Definition
The accumulated cost of expedient, undocumented, or ungoverned AI service integrations into enterprise ERP systems, showing up as increased maintenance burden, reduced reliability, and governance risk over time.
First published July 3, 2026 in Your ERP Has a Hidden AI Problem, and It's Growing Every Day.
The idea builds on two earlier lines of work. Ward Cunningham introduced the technical debt metaphor in 1992. Researchers at Google extended it to machine learning systems in 2015, showing how hidden costs build up in ML systems that are built and maintained carelessly.
AI Integration Debt is the next layer, and it is specifically an enterprise problem. The key word is accumulated. It is invisible until it is expensive, building up integration by integration until an organization has AI dependencies it cannot fully inventory, audit, or update.
Why ERP is especially exposed
ERP is the operational spine of the enterprise. When AI is integrated there, the integration becomes a business process dependency. A faulty AI call in procure-to-pay doesn't produce a bad prediction. It produces a bad purchase order.
Long system lifespans
ERP platforms run for decades, so each integration must outlive several generations of AI models and APIs.
Deep process coupling
An undocumented AI step in a payment workflow is a financial control risk, not only a technical one.
Layers of customization
AI sits on top of years of bespoke business logic, so the debt compounds with every layer.
Governance gaps
Change management covers ERP code but rarely model versions, drift monitoring, or rollback. That gap is the debt.
Patterns in the field
Four patterns show up again and again across enterprise environments.
The orphaned integration
Built by someone who has since left. It mostly works, but nobody knows how the prompt is built, what data goes to the model, or how it fails.
The undocumented dependency
A critical process depends on an AI service that appears in no architecture diagram, recovery plan, or SLA.
The accumulating drift
A model calibrated on old ERP data grows less accurate as suppliers, SKUs and processes change, with no monitoring to notice.
The governance void
The integration skipped change control because it was classed as a service call rather than a system change.
Research program
1. Conceptual taxonomy and assessment profile
A conference paper, “AI Integration Debt in ERP Systems: A Conceptual Taxonomy and Provisional Assessment Profile”, is in preparation. It defines a five-dimension taxonomy and introduces the AIDI assessment instrument. It profiles debt across dimensions on purpose, rather than collapsing it into a single composite score.
2. Scoping review
A systematic scoping review following PRISMA-ScR, across five databases: Scopus, Web of Science, IEEE Xplore, ACM Digital Library and AIS eLibrary. It uses four search families:
- Technical debt in AI and machine learning systems
- AI in ERP systems
- ERP debt and legacy systems
- AI governance
3. Practice
The 5-Layer AI-Augmented Development Model, presented at InFocus 2026, applies these ideas to day-to-day JD Edwards development. Its Governance layer is where AI Integration Debt is prevented.
Papers and materials will be posted on this page as they are published.
Timeline
- July 3, 2026Term introduced in the essay “Your ERP Has a Hidden AI Problem, and It's Growing Every Day”
- September 2026Practice framework presented at JD Edwards InFocus 2026, Denver
- In preparationConference paper: conceptual taxonomy and provisional assessment profile
- OngoingPRISMA-ScR scoping review across five databases
How to cite
Until the paper is published, please cite the essay that introduced the term.
Equbal, M. S. (2026, July 3). Your ERP Has a Hidden AI Problem, and It's Growing Every Day. Shahid Equbal on Substack. https://blog.meshahid.ai/p/your-erp-has-a-hidden-ai-problem