Using AI to express the ‘So what’ of data
Designing an AI-powered tool to bring 20 years of consultancy experience onto analytic dashboards.
Product: AI B2B SaaS | Role: End-to-end design process
Overview
How can we help users move past raw survey data and to strategic guidance instantly? This project aimed to turn 20 years of consultancy frameworks and data into an AI-powered tool that provides actionable insights within our dashboards. I led the end-to-end design process, working closely with analytics, consultants, and early users.
Approach & Methods
Firstly, before I could design the interface, I immersed myself in the backend capabilities of our AI systems. I took an AWS course on AI and machine learning to understand the logic, tools, and feasibility behind what our data science team was building. This technical grounding helped me design with intent, not assumption.
To ensure alignment, I led biweekly working sessions with product, analytics, and consulting teams. These cross-functional reviews were key to maintaining the integrity of our consultancy voice—this couldn’t be just another AI plugin. The tool had to reflect our frameworks, benchmarking and uphold the trust clients place in us. We were especially mindful that any automated insight must support, not jeopardise, client relationships
I also interviewed users to uncover how they currently worked with survey data: what they understood easily, where they hit friction, and how they were already experimenting with AI regarding the survey data we provided them. These conversations highlighted major pain points in existing dashboards and shaped key interface decisions.
As our insights team trained the model by reinforcement learning, I gained the clarity I needed to begin designing the interface that would be available on the dashboard
Internal tool for reinforcement learning:
Results & Impact
I designed an interface that provides users with instant summaries of their data upon landing on the dashboard, along with an intuitive section for exploring deeper insights through guided prompts and easy data filtering. We launched a beta version to a select group of early adopters, enabling them to ask questions about their data and receive rich, contextualised insights. These insights draw on the company’s 20 years of expertise and are delivered in our consultants’ tone of voice, enhancing clarity, trust, and engagement.
‘Getting these kinds of insights without waiting for our quarterly relationship review with consultants is a game changer.’ — Early Beta User
High-level AI powered summary displayed on the dashboard landing page.
A dedicated page to gain AI-powered insights tailored to specific questions about selected accounts and markets.
Learnings & Next Steps
This was my first time designing for an AI-powered system, and I didn’t pretend to have all the answers. At the start, the boundaries of my role were blurry, which actually mirrored how many of us on the team felt. But through ongoing collaboration, learning, and a shared curiosity, we built something that felt both ambitious and grounded.
Understanding Limitations:
Understanding the AI Agent’s limitations and cost structure was essential in shaping the interface design. As a designer, the earlier you can access the system and grasp what feeds into the machine learning process, the better your design decisions will be.
Outcomes:
We're continuing to iterate with feedback from early adopters and plan to expand access soon. I'm excited to keep designing at the edge of emerging tech, especially projects that translate complexity into clarity.