AI Agents Marketplace
The study showed that too many parallel discovery routes, plus an open AI input field, made users hesitate when they did not arrive with a specific goal.
From open-ended discovery to clearer guidance
- Starting point
Initial approach
An AI search field and several routes for discovering agents.
- Evidence / constraint
Observed difficulty
Participants without a clear goal hesitated at the blank input and competing choices.
- Decision
Revised direction
Clearer categories by business need, with guidance through configuration and next steps.
| Role | Product Designer. I owned the UX direction, the synthesis of external research, the functional prototype, and the evaluative study. |
|---|---|
| Context | A new Vodafone Business marketplace where SMEs could discover, evaluate, configure, and install AI agents from third-party providers in one place. |
| Users | Small and medium businesses, focused on first-time buyers: limited technical background, curious about AI and cautious about it. |
| Timeframe | 2025, from joining the project to a functional prototype tested with 15 users. |
| What I owned | Research synthesis into a user knowledge profile, tasks, jobs to be done, and journeys; UX direction; the Figma Make functional prototype; the evaluative study and its readout. |
| Status | Unmoderated study with 15 participants in the UK, Germany, and Switzerland. Findings informed the business case and ongoing vendor discussions. No development implementation and no production launch. |
The challenge
Vodafone Business saw an opportunity in a fragmented market: small and medium businesses needed AI help but rarely had the time or the technical depth to evaluate, buy, and set up agents from a growing field of providers. The idea was a marketplace where an SME could find, trust, and install AI agents in one place, with Vodafone's name behind it.
I needed to define how a first-time buyer would understand the offer, find a relevant agent and work through configuration. The experience had to provide enough guidance for people with limited technical knowledge.
I also presented this project as my master's thesis in Customer Experience and Innovation.
My role
An external agency ran the foundational qualitative and quantitative research. My job was to turn it into something the design could be built on, then to own the design and the evaluation:
- Synthesised the agency research into a user knowledge profile, task list, jobs to be done, and customer journeys, so every design decision had a reference point.
- Owned the UX direction for discovery, evaluation, configuration, and purchase, then revised it in response to the study.
- Built a functional prototype in Figma Make, deliberately focused on flow and interaction rather than finished visuals.
- Designed and ran an unmoderated evaluative study with 15 SME-profile participants across the UK, Germany, and Switzerland, and turned the sessions into findings for stakeholders.
What we learned
The study showed that too many parallel discovery routes, plus an open AI input field, made users hesitate when they did not arrive with a specific goal. The findings supported clearer categories related to the areas of the business users had in mind.
It also showed that configuration needed guidance about what to set up and why, and that the post-purchase moment needed clear expectations of what happens next.
These findings drove a revision of the prototype toward simpler discovery and stronger guidance. The revision was informed by the research but was not retested, and no development team implemented it. The project did not reach production.
Where it stands
The findings informed the business case and ongoing vendor discussions. I recommended another evaluation of the revised experience; that study has not taken place.
I introduced AI-assisted functional prototyping within my immediate team. This supported the existing user-centred design process: we could invest in user tasks, diagrams and stakeholder alignment before moving into screens, then turn the agreed structure into a testable experience quickly. The team has begun applying elements of this approach on other projects.
Reflection
The lesson I keep returning to is watching evidence override my own instinct. The AI search box felt like the obvious hero of an AI marketplace until the study revealed hesitation among participants without a clear goal. Front-loaded definition, the profile and the tasks, is what made the fast loop possible; the AI tooling only accelerated a direction that was already clear.
Methodology grounding: User-Centred Design (ISO 9241-210); Jobs To Be Done; unmoderated usability evaluation.