INTRODUCTION

AI Enquiries Insights

AI Enquiry Insights is an intelligent assistant that helps users increase the success rate of their enquiries. By analyzing historical data, the model identifies which factors drive supplier engagement, such as payment terms, delivery mode, Incoterms, or quantity. Based on these insights, the AI provides real-time recommendations while users fill out the enquiry form.

This was my first time acting as a Product Owner, leading a cross-functional team with data science, engineering, and design. The data team developed the predictive model, while I was responsible for shaping it into a usable product experience. The project served as a pilot for AI adoption at Metalshub, testing how advanced analytics could be integrated into our workflows and delivered as tangible value to users.

Overview

Brainstorming

Early ideation on AI use cases

As the Product Owner, I facilitated a workshop with design, engineering, and data teams to align on scope and explore how AI could support enquiry success. This session was key to transforming the data model into a practical product feature.

Mapping the enquiry journey

In the workshop, we mapped the entire enquiry journey step by step.By combining this with historical data, we identified where users struggle most.These pain points became the foundation for defining AI opportunities.

Feature 1

Enquiry Recommender

The goal of this feature was to increase the success rate of enquiries by making them more attractive to suppliers. Historical listing data was analyzed to identify patterns that led to higher supplier participation. For example, payment terms under 30 days or delivery via DDP Rotterdam consistently showed stronger engagement.

The model translated these insights into real-time recommendations during enquiry creation. As buyers filled out the form, the system suggested optimal parameters such as payment terms, delivery mode, or Incoterms. Importantly, users could accept or adjust these suggestions, maintaining full control.

Feature 2

Document Importer

The second feature aimed to reduce the manual effort of transferring contract or Excel data into the platform. Since building a full importer would require significant development resources, we decided to prototype the workflow with AI first and test whether it could actually improve user efficiency.

Challenge

Prototyping in uncharted territories…

Our users are more traditional, and during past testing sessions we noticed that many struggled to understand the concept of a Figma prototype — they assumed every button should be fully functional. This led to unexpected insights: instead of focusing on the flow, users reacted to the missing interactions.

To address this, we explored new tools such as V0 with Cursor to build higher-fidelity prototypes. Compared to Figma’s text-to-image mockups, V0 generated cleaner visuals and allowed us to publish directly or export code to GitHub. This approach made the prototypes behave closer to a real product, which in turn created more reliable user feedback and smoother interactions during testing.

Hypothesis workflow

AI outputs were unpredictable in testing, we split the workflow into stages — filtering and validating at each step — to improve overall accuracy and reliability.

Resilience

AI analysis times were unpredictable — sometimes very fast, sometimes significantly slower. To avoid blocking the workflow, we allowed users to run the importer in the background and continue with other tasks until results were ready.

During training we observed a high rate of AI hallucinations. While scoring systems could flag risky outputs, we also introduced quick-edit capabilities so users could easily correct errors. This combination of transparency and control helped build trust in the feature.