Many enterprise IT leaders express high confidence in their organization’s readiness for AI-driven commerce. A recent global survey of B2B and B2C senior IT and technology decision-makers, commissioned by Akeneo, reveals that 95% believe their organization is ready for AI, and 87% anticipate an increase in their AI budgets over the coming years. However, this outward confidence often masks significant underlying challenges related to data quality, governance, and foundational infrastructure. The survey uncovered ten “hard truths” demonstrating that true AI readiness is less about technological ambition and more about disciplined investment in the core data assets that underpin all AI initiatives.
The Foundation IT Owns: Governance, Data, and Strategic Alignment
Effective AI implementation in enterprise commerce requires a robust, well-governed product information foundation. Without this, AI initiatives risk inefficiency, error, and significant operational debt. The survey indicates that IT is increasingly shouldering the responsibility for this critical foundation.
AI’s Demand for Robust Enterprise Foundations AI is not merely a layer to be added on; it exposes and amplifies existing inefficiencies within an organization’s product data infrastructure. The survey found that 63% of IT organizations now own the AI strategy, indicating a shift in accountability for these new capabilities. This means IT leaders are tasked with integrating AI within existing systems, including ERP, PIM, DAM, supplier feeds, and commerce platforms. Brittle point-to-point integrations, localized taxonomies, manual mapping, and disconnected workflows become major impediments, increasing the cost of technical debt and hindering scalable AI adoption. For instance, a major retail corporation attempting to deploy AI for personalized product recommendations found its progress stalled due to inconsistent product attribute definitions across multiple regional ERPs, requiring extensive manual reconciliation efforts.
The Criticality of Product Content Governance While AI accelerates content generation, concerns persist regarding the provenance and approval of that content. The survey highlights that nearly 80% of IT teams are involved in AI approval and governance processes, and 54% of senior IT decision-makers are directly responsible for AI data quality, governance, or compliance. This reflects a growing recognition of the risks associated with “Shadow AI,” where business units independently deploy AI tools without centralized oversight. An example in financial services might be a regional marketing team using a generative AI agent to draft compliance statements for a new investment product, only for it to use outdated regulatory language not approved by the legal department, leading to significant exposure and potential fines. Without clear governance and audit trails, AI-generated content can lead to incorrect pricing, outdated product claims, or misaligned brand messaging, directly impacting customer trust and regulatory standing.
Bridging the Governance Policy-to-Operation Gap Having an AI governance policy in place is a crucial first step, but operationalizing it for accountability and traceability remains a significant hurdle. The survey reports that 99% of organizations either have an AI governance policy or are actively developing one. However, the real challenge lies in establishing decision lineage: knowing precisely who or what (e.g., which agent, which model) initiated a change to a product record, what data was used, and which human ultimately approved it. In a telecom company, an AI agent might propose a new promotional bundle price. Without AI Site Reliability Engineering (SRE) principles—which extend operational rigor, monitoring, incident response, and change control from traditional software to AI systems—it is nearly impossible to track the full approval chain, posing risks to pricing integrity and compliance. This lack of an auditable record transforms governance from a strategic statement into a fragile operational control.
Summary: IT’s expanded ownership of AI strategy places a direct responsibility on them to address foundational infrastructure weaknesses, establish rigorous content governance, and operationalize AI policies with clear decision lineage and audit capabilities.
Addressing the Product Data Readiness Gap
Despite widespread belief in data readiness for AI, the practical reality of data preparation consumes significant resources and effort. The gap between perceived readiness and operational reality is substantial.
The Reality of “AI-Ready” Data Most IT leaders express confidence in their data, with 95% stating readiness for AI commerce and 92% rating their data quality as “good” or “excellent.” Yet, a stark contradiction emerges: 53% of these organizations spend 26–50% of their total AI project effort on data preparation, with 36% spending more than half. This “data preparation” often involves labor-intensive tasks such as chasing missing attributes across teams and suppliers, reconciling duplicate SKUs, normalizing units and taxonomies, resolving conflicting values, and reformatting data for various channels. For an e-commerce platform, achieving “AI-ready” product descriptions often means manually correcting thousands of inconsistent product dimensions, translating marketing jargon into structured attributes, and ensuring price parity across multiple regional catalogs. This hidden manual effort, rarely measured or funded, is what truly underpins perceived AI readiness.
Structured Systems as AI’s Essential Control Layer Raw enterprise data, fragmented across numerous systems and business units, is often too messy for direct AI consumption. As a protective measure, 65% of organizations now leverage structured enterprise systems like Product Information Management (PIM) and Enterprise Resource Planning (ERP) to power their AI models. These systems act not as temporary workarounds but as essential control layers. For example, in a B2B SaaS company, customer entitlement data might reside in one ERP, pricing in another, and product features in a PIM. An AI assistant recommending upgrades requires a unified, trusted source of truth to avoid suggesting incompatible features or incorrect pricing. PIM and ERP provide the necessary definitions, update cycles, and reconciliation capabilities to ensure AI models access trustworthy information, mitigating risks of errors that could be amplified across customer interactions.
What to do:
- Conduct a comprehensive data audit: Map all product data sources, identify inconsistencies, duplicates, and gaps. Prioritize data quality initiatives (e.g., standardizing taxonomies, attribute definitions).
