signals that prompt action while preserving user control. The result is fewer urgent purchases, fewer last-minute compromises, and more predictable procurement flows.

Turning vendor chaos into structured decisions

Vendor communication proved to be one of the largest sources of inefficiency. RFQs and quotes arrive in different formats, through different channels, and require extensive manual comparison. We streamlined this process by structuring RFQs and normalising incoming vendor responses, allowing teams to assess offers based on what actually matters — availability, fit, and cost — instead of wrestling with documents.

Designing this experience meant focusing on decision-making rather than documentation. By removing repetitive follow-ups and manual comparison work, procurement teams reclaim 20+ hours per week per person — time previously lost to administrative tasks.

Using benchmarks to challenge habitual decisions

Procurement decisions are often driven by habit rather than context. Teams repeat the same purchasing patterns simply because there is no reference point to question them.

We introduced benchmarking against peer hospitals, giving procurement managers visibility into how their usage and spend compare to similar organisations. This helps teams identify inefficiencies, over-ordering, or missed opportunities — without prescribing what decisions they should make.

From a product perspective, benchmarking was framed as guidance rather than judgement, supporting better decisions without adding friction or internal tension.

Turning procurement data into actionable analytics

Spreadsheets, reports, and fragmented systems make it difficult to understand where time and money are actually being lost. Analytics tend to describe what already happened, rather than support better decisions going forward.

Our team approached analytics as a decision-support layer, not a reporting tool. By combining usage data, peer benchmarking, and early shortage signals, the platform now helps procurement teams spot inefficiencies before they escalate — from over-ordering to slow-moving stock and recurring manual bottlenecks.

Instead of asking teams to analyse data themselves, we designed analytics to surface where attention is needed and why, at the moment decisions are made. This shifts analytics from passive observation to practical guidance — supporting faster procurement cycles, up to 30% supply savings, and a significant reduction in manual work across teams.

Designing for measurable impact, not abstract AI value

Across the product, AI is treated as an enabler rather than the centrepiece. Every AI-assisted capability was evaluated through a practical lens: does this reduce time, cost, or risk? If the answer was unclear, the feature did not move forward. This principle helped keep the product focused on operational impact rather than technical novelty.

That focus is reflected in AMS AI's publicly shared results: 75% faster procurement, up to 30% supply savings, and significant reductions in manual work.

These outcomes are not abstract promises — they are the cumulative result of deliberate design choices aimed at simplifying a complex system.

Results and key takeaways

By the end of the project, we delivered more than a set of interfaces. We helped shape a coherent product narrative that translated AI capabilities into concrete operational value and could be clearly communicated to investors, hospital teams, and vendors.

From a product perspective, our work resulted in a structured, AI-assisted procurement flow built around early visibility, reduced manual effort, and clearer decision points. For procurement teams, this meant faster sourcing and fewer last-minute shortages; for vendors, a more predictable and structured way to respond to hospital demand.

From a business standpoint, this clarity translated into tangible, publicly shared outcomes:

• 𐩒75% faster procurement processes
• 𐩒Up to 30% supply savings
• 𐩒20+ hours per week saved per person by removing manual work across sourcing, RFQs, and analysis
• 𐩒A complete set of go-to-market assets, including the website, product videos, and a pitch deck
• 𐩒Support for fundraising and sales conversations, contributing to a successful $1.2M funding round in 2025

For us, the key takeaway is simple. Designing AI products in complex domains like healthcare is not about showcasing intelligence, but about placing it precisely where it removes friction, risk, and uncertainty. When done right, AI stops feeling like a feature and starts working as infrastructure.