Envisioning a future where dashboards stand the test of Generative AI in Digital Analytics

Lately, I've been asking myself some questions about the future of data visualization in a world where generative AI is becoming increasingly important.

Generative AI in digital analytics

We now have many solutions of projects of solutions that integrate AI features to:

• Help model or explore data in warehouses
• Address analytical questions from datasets
• Annotate reports with generated insights
• Automate all the configuration needed for measurement across different tracking and collection tools
• Handle quality assurance aspects automatically
• Or even reveal insights from an ordinary chart in a dashboard

Dashboards

Ah, dashboards...some even say that dashboards will no longer be necessary, that it will be enough to just ask "the AI" questions and get the right answers.

For now,

• if there is one cognitively effective way to represent and convey a message, it's visualization, the right visualization
• if there is one thing that reassures humans and organizations, it's routine, stability, and proof, making it easier to spot what's remarkable

What's the synthesis of the two? ...dashboards and data visualizations.

Analytics transformations with generative AI, under conditions

And thinking that we only need to be alerted is to forget that our profession, our value, our vocation also consist of absorbing knowledge, context, and demanding that we sharpen our minds. Scan, understand, explore, find, confirm.

We want the journey AND the destination.

Some will say we already have plenty to work with... but I wonder if we might risk losing, without realizing it, what drives us, and sometimes what inspires us, elevates us, makes us creative.

My words are deliberately exaggerated, but it seems to me the risk is very real:

• Focusing only on productivity metrics while forgetting the rest
• Paying attention only to the noise, when silences are just as important
• Teaching only the machine while neglecting the importance of skill development and knowledge acquisition among our peers

Moreover, we also know that using generative AI necessarily requires maintaining a form of control, the ability to perform intermediate checks, and therefore by no means being content with a black box into which we feed requests and receive ready-made solutions in return. No magic, no reliable black box, no rabbit out of a hat. Quite the opposite, and we've been experiencing this for a little while now. We must follow the rabbit down the hole.

If there is one essential reflex when it comes to adopting AI within Analytics activities, it's to preserve our analytical mindset. It must come back with even greater force, be ever-present.

Data visualization and generative AI

Let’s return to data visualization, in an environment controlled by the digital analyst and conducive to knowledge acquisition and intellectual growth.

Respecting the following usual principles greatly simplifies the participation of AI:

• Segmentation: Break down a complex problem into smaller problems
• Specialization: Assign a specific function to each building block or technical layer
• Progressiveness: Refine the data step by step, layer by layer, to facilitate its control, quality assurance, and promotion as a source of truth
• Catalog / Documentation: Inject meaning, context, and business logic
• Automation: Promote programming to encourage automation

Critical bottleneck of historical BI and data viz tools

Looking ahead at data visualization, one cannot help but notice a critical bottleneck in many BI and data viz solutions: a WYSIWYG editor designed exclusively for humans, with no option for programming, and therefore no easy automation:

• An interface design not intended for programmatically creating or modifying customized reports or dashboards
• Limited possibilities afterward, restricted to enhancements through corollary features or overlays
• Complex components that can only be adapted with certain customizations, or that allow only simple actions such as duplication or replacing a data source that shares exactly the same model
• ...

For these tools, removing these limitations will be crucial in the medium term. They will have to move beyond the user who clicks and waits. The challenge is to be able to generate reports, or report prototypes, via AI agents orchestrated by more or less detailed, multimodal instructions.

And this will have to be done by opening their software to protocols such as MCP (Model Context Protocol) or by developing their own proprietary protocol for programmatic report generation and AI agents.

"BI as code", episode 2: On steroids with generative AI

In this context and at this stage, a family of BI tools seems to stand out, thanks in particular to the mindset in which they were designed a few years ago: BI as code.

All the benefits of industrialization and software code controls, applied to data display. Your reports are generated through configuration and layout files.

In short, a very favorable context for adding your AI agents to generate structured reports by following your instructions, leaving you only to fine-tune the AI deliverable so that it becomes a business deliverable.

As part of a lot of advantages, this would be a way to include AI in your data presentation layer without it, or its results, haphazardly replacing the interactive reports and dashboards we currently produce manually.

Another obvious benefit is democratizing code-based solutions for a wider audience. Since code is currently the entry point for this type of visualization generation tool, it can easily be enriched through a graphical editing interface, a generative AI chatbot, or an automation program...or even all three at once.

BI as code for digital analytics and digital marketing

On the other hand, certain current but surmountable constraints prevent or slow down the rapid adoption of "BI as code" or "data viz as code" in analytics and digital marketing in general. But many benefits encourage us to consider its use in our activities, including as example the availability of solutions in Open Source form with GPL licenses, and even MIT licenses for some.

Everything suggests that this trend, born in the BI world, will branch out into the digital analytics field. In light of this, the historical providers of data visualization and BI solutions face significant challenges to overcome the generational limitations “by design” in their tools. They’ve moved some of their pieces forward, but the road ahead is long.

Final thoughts

We will be here to implement the best strategies, with the most suitable solutions, for the benefit and sustainability of our projects.

In the meantime, the time for experimenting with available solutions is well underway, and we will certainly have the opportunity to discuss the changes, still somewhat tentative at this stage, in data visualization for digital analytics, but which are likely to bring about greater activities transformations in the future...just like the rest.

This article remains an appetizer regarding possible solutions for the best collaboration between generative AI and data visualizations for the sake of our deliverables. There is so much to say about the benefits and the operational means to ensure them.

On a personal note, I’ve started benchmarking and testing solutions and will probably share some examples in the future. So, see you soon for the next chapter of these adventures.