Data Quality software and AI: adding features is not enough to create value
Artificial intelligence is becoming a driving force in software evolution, fuelled by rapid adoption and rising market expectations. According to the Top 250 Numeum EY report, 83% of software publishers now rank AI among their technological priorities, while more than 60% have already integrated generative AI capabilities into their solutions. This acceleration marks a clear shift: software is no longer limited to executing predefined rules. It is increasingly expected to learn, analyse and support decision making.
However, AI integration is often superficial and remains largely approached through the lens of feature accumulation. This raises a critical question: how can AI’s technological potential be translated into real, sustainable operational value? In domains such as Data Quality, where reliability and consistency are essential, the challenge lies in combining AI with existing mechanisms in a way that keeps value controlled and measurable over time.
Beyond AI integration: transforming software value
Today, AI is reshaping software market standards and forcing publishers to rethink how their solutions create value. In practice, however, there is still a strong temptation to add AI as an additional layer that is visible, but poorly integrated and loosely connected to real business use cases.
Beyond this surface level approach, a deeper transformation is already under way. Software is moving beyond simple rule execution to handle more complex processes that blend rules, analysis and decision support. As a result, integrating AI is neither about multiplying features nor about replacing existing application logic. It is about selectively enhancing processes with targeted capabilities. The goal is clear: to create long term, controlled value by making these different components work together effectively.
In a Data Quality solution, this approach consists of integrating AI alongside established processes that guarantee data reliability. Core mechanisms such as standardisation, validation and deduplication continue to rely on explainable, fully controlled software logic. AI complements these foundations by bringing additional analytical and contextual capabilities, particularly when interpreting heterogeneous or unstructured data. AI is applied in areas where rule-based approaches fall short, especially when working with ambiguous data, free text inputs or situations that require interpretation. The principle is straightforward: use AI where it adds value beyond what traditional processing can achieve.
Hybridisation: enhancing what exists rather than replacing it
Integrating AI into an existing software solution is not a single binary choice, but a series of trade-offs guided by business use cases. A common mistake is to reduce AI integration to generative models alone, without fully considering their limitations. Their potential inconsistency and lack of explainability make them unsuitable for certain critical processes, particularly in data quality and compliance-driven environments. In practice, technology choices should follow a simple principle: using the right engine for the right use case, based on the required levels of reliability, explainability and user interaction.
This approach naturally leads to combining several families of models in a complementary way. On the one hand, explainable approaches are often based on probabilistic models. They are particularly well suited to structured processing, where robustness and result traceability are paramount. On the other hand, deep learning models are used in user interaction contexts, notably when processing natural language queries or
supporting real-time actions. Rather than prioritising a single technology, the objective is to orchestrate these building blocks according to specific needs.
In the field of Data Quality, this hybrid approach translates into a clear division of roles:
- Core data processing (including standardisation, validation and deduplication) is handled through deterministic rules and proven, domain specific algorithms built into the software. This approach ensures reliable, traceable and reproducible results, even at scale.
- AI models are integrated as complementary capabilities, applied where they create distinctive value: interpreting free-text data, analysing the root causes of anomalies or assisting users in configuring quality rules.
This approach introduces a form of operational intelligence into existing processes. AI no longer merely detects errors; it helps explain why they occur, suggests corrective actions and supports users in their decision-making. It can also provide additional insight into results by placing them back into their business context. In doing so, AI acts as a lever for targeted enhancement without calling into question the mechanisms that ensure data quality.
Structuring sustainable AI within software
Beyond simple integration, AI raises questions about long term sustainability. Not all use cases deliver the same value, and their relevance cannot be assessed solely in the short term. For a software publisher, the challenge is to embed AI within a coherent product roadmap; one that delivers continuous value without undermining control over the software.
Building on its business-driven and hybrid approach, DQE’s recommended strategy is built around three core pillars, which also includes a strong focus on responsible AI usage:
- Business driven AI - Integrating capabilities that address concrete, clearly identified business problems. Not using AI as a technological showcase. In the field of Data Quality, this means targeting use cases where AI genuinely improves data quality, without weakening existing, trusted processes.
- Pragmatic hybridisation - Combining technologies according to actual needs, applying the right engine to the right use case rather than generalising a single model. This approach makes it possible to articulate AI models with proven application logic, preserving the reliability and explainability of results.
- Responsible by design - Embedding performance, cost and efficiency constraints from the outset. In practice, this translates into a principle of digital frugality: limiting the use of expensive or unstable models for critical processing, favouring approaches suited to data volumes and operational requirements, and avoiding unnecessary resource consumption.
At software level, the challenge is therefore not simply to integrate AI, but to control its use over time. A software publisher is not just another AI user: it acts as the orchestrator of AI usage, ensuring reliability and controlling its impact. Structuring sustainable AI ultimately means striking the right balance between innovation and control in order to preserve relevance, stability and independence from individual models.
About DQE
Because Data quality is essential to customer knowledge and the construction of a lasting relationship, since 2008, DQE has provided its clients with innovative and comprehensive solutions that facilitate the collection of reliable data.
18
Years of
expertise
800
Clients in all
sectors
10Bn
Queries per
year
240
Internationnal
repositories
