Artificial Intelligence: revolution or illusion in Digital Analytics?
Over the past two years, AI has aroused a mixture of fascination and skepticism in the world of Digital in general and Digital Analytics in particular.
A revolution, say some. An illusion, say others.
So, what's really going on? Let's explore this glass half-full... or half-empty.
AI: a powerful tool, but not a miracle solution
Ever since it burst onto the digital scene, AI has been supposed to be an essential marketing tool.
Promises of total automation, infallible predictions and unprecedented efficiency dominate the discourse.
The promise is big, it is strong, it is relentless. And so is the expectation.
But what's the reality when you take a closer look?
The reality is that many advertisers and agencies are struggling to see these promises materialize. And why is that?
- Current models lack business integration.
AI doesn't always understand the operational subtleties of companies.
Without this integration, it produces results that are difficult to use. - Data quality is insufficient.
Without reliable data to learn from, AI becomes ineffective.
It produces approximate results and reproduces the same biases and shortcomings as the processes it was supposed to overcome.
And yet, in marketing, precision is essential and leaves no room for approximation. Disappointment is therefore commensurate with the gulf between promise and reality, for both end-customers and agencies; from a business, financial and ROI point of view.
These pitfalls take us back to a time when marketing decisions were based on fragile assumptions, or even intuition.
Ironically, AI, which is supposed to symbolize progress, could actually set back certain practices.
Rethinking the uses of AI: from promises to concrete solutions
Let's be clear: AI in its current state is not a failure. Almost everyone is still in the learning phase. The potential of AI remains immense, but it depends on one crucial factor: data quality. With data that is qualitative, i.e. accurate, consistent and resilient, AI can truly become a transformative tool.
Take, for example, the wonderful world of Digital Analytics, with its array of cookies, tags, MarTechs and other tracking technologies designed to collect data from the Web and Apps in order tooptimize UX, personalization, conversion and produce actionable Insights.
Every day, these technologies collect mountains of data.
Yet the reliability of this data remains a challenge.
Rather than focusing on the sustainability of tracking processes, many analytics vendors have chosen to integrate AI into their technologies.
But why not focus on improving processes first?
Because integrating AI is easy, not too expensive, and trendy?
Because AI is a fantastic tool that makes life easier for editors... even if it does nothing to solve the historical and ongoing problems of tracking resilience.
To date, the use of AI in Digital Analytics is still not a proven, demonstrated, measured factor in improving data quality as we defined it earlier.
Adding AI to a failing model is like building on an unstable foundation. It won't hold.
Nor will TMS solve process and data quality problems.
In 20 years, little or nothing has changed.
Just take a look at your Chrome console a few hours to a few days after a production launch (depending on the nature of the site) to see for yourself...
Data On Duty's 2025 guidelines for pragmatic, operational use of AI
At Data On Duty, we believe AI can revolutionize Digital Analytics, but not by skipping steps.
If AI is to have a significant impact on digital analytics, it will inevitably require the ability to understand businesses and processes, as well as relevant and resilient tracking, for both in-site and media, and any other technological model likely to contribute to data collection and Insights production.
Insights: the Holy Grail of digital analysts, business owners, decision-makers and many software publishers. Our experience has often shown that technology alone, however innovative and powerful, is incapable of producing truly exploitable Insight.
We believe that it is the modeling of relevant and accurate data, derived from multiple tracking technologies over multiple paths at a given moment in time, that will produce relevant and usable Insights.
Data On Duty already offers solutions for managing tagging processes, centralized tracking and data repositories, and for solving data collection problems at source.
Of course, this introduces an unprecedented level of complexity into pre-conversion and pre-acquisition data processing.
But what a challenge!
And what an extraordinary field of application for AI!
All the more so with the introduction of the temporal dimension.
Let's be pragmatic and realistic.
As we have seen, for the time being, the contribution of AI to Digital Analytics is hampered by the lack of integration of business models and the poor quality of learning data.
This reduces AI to a facilitator's role, without guaranteeing reliable and sustainable results.
Our vision of AI use is therefore based on improving existing technology, thanks to finite and controllable models.
The choice of AI model to address a subject will be decisive. Generative AI alone cannot do everything.
Depending on the type of questions to be addressed, other models will need to be added, notablydescriptive, predictive or prescriptive AI. The latter is particularly critical in understanding business issues and processes.
It is a prerequisite for exploiting data sets that are as qualitative and targeted as possible.
These will then be processed by generative, descriptive or predictive models, depending on the objectives to be achieved.
To illustrate, in an e-commerce environment, a descriptive model might identify that 30% of cart abandonments occur at the checkout stage. A predictive model could determine which customers are most likely to abandon, and a prescriptive model would suggest solutions such as adding payment options or promotional codes.
At Data On Duty, we believe that a pragmatic and structural approach to the uses of AI in Digital Analytics requires :
- Processes shared by the various contributors on the same business platform
- A centralized tracking and data repository
- Appropriate tracking and monitoring
- Relevant, accurate and compliant data
- This is what Data On Duty - Governance Manager already does.
Then we'll need a self-correcting data collection system that guarantees resilient data, which is what Data On Duty - Enforcement Manager will be doing in spring 2025.
By achieving this, we will have propelled our industry towards unrivalled excellence in tracking while paving the way for the production of relevant, actionable Insights thanks to AI.
We will then be able to focus on the key stages that are priorities for the vast majority of the major groups we work with:
- Produce relevant and exploitable Insights: by introducing the dimensions of path and temporality. AI will enable the combined analysis of relevant tracking sources and valid generated data to produce these Insights.
- Model recurring success patterns with a systemic approach in applying AI models to the results produced by our DoD platform to:
a) anticipate the appearance of anomalies in data collection and ensure their correction
b) understand and model conversion success patterns and introduce them into a reproducible success model
c) understand and reverse conversion failure patterns and introduce them into a reproducible success model.
The story of AI in digital analytics is just beginning.
For it to become a real lever, we must first and foremost focus on the essentials: solid, automated processes and accurate, resilient data.
There's nothing magical about it.
Like any form of intelligence, including human intelligence, AI needs this engine and this fuel.
So let's move forward step by step, but with determination.
At Data On Duty, this is the philosophy we adopt: combining method, technology and pragmatism to build a future where AI will finally live up to its promises.
For the application of creative AI in digital analytics, we'll have to wait a little longer.
