Sourcing AI
Pietra
—
2025
I designed the AI powered sourcing experience at Pietra, helping emerging brands move from vague product ideas to supplier ready conversations, while improving response rates, reducing back and forth, and strengthening trust across the marketplace.
Project overview
01
Problem
Brands coming onto Pietra were attracted by its trusted network of 1300+ vetted factories, but many long tail users struggled to clearly articulate what they wanted to build. As a result, they sent vague sourcing requests that suppliers could not act on, leading to low response rates, excessive back and forth, and frustration on both sides of the marketplace.
Solution
I led the design of Pietra’s sourcing AI agent to help brands move from early product ideas to supplier ready conversations. The experience guided users in defining product requirements, generated structured product briefs, recommended relevant factories, and automated outcome driven supplier conversations, improving clarity, efficiency, and trust across the sourcing workflow.
Impact
The sourcing AI agent saw strong adoption, with 230+ AI generated product briefs created. This resulted in 140 sourcing invoices, with nearly 50% converting into paid transactions.
Tools
Team
Lauren Schiller, PM
Xinyu Wu, AI Engineer
Noelle Claessens, UI developer
As Head of Design and an individual contributor at Pietra, I led the end to end design of Pietra’s first AI powered product, the sourcing AI agent. I partnered closely with founders, product, engineering, and supply teams to define how AI could streamline sourcing workflows while preserving trust across the marketplace.
Problem
02
A growing number of long tail users were coming to Pietra with the ambition to start a brand, but without a clear understanding of what they wanted to build. These brands were attracted by access to trusted suppliers, but often lacked the experience to translate an idea into clear sourcing requirements.
Suppliers, on the other hand, needed structured product details to assess feasibility, pricing, and timelines. When that information was missing, conversations slowed down or stopped entirely.
Brands reached out to suppliers with vague product ideas, leading to stalled conversations, low response rates, and frustration on both sides of the marketplace.
Hi, I want to manufacture hoodies. Oversized fit, premium quality. What would it cost?
Thanks. Do you have a product brief with
Fabric type
Weight or GSM
Colors
Quantity
Target unit cost
Timeline
Not sure about GSM. Probably cotton or something soft. Quantity maybe 200 to start.
Is this men or women? Any reference images?
Both. I’m still figuring things out. Just want to see what’s possible.
Supplier has ended the conversation
Research
03
To unblock sourcing conversations, I partnered with the sourcing team and reviewed historical supplier conversations to understand where and why outreach was breaking down. I needed answers to two questions:
What information helped suppliers progress a chat toward a match?
Suppliers preferred a structured product brief (materials, construction details, quantities, pricing targets) to evaluate feasibility, costs, and timelines.
Why were brands unable to share these requirements?
Most long tail user were early stage and unfamiliar with sourcing. They did not know how to translate an idea into structured product requirements.
Insight
The gap was not intent or effort, but knowledge and structure. If brands are guided to create clear, detailed product briefs, sourcing conversations progress faster and supplier engagement increases.
Problem statement
04
How might we help brands move from early ideas to clear, supplier-ready requirements, so sourcing conversations can progress forward?
AI Design opportunity
05
As AI became increasingly capable of interpreting and structuring unstructured input, we saw an opportunity to use it as a guide rather than a shortcut.
t-shirts
oversize fit
soft
GSM?
cotton maybe
Product overview
Fabric
Rib-Knit Collar
Stitching Detail
220gsm, 100% Combed Ring-Spun Cotton Jersey
95% Cotton, 5% Spandex Blend
Double-Needle Stitching
Product overview
Product brief for cotton t-shirt
Product Name
Product Category
Target Market
Season
Intended Use
Oversized Fit T-Shirt
Apparel
Unisex
Year-round
Casual Wear
AI could
Help brands refine early product ideas
Generate product briefs
Connect brands with right factories
The goal was not automation, but clarity. AI was used to guide and educate brands through the sourcing process so they could approach suppliers with confidence and receive faster, more meaningful responses.
Design and iterations
06
I designed an AI-led sourcing experience that helped brands translate early ideas into structured product briefs and connect with the right factories to move sourcing conversations toward production.
Idea intake
—
1
Design to support uncertainty, not punish it.
The intake experience was designed to meet users where they were in their thinking, allowing them to share intent first and refine details progressively.
The input screen was designed to feel lightweight and approachable, allowing brands to describe what they wanted to build in plain language. The goal was to lower the barrier to getting started while setting users up for clearer, more actionable outcomes downstream.
30% of users who saw the input screen progressed to generating a product brief.
This was a strong signal given that sourcing is not a frequent, everyday task and many users were exploring the feature out of curiosity at launch.
Minimal input to avoid early intimidation
First-time brand owners often don’t know how to answer detailed sourcing questions upfront.
We kept the intake to a simple text input and trained the AI to infer missing details and prompt for them later, once users had more context. They could start with what they knew, increasing completion and reducing abandonment while still capturing intent.
Example inputs to reduce prompt uncertainty
AI experiences assume users know what to write, but most users don’t.
We provided example inputs at different levels of detail to show what kind of information works, without requiring users to learn how to prompt.
Improving input quality before generation
Because product brief generation relies heavily on context, vague inputs would lead to more guesswork and additional refinement later.
We added a “Help me refine my idea” step that generated 2-3 enhanced versions of the user’s input, which users could select and edit before moving forward. This lightweight step improved clarity early, leading to more reliable product briefs and less rework downstream.
Product brief
—
2
Product brief as the core artifact
Pietra sits at the center of a large, trusted supplier network and has deep institutional knowledge of what factories need to evaluate a project. Over time, this has resulted in thousands of data points across product categories, materials, pricing structures, MOQs, and production constraints.
