The brief was to design a tool.
I proposed starting with current state reality.

When I joined Volvo Group Trucks Technology & Industrial in August 2024, the Purchasing division managed supplier collaboration across a fragmented landscape of disconnected tools: no single source of truth, no unified experience, and no shared understanding of what the end-to-end journey actually looked like. One of the core tensions: designing for common processes while accounting for the specific exceptions each of the 14 Volvo Group brands operated by.

The dynamic will be familiar to anyone who has worked on marketplace products: when the platform fails, one side improvises and the other compensates, and the cost lands somewhere nobody is measuring.

What we found was not a technology problem. Suppliers were improvising to fill gaps the tools couldn't handle. Buyers were compensating, absorbing friction invisibly. The cost was real but scattered: clarifications, re-routing, double-checking, late visibility. Individually small. Strategically expensive.

Nothing about this project
was straightforward.

17 months from first interview
to portfolio-wide design direction.

Both sides of the collaboration.
All the layers connecting them.

Over 17 months I led a mixed-methods programme spanning perspectives that had never been brought into the same conversation: buyers, suppliers, and the organizational layers sitting between them.

The first service blueprint
produced in 15 years for this area.

Before we could design improvements, we needed to agree on what the current reality actually looked like. The result was a complete end-to-end supplier collaboration blueprint, covering every stage, every role, and every handoff across the journey.

This blueprint became the basis for prioritizing which pain points to address first. It directly shaped the scope of the Supplier Collaboration Portal (Case Study 02) and the supplier registration process, and it provided the foundation for fitting a micro-journey for Raw Material Agreements (covered in Case Study 03).

Service blueprint · end-to-end supplier journey

Jobs to be done · supplier main job map

Proto-personas · four supplier archetypes

Research that keeps working
after the interviews end.

To scale insights beyond a one-time deliverable, I built a dedicated Copilot research agent trained on anonymized interview transcripts. Designing the interaction patterns for how the team queries findings was as important as the technical setup. The agent needed to surface the right level of detail without overwhelming, and to prompt follow-up questions that led to better product decisions. Rather than insights living in a static report, the team can query findings continuously, connecting new product questions back to original research in real time.

Built with Microsoft Copilot Studio · anonymized transcripts · daily use by the product team

Six changes that outlast
the research itself.

The data existed.
The ownership didn't.

The Supplier Collaboration Platform is a new product built for the Purchasing portfolio at Volvo Group — replacing a fragmented legacy system where certificate management and contact data lived in disconnected tools, managed largely through email.

This case study covers the first two use cases taken into production: certificate management and supplier contact management. Both address the same underlying problem — critical supplier data that nobody owned, nobody monitored, and nobody could trust. What made this work consequential wasn't the features themselves. It was the approach: research before design, users before assumptions, and a prototype that felt real enough to test properly.

No data. No roles.
No baseline to start from.

Two use cases.
One shared failure mode.

Research first.
Prototype only when ready to test.

A full research-to-prototype cycle across both use cases, run in parallel — each informing the shared platform architecture underneath.

Built to feel real.
Tested like production.

The prototype was built using GitHub Copilot and Claude in VS Code — both used as active collaborators in the development process, not just autocomplete tools. The goal was a React + .NET application close enough to production that users could respond to it as they would the actual product.

This approach compressed the design-to-tested-interaction cycle significantly — and produced a prototype credible enough to shift stakeholder expectations of what a prototype needs to be.

Not a clickable mockup.
A working environment.

The prototype was deployed to a test environment, so sessions ran against something that behaved like the product rather than an approximation of it. Users could search, open, and act on real structures — which is what made their responses worth listening to.

Three levels of proof.
One baseline survey in progress.

Metrics were proposed as part of the design deliverables — structured to demonstrate value at adoption, behaviour change, and outcome level. Target values are being established through an end-user survey.

Five shifts that hold
beyond the deliverables.

Two surfaces.
One supplier relationship.

Supply chain compliance is not a stable problem. EU deforestation regulations, shifting tariff frameworks, raw material volatility — the regulatory ground moves faster than most enterprise software can follow. The Volvo Purchasing portfolio needed two things at roughly the same time: a brand new tool for collecting supplier compliance declarations, and a redesign of an existing raw material pricing application that had drifted out of alignment with the design system and accumulated usability debt along the way.

Running them in parallel was a deliberate call: if suppliers would eventually touch both, the logic holding each app together needed to match. That shared context also meant findings from one fed directly into the other. The AI evaluation approach developed for the compliance app became the starting point for how we evaluated the pricing app — not because we were short on time, but because the problem space genuinely overlapped.

Two apps,
two kinds of difficulty.

What we did —
and what it changed.

A full research-to-prototype cycle across both tools — each informing the other, with AI-assisted prototyping establishing a reusable method along the way.

The price update cycle,
end to end.

Mapping the raw material price update across both sides — supplier and buyer — showed where the work leaves the application: manual verification in Excel, parts excluded and tracked by hand, discrepancies resolved over email. The tool was not broken so much as partial.

RM price update journey · supplier and buyer

Not just using Copilot —
establishing how to use it.

The vibe-coded prototype approach covered more than tooling. We established a specific set of instructions to keep generated code in a state developers could work from rather than rewrite. The rules mattered as much as the technology.

Key principles: prioritise VCDK components before inline alternatives; write in small modular functions; mock all UI data and keep REST API as live data source.

Same data.
A different tool to read it with.

These pairs show the raw material pricing app — the tool the team had stopped seeing the problems in. The redesign was proposed in two stages: a like-for-like component update first, structural change second. Left, the existing application. Right, the redesigned equivalent rebuilt on the Volvo design system.

Before

Dashboard · existing application

After

Dashboard · redesign

Before

Review · existing application

After

Review · redesign, part detail expanded

What the work
actually revealed.

What moved —
and what stayed moved.