- Invest in master data management (MDM) solutions: Implement or strengthen PIM and ERP systems as central hubs for trusted product data, ensuring a single source of truth for AI consumption.
- Allocate specific budget for data preparation: Recognize that data readiness is an ongoing operational cost, not a one-time project. Budget for data engineers and specialists who can maintain data quality.
- Establish data ownership and SLAs: Clearly define who is responsible for specific data domains and set service level agreements for data accuracy and completeness.
What to avoid:
- Assuming data is “AI-ready” without verification: Do not rely solely on self-reported readiness; conduct rigorous, practical assessments of data quality.
- Bypassing structured systems for raw data feeds: Resist the temptation to directly feed AI models with unverified, fragmented data from disparate sources.
- Underestimating the cost and effort of data preparation: Failing to adequately budget and staff for data quality initiatives will impede AI scalability.
- Developing AI models in isolation from data governance teams: Ensure tight collaboration between AI development and data governance from the outset.
The Automation and Agent Paradox: Scaling with Trust
The strategic deployment of AI agents and automation is critical for scale, yet it introduces complexities related to orchestration, consistency, and trustworthiness.
Automation as a Strategic Imperative, Not a “Flashy Use Case” For IT leaders, the primary driver for AI investment is often not flashy customer-facing applications, but core automation and operational efficiency. The survey reveals that IT automation (57%) and IT operations (55%) outrank conversational AI (48%) and personalized search (40%) as investment priorities. This is because the sheer scale and complexity of modern commerce operations—managing millions of SKUs across numerous markets and channels—exceeds human capacity. Automation can maintain product attributes, ensure compliance updates, and facilitate localization without proportional increases in headcount. However, each narrow AI agent deployed without a shared orchestration layer can introduce new security, monitoring, and coordination challenges, leading to “agent sprawl” where multiple agents act on the same product records without understanding each other’s actions or cumulative risks.
Ensuring Coherent and Trusted AI-Powered Experiences AI agents are projected to be the top channel influencing future customer interactions (57%), with customer support already a leading use case (49%). Yet, significant obstacles remain, including integration challenges (28%) and data silos (18%). The core issue is ensuring every AI-assisted touchpoint accesses complete, current, consistent, and governed product information. Consider a healthcare provider using an AI assistant for patient support. If the assistant recommends an incorrect dosage or provides an outdated procedure guideline due to fragmented data, it can lead to patient harm, regulatory violations, and severe reputational damage. An AI sales tool in retail that pulls an incorrect warranty claim or an unapproved sustainability claim can erode customer trust and increase returns. Protocols like Agent2Agent communication can standardize information exchange, but the overall experience remains fragile if the underlying data handoff is weak.
The AI Harness: Enabling Reliable AI Action The missing element in many AI deployments is the “AI harness.” This concept, emerging from AI engineering research, describes the comprehensive infrastructure built around an AI model that enables it to act reliably, not just respond. An AI harness encompasses memory, permissions, guardrails, orchestration, and audit trails. In commerce, where AI actions impact pricing, compliance, and customer perception, this infrastructure is non-negotiable. It extends the established disciplines IT teams have built into ERP, PIM, and DAM systems for years—such as approval workflows and access controls—to AI agents. For example, a financial institution deploying AI for customer service interactions must ensure the AI operates within strict regulatory compliance guardrails (e.g., “no personal financial advice”), adheres to data privacy policies (e.g., “do not store Personally Identifiable Information indefinitely”), and logs every significant interaction for audit. A model with a harness can safely execute actions within systems of record; a model without one can only generate plausible text.
Immediate Priorities (First 90 Days):
- Define AI Agent Orchestration Strategy: Map existing and planned AI agents. Develop a framework for centralized management, monitoring, and inter-agent communication to prevent conflicts and ensure consistent information delivery.
- Establish AI Data Access Policies: Implement granular access controls and data masking policies for AI models, ensuring they only access approved and necessary data (e.g., using role-based access control (RBAC) for data endpoints).
- Pilot AI SRE Practices: Introduce basic AI Site Reliability Engineering principles, focusing on monitoring AI model performance, data drift, and unexpected outputs. Establish clear incident response and escalation paths for AI errors.
- Integrate Key Data Sources: Prioritize integrating essential product, customer, and compliance data from PIM, ERP, and DAM into a unified, governed foundation accessible to AI.
Summary
The growing investment in AI is undeniable, with 87% of IT leaders anticipating budget increases. However, the survey’s “hard truths” reveal a critical disconnect: 89% of organizations are still spending over a quarter of their AI project budgets on data preparation. This suggests that without a robust, reusable product information foundation, increased AI spending risks merely layering new tools on top of existing fragmentation, slowing progress, increasing risk, and failing to scale.
True AI readiness for agentic commerce is not about the sophistication of the AI model itself, but the strength of the underlying “AI harness”—the infrastructure of permissions, memory, orchestration, and governance that allows AI to act reliably and trustworthily within enterprise systems. Organizations that prioritize building this foundation, ensuring data quality, clear governance, and auditable decision lineage, will be the ones that effectively scale AI, secure customer trust, and realize measurable outcomes in the competitive landscape of AI-driven commerce. The time to build this essential harness is now.
