MaPS AI: a governed AI orchestration architecture for your product data

Artificial intelligence is gradually transforming the role of product information management platforms.

Having centralised, structured and improved the reliability of data, PIM and MDM solutions must now enable businesses to use that data faster, at greater scale and in accordance with their corporate governance policies.

With MaPS AI, MaPS System adds an intelligent orchestration layer to its platform. Its purpose is not to impose yet another artificial intelligence model, but to enable each organisation to select, configure and manage the technologies best suited to its requirements.

The objective is clear: to make AI an operational component of the product information system, without relinquishing control over the architecture, data validation or costs.

From a centralised PIM to an intelligent, operational platform

For many years, the primary role of a PIM or MDM system was to create a reliable central repository: centralising catalogues, harmonising attributes, monitoring information quality and distributing data to the relevant channels and stakeholders.
This function remains essential. However, it is no longer sufficient to meet today’s challenges.
Businesses must manage increasingly large catalogues, data from a wide range of sources, growing personalisation requirements and an ever-expanding number of channels, languages and formats.
In this context, the challenge is no longer simply to store and organise information. Businesses must also be able to enrich, validate, translate, reformat and process it at scale.
With MaPS AI, the PIM becomes a genuine execution platform.
Embedded within product data management processes, artificial intelligence can be used at different stages of the information lifecycle, including extraction, modelling, classification, enrichment, quality control, translation, editorial adaptation and the preparation of data for different distribution channels.

AI orchestration: choosing the right technology for each task

Not all artificial intelligence models have the same capabilities.
Some are particularly effective at content generation. Others are better suited to document analysis, classification, multimodal processing, translation or the large-scale execution of repetitive tasks.
Relying on a single model for all these use cases can restrict performance, increase costs or create excessive dependency on one provider.
MaPS AI adopts a model-agnostic, multi-model approach. Depending on the chosen configuration, the platform can connect to different artificial intelligence model providers, including OpenAI, Anthropic, Google and Mistral AI. It can also integrate models running within infrastructure controlled by the customer, particularly through Ollama.
MaPS AI therefore acts as an orchestration layer between data, business workflows and different artificial intelligence models.

This approach makes it possible to:

  • select a provider or model according to the nature of the task;
  • reserve the most advanced models for operations requiring more complex reasoning;
  • allocate fast, cost-effective models to repetitive or high-volume processes;
  • change or upgrade the models being used without rebuilding the entire business process;
  • reduce the architecture’s dependency on a single provider.

Orchestration is therefore not simply about creating connections to multiple models. It makes it possible to apply routing, control and governance rules to every use of artificial intelligence.

An architecture aligned with your data governance policy

Integrating artificial intelligence into a PIM raises legitimate questions.
What data is being transmitted? To which provider? In which region is it being processed? What retention policies apply? Who owns the access credentials? How can processing activities be monitored and audited?
MaPS AI is designed to enable the IT department to retain control over these decisions.
Depending on the chosen architecture, connections to AI services can be established using the customer’s own accounts and API credentials, rather than through a generic account shared and managed by the software provider.

This separation makes it easier to apply internal security policies, monitor usage and manage the relationship with each provider.
The organisation therefore retains the ability to determine:

  • which providers are authorised;
  • which models are available;
  • which types of data may be processed;
  • which users are authorised to initiate processing;
  • which processes may involve AI;
  • which stages require human validation.

This architecture helps protect the confidentiality of strategic information, including product specifications, supplier data, technical documentation and pricing information.
Data sovereignty nevertheless remains a comprehensive policy issue. It also depends on the platform’s hosting arrangements, the selected processing regions, the contractual terms of each provider, data retention policies and the security rules implemented by the organisation.
MaPS AI therefore provides an architecture that enables each organisation to apply its own artificial intelligence governance policy.

IT-controlled configuration

MaPS AI enables the IT department to establish a common framework for AI usage, rather than allowing fragmented use cases or ungoverned connections to artificial intelligence tools to proliferate.

Administrators can configure the providers, models and access credentials made available through the platform. They can therefore determine which options users can access and which rules apply to each process.
This centralised approach balances business autonomy with IT control.
Business teams gain access to artificial intelligence capabilities embedded directly within their product information management processes. At the same time, the IT department retains control over technical connections, permissions, available models and governance rules.

Artificial intelligence is no longer an isolated tool used outside the information system. It becomes a governed service integrated into the organisation’s data environment.

A FinOps approach to monitoring and optimising AI costs

Moving from isolated experiments to processing several thousand products fundamentally changes the cost equation.
An operation that is inexpensive when applied to a handful of product records can generate significant consumption when deployed across an entire catalogue, in several languages and on multiple channels.
MaPS AI includes consumption-monitoring capabilities that make it possible to view token usage by model or service.
This visibility helps teams understand the costs associated with different processes and identify the most resource-intensive operations.
Orchestration also makes it possible to apply a cost-performance optimisation strategy by:

  • using cost-effective models for high-volume classification or validation tasks;
  • using specialist models to extract data from documents;
  • reserving the most advanced models for complex enrichment tasks;
  • reducing unnecessary calls through workflow rules;
  • validating results on a sample before launching large-scale processing.

