AI Chatbot

Designing an Agent-Driven AI Assistant for Data Compliance

Overview

Aiimigo is an AI-powered chatbot assistant embedded within Workplace AI — a secure enterprise search platform that unifies data silos and surfaces insight with explainability and performance. Aiimigo’s purpose is to transform discovery into actionable insight by answering user queries, summarising documents, analysing relationships, and supporting complex tasks like Subject Access Requests (SARs).

Where traditional search returns links or keywords, Aiimigo is designed as an intelligent interface that reasons over data, interacts conversationally, and helps users get meaningful answers without manual context-switching.

Product: Workplace AI (Enterprise SaaS Search & AI)
Role: UX Research & UX/UI Design Lead (Research, Low & Hi-Fidelity Design)
Team: Head of UX, CTO, Engineering
Duration: 10+ Sprints

Strategic Problem

Workplace AI customers rely on search to unlock enterprise information scattered across systems. However:

Traditional search places the cognitive burden on users to interpret results manually.
Complex legal and governance tasks — such as SARs (Subject Access Requests) — involve multi-step processes, regulatory risk, and domain-specific subtleties that standard interfaces struggle to support.

Our challenge wasn’t just to make a chatbot — it was to design an AI interface that feels intuitive, trustworthy, and meaningfully useful in real workflows. This meant reconciling agentic AI behaviour (i.e., an assistant that acts rather than merely responds), with enterprise expectations around transparency, explainability, and compliance.

The key product question became:

How might we design a conversational AI that supports complex tasks while embedding trust and control into the user experience?

Research & Discovery

Before jumping into interface decisions, we grounded our design in industry and domain research.

Heuristic Evaluation

I audited existing AI assistant experiences — focusing on how agents handle:

Task completeness vs. ambiguity
Response accuracy and trust cues
Integration with underlying search and system data

This revealed a pattern: agents often excel in open-ended scenarios, but enterprise tasks like SARs demand structure, traceability, and clear intent signalling — not just conversational flair.

User Journey Mapping

To understand where Aiimigo could add value, we mapped the full SAR workflow:

Starting with an initial request
Through data collection and classification
To redaction and reporting

We evaluated where agentic AI could accelerate tasks, and where it could dangerously misinterpret intent. This led us to focus the initial phase on well-bounded, high-impact sub-journeys of SAR rather having to explore than the entire end-to-end process.

By thinking in terms of user journeys rather than screens, we established a focused phase one scope with clear success criteria.

Design Exploration

Planning for AI-Driven Interaction

We began by identifying key interaction elements:

Triggering agentic behaviour: What user actions prompt Aiimigo to meaningfully intervene?
Context awareness: How does Aiimigo surface relevant metadata and where does it draw from?
Transparency & control: How can the assistant explain its outputs and let users correct or refine them?

Before any UI was sketched, these questions shaped design principles — focusing on clarity, control, and accountability.

Wireframes & Structure

Wireframes explored how the AI assistant would exist alongside search results. We tested options such as:

A dockable chat panel vs. full-screen views
Contextual vs. standalone agent invocation
Inline prompts that suggest Aiimigo’s help based on user tasks

This period clarified the information architecture, ensuring the agent didn’t feel isolated from users’ existing workflows.

Usability Testing

We ran multiple prototype tests with Aiimi customers and internal users. The focus was on:

Whether the agent’s responses felt actionable
How users understood the agent’s confidence and source citations
Navigation between chat and underlying data

Feedback was overwhelmingly positive on component placement and familiarity, with specific refinements made around navigation clarity.

Final Design & Delivery

Prototyping

Once patterns were validated, I crafted a high-fidelity prototype in Figma, extending our design system to support:

Message bubbles, typing indicators, and conversational cues
Context panels that show why Aiimigo recommended something
Action prompts tied directly to search and classification results

The interface preserved a clean, minimal aesthetic — letting content and intent take centre stage while communicating AI confidence and evidence. This reinforced user trust, especially in regulatory tasks where explainability is non-negotiable.

Implementation & Delivery

We delivered the first phase of Aiimigo’s agentic SAR support. Post-release, we monitored adoption and bug feedback, iterating quickly to improve accuracy and reduce friction.

Key aspects of delivery included:

Alignment with engineering on backend data retrieval and response generation
Continuous adjustments based on customer usage patterns
Quick reaction to interface issues post-rollout

Impact & Outcomes

Business Impact

Aiimigo significantly enhanced Workplace AI’s competitive positioning by:

Making AI search proactive rather than reactive
Enabling customers to handle complex governance tasks more efficiently

The feature became a strategic differentiator in marketing and sales conversations — demonstrating how agentic AI can extend beyond simple Q&A to support regulated workflows.

User Impact

Enterprise users reported:

Reduced manual workload on compliance tasks
Lower operational risk through structured, guided processes
Faster turnaround on SAR-related work with clear signals and actionable suggestions

These benefits translated into higher engagement and satisfaction scores among early adopters.

Reflection & Lessons Learned

Design Before Development Matters

Aiimigo’s initial build began without UX input, adding friction as we retroactively shaped interaction logic. Starting with design exploration would have halved ambiguity and iteration cycles.

Balancing Agency with Control

Agents need autonomy — but enterprise users need visibility into reasoning and control over actions. Designing transparency mechanisms was as important as designing answers.

Chat as a UX Pattern

Chat interfaces are familiar and intuitive, but they require structured decision pathways when applied to mission-critical tasks (e.g., legal workflows) rather than casual Q&A.

Cross-Discipline Collaboration

Working closely with the CTO and engineering ensured our designs were technically feasible and scaled gracefully with AI capabilities that continue to evolve.

Next Steps

To mature Aiimigo further, future work could include:

Adaptive AI prompts that adjust based on user task history
Fine-grained task handoffs tied into workflow systems (notifications & assignments)
Performance insights that show how Aiimigo’s suggestions improve efficiency over time

Continue Exploring...

Classification: AI Studio

SaaS Product Feature Design

The one-stop-shop for configuring and managing document classifications in Aiimi's Workplace AI.

Notifications: Assignments

SaaS Product Feature Design

Improving the way users manage their assignments through an intuitive notifications experience.