The Future of Data Experience: From Static Dashboards to Conversational AI
We recently sat down to discuss the current state and future of Data Experience (DX). What followed was an insightful discussion about our history with DX, the journey we’ve taken at X is Y, and how we plan to shape the future.
It all started with an initial development process heavily based on product development methodologies, keeping designs simple and presenting data through the right charts and visuals. But as we looked back, we realised the true value wasn't just in the charts; it was in the journey we took with our users.
THE EVOLUTION: BRINGING USERS ON THE JOURNEY
Early on, we found that bringing users along for the ride was critical. It gave them "skin in the game." By seeing each other's designs and voting on them, they didn't just engage with the way of working, they engaged with each other.
This collaboration helped users understand why specific designs were chosen, helping them navigate away from tables towards visuals and dashboards. The major benefit was consensus: accountability and responsibility shifted from the developers back to the business.
While this worked well, we knew we could do better.
FORMALISING CONSISTENCY: THE DESIGN SYSTEM
From this foundation, our first Design System was created in Figma. The goal was to formalise and maintain consistency in elements like colors, typography, spacing, and content hierarchy, while incorporating reading techniques like the F-shape or Z-shape.
Through multiple iterations, the design system has become more constrained to drive velocity. We focus on essential components such as line charts, bar charts, and comparators, while still allowing for necessary custom tiles.
This system addresses two key problems:
User Experience: Providing an easy-to-understand, consistent interface.
Velocity: Reducing the decisions developers need to make, thereby speeding up delivery.
We also found that rapid prototyping helps users become emotionally attached to the product they co-created. Showing a prototype helps clarify needs sooner, getting to the heart of what is actually required.
THE HIERARCHY OF DASHBOARDS
We implemented a hierarchy of dashboards from high-level metrics in Strategic dashboards to dynamic detailed drill-downs as the agreed-upon method for data exploration.
This allows users to quickly identify key areas or problems and take that information to the teams responsible or chat with analysts for a deeper dive. Our view is that dashboards (and tools like Tableau or Power BI) are not analytical investigation tools for business users or so-called "self-service." They only go so far.
Example of Metric Hierarchy:
Start with Net Profit.
Drill down into Revenue and Costs.
Explore the specific areas that make those up.
Crucially, dashboards are intended to guide users to a point where they can make decisions or talk to the right teams not to answer every single question.
WHERE WE ARE NOW: A WELL-OILED MACHINE
This journey has led us to today. Data experience at X is Y is a well-oiled machine. We have tools and processes in place that streamline work, reduce complexity, and increase velocity. The unknown has become the known.
So, what next?
THE NEXT FRONTIER: ARTIFICIAL INTELLIGENCE
The current hype is AI, and we believe it. For Data Experience (DX), it can make our job easier and allow us to go even faster. There is massive potential, but we’ve quickly recognised that you can’t just jump in and go crazy.
D&A teams already carry a lot of technical debt, and AI risks increasing this. Implementing it must be done with purpose. It’s impossible to build a car without first inventing the wheel, and users naturally resist change. AI is a big step away from their beloved Excel file with pie charts.
Short Term: Acceleration & Elevation
Our short-term goal is to utilise AI to accelerate our current methodology. We aren't replacing the dashboard yet; we are drastically reducing the time it takes to build one.
The workflow shifts: We can take a design prototype from a workshop and hand it directly to an AI agent. We ask the agent to "build this in PowerBI," and it figures out the model and constructs the visuals.
This doesn't just cut down on hours; it elevates the DX team. Instead of spending days on the technical minutiae of building visuals, our team is "moved up a layer." They can focus on the strategic definitions, the user experience, and ensuring the output aligns with business goals. We work alongside our Data Engineering team who are using AI for modeling and data ingestion to tighten the loop between design and deployment.
Medium Term: The Hybrid Experience
As we look to the medium term, we begin to change the experience itself. However, it would be foolish to simply "cut over" business users entirely to a chatbot.
Conversation is hard for most humans to parse rapidly. There is a reason websites use GUIs and not just text—visuals help humans get to a decision quicker. Therefore, the medium-term future is a hybrid of Visual + Conversational.
Dashboards will still exist to provide an immediate status update, but they only need to go to a certain point. Instead of building endless drill-down pages, the user will pivot to AI to ask the deeper questions. The AI becomes the analyst sitting next to them, explaining the "why" behind the chart.
The Critical Need for Guardrails
Before we reach the long term, we must address governance. Human brains are essentially "random thought generators." If we turn an AI loose on operational data without constraints, it becomes an exponential mess of bias and distraction.
We must put the same guardrails on our agents that we used to put on our drill-down dashboards. The AI shouldn't just answer any question; it should guide the user back to the company strategy. If a user asks a distracting question, the AI should be sculpted to say, "That’s an interesting idea, but we are currently focused on [Strategic Goal X]. Here is the information related to that."
This prevents the "scatter-gunning" of ideas and keeps the business aligned.
There is also a misconception that AI can magically interpret messy data; in reality, AI acts as an accelerant. It follows the same rule as any human analyst: bad data in, bad information out. The difference is that while a human might pause to question an anomaly, an unconstrained AI might confidently present a hallucination as fact.
To prevent this "exponential mess," we must provide the AI with strict, approved definitions and clean data pipelines. We need to create guardrails that force the AI to align with company goals, ensuring it focuses on valid business truths rather than statistical noise.
Long Term: Orchestration & Action
The long-term vision shifts from "Information" to "Action."
Currently, data teams deliver information, but the onus is still on the business user to figure out what to do with them and take the risk of being wrong. The long-term goal is Agent-to-Agent communication, where the human is taken out of the processing loop and placed into an orchestration role.
Imagine an Operational Agent talking to a Metrics Agent to analyse a problem. The outcome isn't just a chart; it is a series of "bets." The AI will say:
We have identified a problem in X.
Here are 3 potential actions we can take based on historical scenarios.
Action A has a projected outcome of Y.
Which action would you like to take?
The human remains in the loop, but as a decision-maker (an orchestrator) rather than a data-miner. This allows the AI to handle 80-90% of the routine queries, freeing up human analysts to handle the complex, creative, or edge-case strategic problems that require human nuance.
We’ll go further into the technology, principles, best practices, and the power of design in our following blogs on this DX future series.