Cost control therefore involves more than simply monitoring an invoice. It begins at the workflow-design stage, by allocating the appropriate level of computing capability to each task.
This FinOps approach enables the IT department to monitor usage, compare costs and progressively adjust the models being used according to observed performance.

Workflows tailored to each organisation’s requirements

Artificial intelligence use cases do not all involve the same level of risk.
Generating a suggested product category does not necessarily require the same level of control as creating a description that will be published automatically on an e-commerce website, or modifying pricing data.
MaPS AI makes it possible to embed artificial intelligence processing within workflows configured according to the organisation’s own rules.
An AI stage may, for example, suggest a description, classify a product or extract technical specifications. The result can then be:

  • checked automatically;
  • compared with existing data;
  • assessed against quality rules;
  • sent to a colleague for approval;
  • accepted, rejected or returned for correction.

The organisation can therefore define different levels of autonomy according to the criticality of the process:

  • automatic execution for simple, reversible tasks;
  • sample-based checks for industrialised processing;
  • systematic human approval for sensitive content;
  • workflow interruption when quality requirements are not met.

This flexibility allows artificial intelligence to be deployed progressively, starting with controlled use cases before extending it to more complex processes.

From unmanaged AI to governed AI

Artificial intelligence does not replace business rules or human expertise.
It automates certain operations, accelerates processing and makes it easier to work at scale. However, its value depends on the framework within which it is used.
Without governance, AI can lead to a proliferation of tools, reduced visibility over transmitted data, uncontrolled cost increases and inconsistent results.
With MaPS AI, the organisation defines which models are authorised, how they may be used, which workflows are concerned and where controls must be applied.
AI performs time-consuming tasks, supports users and helps maintain consistency across processing activities. Business teams retain control over final quality, while the IT department manages the architecture, access rights and consumption.
By integrating orchestration directly into the PIM, MaPS System turns artificial intelligence into a governed driver of operational performance.

Your data. Your way. Your growth.

F.A.Q : MaPS AI and product data governance

MaPS AI uses a model-agnostic architecture.

Depending on the configuration and available connectors, the platform can connect to the main model families provided by OpenAI, Anthropic, Google and Mistral AI.

It can also integrate models running within infrastructure controlled by the organisation, particularly through Ollama.

As models and their versions evolve rapidly, the MaPS AI architecture makes it possible to update the services being used without having to redesign the entire set of business workflows.

MaPS AI permet à l’entreprise d’associer un modèle à chaque type de traitement en fonction de critères définis par ses équipes : nature de la tâche, niveau de qualité attendu, volumétrie, délai d’exécution, coût ou politique de sécurité.
Un modèle spécialisé dans le traitement documentaire peut, par exemple, être configuré pour extraire des caractéristiques depuis un PDF fournisseur.
Un modèle rapide et économique peut prendre en charge une classification à grande échelle, tandis qu’un modèle plus avancé peut être réservé à la création de contenus nécessitant davantage de raisonnement ou de contextualisation.
MaPS AI applique ensuite les règles de routage et les choix de configuration établis dans le cadre de la gouvernance IA de l’entreprise. La plateforme ne sélectionne donc pas de manière autonome le modèle à utiliser : ce choix reste défini et piloté par l’organisation.

The model-selection process can be configured according to a range of criteria, including the nature of the task, the required quality level, processing volume, execution time, cost and security policy.

For example, a model specialising in document processing may be used to extract specifications from a supplier PDF.

A fast, cost-effective model may handle large-scale classification, while a more advanced model may be reserved for content-generation tasks requiring greater reasoning or contextual understanding.

These choices are defined as part of the organisation’s AI governance framework.

A model specialised in document processing can, for example, be configured to extract characteristics from a supplier PDF.

A fast and cost-effective model can handle large-scale classification, while a more advanced model can be reserved for creating content that requires greater reasoning or contextual understanding.

MaPS AI then applies the routing rules and configuration choices established as part of the company’s AI governance framework. The platform therefore does not select the model autonomously: that choice remains defined and controlled by the organisation.

 

MaPS AI can use the accounts and API credentials owned by the customer to connect to authorised providers.

This architecture avoids routing all customers through a single provider account shared by the software publisher and helps maintain separation between different uses.

Actual confidentiality also depends on the selected providers, contractual options, processing regions, retention policies and environment configuration.

These parameters should be assessed by the IT department, together with the teams responsible for security and compliance.

MaPS AI enables organisations to monitor consumption by process and model.

This visibility can be supplemented by orchestration rules, including the use of cost-effective models for high-volume processes, limits on the number of calls, sample testing and approval before large-scale execution.

The aim is to reserve more expensive models for tasks that genuinely require their capabilities.