The product brief system was designed to leverage this knowledge to translate unstructured brand input into a structured, supplier ready brief.
The product brief is the core artifact suppliers rely on to evaluate a project. Because it directly impacts response rates and sourcing outcomes, it was designed as the primary workspace, always visible, editable, and supported by AI guidance.
55% of users who created a product brief moved on to factory recs
This indicated that once users invested in creating a structured brief, the experience successfully carried momentum forward instead of dropping users between steps.
Guiding users through complexity
Moving from a simple idea to a detailed product brief is a big jump, especially for first time founders.
We used the AI agent to explicitly guide users through the brief section by section, explaining what was happening and what would be addressed next.
Locking fields to respect user intent
Once users make a decision, repeated questions feel inefficient and undermine trust in the system.
When users clearly defined a requirement, that field was automatically locked and skipped by the AI. Users could also manually lock fields at any point, ensuring the system respected final decisions and focused only on unresolved parts of the brief.
Visualizing decisions to reduce abstraction
Many decisions are hard to evaluate without seeing a physical outcome. Imagining how materials, finishes, or construction choices will look post-production can be challenging and often leads to hesitation or guesswork.
We introduced a visualization option that used the full context of the product brief to generate visual representations of different options, helping users reduce uncertainty and make more informed decisions.
Using conversation to drive completion
Static forms are hard to complete when users don’t yet know all the answers.
The AI guided users through the brief one field at a time, asking focused questions, explaining options when needed, and updating the document in real time.
Making AI assumptions visible and reviewable
When information is missing, AI systems often make silent assumptions. These hidden guesses can introduce errors and reduce user trust, especially in high-stakes workflows like sourcing.
We surfaced multiple options and marked the field as requiring review. Users had to explicitly confirm a choice before proceeding, keeping uncertainty visible and decisions intentional.
Supporting both guided and direct editing
While some needed step-by-step guidance, others wanted to move quickly without being slowed down by prompts or explanations.
The product brief supported direct, inline editing alongside AI guidance. Users could manually edit any field at any time, and the AI acknowledged and adapted to those changes in real time.
Factory outreach
—
3
Designing for confidence at the moment of commitment
Users have already invested significant effort in clarifying their idea and creating a detailed product brief. At this point, their primary concern is simple "Will this increase my chances of getting a response?"
The final step of the journey was designed to protect that effort by ensuring users reached out to the right factories, in the right way.
Factory selection is the highest drop-off point in the sourcing flow. Too many options, unclear differences, or low-confidence outreach can stall progress completely.
We leveraged AI to use the completed product brief to recommend a focused shortlist of factories that matched the user’s requirements across pricing, MOQs, product category, and production capabilities.
To reduce friction and decision making, the system also generated a supplier-ready introductory message, which users could review and edit before sending.
~80% of users who saw factory recs went on to contact suppliers
This validated that the final step removed hesitation at the point of outreach and translated preparation into action.
Automating conversations
—
4
Extending AI into messaging
As AI started helping brands identify more relevant factories, users were often reaching out to a larger number of suppliers than before. While this improved reach, it also increased the effort required to manage multiple parallel conversations.
To reduce this overhead, we extended AI support into supplier messaging.
Brands could choose to either manage conversations manually or deploy an AI agent to handle outreach and follow-ups on their behalf. When automation was enabled, users defined simple goals for the agent, such as what information to collect and when to report back.
This allowed users to scale supplier conversations while staying focused on decision-making, rather than coordination.
~90% of users who turned on automated messaging kept it enabled.
This indicated that goal-based automation reduced overhead without compromising trust or control.
Preventing failed automation through proactive error handling
Automated outreach fails when critical information is missing.
Before allowing users to deploy the agent, we introduced an LLM-based information validation layer. Errors were surfaced clearly, with specific guidance on what needed to be added. This ensured the agent entered conversations with enough context to be effective.
Allowing users to intervene, adjust, or take over at any point
Users needed confidence that automation wouldn’t lock them into decisions or conversations they could no longer control.
I designed explicit control affordances within the messaging experience. Users could edit the goals or stop the agent altogether. Automation became reversible, reducing hesitation and increasing trust.
Making agent work scannable
As conversations scaled, users needed a fast way to understand status without opening every thread.
AI summarized conversations and tagged them by outcome, helping users quickly identify which threads needed action versus which were still in progress.
Closing the loop with timely, actionable notifications
Users didn’t want to monitor every conversation, but didn’t want to miss decisions
We built a lightweight email notification system that alerted users only when human input was needed.
Solution
07
I designed the end-to-end AI-led sourcing experience that helped long-tail brands move from early, unstructured ideas to supplier-ready conversations, increasing response rates and accelerating sourcing decisions.
Idea intake
The input experience was intentionally lightweight and non-intimidating, allowing users to express intent in plain language. Instead of assuming expertise, the system guides users toward better inputs through examples and optional refinement, improving downstream outcomes.
AI led Product brief creation
The product brief was designed as the core interaction surface, with AI guiding decisions, surfacing uncertainty, and respecting user intent through explicit confirmations.
Deploying the agent
AI narrows the supplier set to the right factories and, when aksed, handles supplier outreach through goal-based automation that users can edit, or stop at any time.
Impact
08
In a span of 3 months,
230+ product briefs were created through the sourcing AI, generating over 140 invoices, with ~50% converting to paid orders.
Evolution
07
The sourcing AI has since been integrated into Pietra’s core co-pilot, enabling users to generate briefs and discover relevant factories through a single, unified experience.
Sourcing AI was Pietra’s first successful AI-native workflow, proving that guided automation could meaningfully improve a complex, high-stakes process like sourcing. This integration marked the transition from an experimental AI feature to a foundational capability embedded in the product’s day-to-day workflows.
Let's work together —
Currently looking for new opportunities