FEXILLON

Rolling out AI
for Datacentres

Directing strategy for £1M+ AI rollout, scaling to 5,000+ users and improving workflow efficiency 300–500%.

SCROLL FOR MORE

Microsoft DC Operations teams rely on fast, accurate access to technical documentation (drawings, procedures, manuals, handover artefacts) across multiple live datacentre projects. Fexillon’s document estate is deep and (historically) time-consuming to navigate.

To reduce friction, we rolled out Fexillon AI Copilot to a nominated Microsoft focus group and ran a structured research programme.

Objectives

Validate whether Copilot genuinely improves findability and day-to-day decision-making.
Understand where trust breaks (accuracy, context, reliability, “AI confidence” vs reality).
Turn real-world feedback into prioritised roadmap recommendations.

Introduction

Role: Head of UX Research & Design
Scope: Entreprise SaaS, AI Copilot rollout
Users: Microsoft Datacentre Operatives
Scale: 100+ participants to 1500+ users

Challenges

Specialist Knowledge

Operational queries often depend on interrupting busy colleagues for answers

Learning Curves

Onboarding new users has long and steep learning curves, requiring dedicated resource

Slow Searching

Traditional search methods are slow, and frustrating due to intricate structures

Complex System

Supports complex ecosystems of many very specialised individuals with broad, far-reaching goals

Big Data

Hosts over 150,000 files and documents across 16 unique datacentre projects

Approach

We needed to understand if what we made solved real problems.

Everything we do for users is an assumption until it’s validated by feedback. Users were able to search through thousands of files and aggregate data through prompts, but we wanted to establish where this new technology sat within the existing paradigm. This would not only ensure that our solution worked, but also build a foundation for future features.

I designed a four-workshop programme with self-directed usage in between, supported by integrated feedback and targeted interviews.

Cadence and activities

Workshop 1: Onboarding & feature understanding

Demo the three Copilot surfaces and set expectations for feedback.

Workshop 2: Copilot in context and general feedback

Capture how tasks are done today, biggest AI frustrations, and most/least helpful aspects.

Workshop 3: Ideal experience and AI interactions

Map ideal workflows, tone/trust requirements, and where Copilot should sit in a “perfect day”.

Workshop 4: Use cases and key challenges

Prioritise the most valuable AI opportunities and customer-specific pain points.

Findings

All integrated feedback, group workshops, and participant interviews were analysed into insights.

User journey map

Behavioural findings

Learning through iterating

Users would repeatedly refine prompts to improve results, forming a common taxonomy to influence their responses.

Trial-and-error exploration

Uses treated Copilot as a sandbox, experimenting and pushing the limits of what the AI could do for them.

Chat-first mentality

Most users gravitated to chat sessions, even for general searches, preferring natural language over search inputs.

Summaries work best

Documents often contain more information than a user needed. Copilot’s summarised bullet points removed the noise.

Repeated use cases

Copilot was most used for locating specific information, learning systems, and understanding procedures.

Roles influenced usage

Different roles demanded different things. Field staff wanted quick look-ups, apprentices supported training, supervisors drafted reports.

Practical impact

Recommendations

Research findings uncovered painpoints about our first implementation that were crucial to future developments of the AI Copilot product. We gathered 40 insights across 12 recommendations over 5 categories.

Affinity mapping

Software performance

Ensuring reliable, fast, and seamless access to Copilot
features

Adoption and awareness

Driving feature discovery, onboarding, and long-term engagement

Workflow and structure

Supporting how users work across projects, roles, and organisational layers

Interaction and communication

Improving how users communicate with Copilot and interpret responses

Information retrieval

Helping users quickly find, understand, and trust critical content

Final thoughts

AI Copilot is integral for efficient DC Ops in Fexillon, but they have unique needs.

Whether it’s optimising our model for specific datatypes, allowing users to control and filter the data that Copilot accesses, or supporting field operatives with mission critical information, we have identified opportunities to expand out scope.

Extending our capabilities further into Mobile, Extended Reality, and Digital Twin offerings will unlock unprecedented capabilities for the datacentre environment.

Building on our body of research, deepening our relationship with our users, is crucial to achieving this vision.