A readiness crisis is unfolding — in enterprise AI. BCG surveyed more than 1,250 companies about their AI initiatives in 2025 and found that only 5% are generating value at scale. Sixty percent reported minimal gains despite substantial spending.BCG, AI Radar 2025. Only 5% of surveyed companies are generating value from AI at scale; 60% report minimal gains despite substantial investment.
That failure mode is remarkably consistent. In the dozens of ERP projects I've guided over the past decade (my engineering teams were also involved), the pattern repeats: companies invest in AI tools, point them at their ERP data, and watch the results disappoint. Any single deployment is a striking case, but readiness gaps are widening everywhere.
This points to the fundamental reason that AI-powered ERP will disappoint rather than transform: the loss of preparation. The first step to adding AI to your ERP isn't choosing a vendor. It's making your ERP ready for one.
Your Data Isn't What You Think It Is §
Some say this is overblown. That ERP vendors can handle AI integration through embedded features. In theory, organizations enjoy less manual processing, more automated decisions, and higher accuracy due to built-in AI agents. From this perspective, adding AI to an ERP isn't a problem but a turnkey upgrade.
Alas, these hopes are misplaced.
Consider your ERP data, often misconceived as a single reliable source of truth. In fact, years of manual entries from dozens of departments are layered onto shared databases, a system of cross-functional data sharing that depends on consistent naming conventions, field definitions, and maintenance routines. As organizations layer AI onto this foundation, the cracks show immediately.Gartner, 2025. 63% of organizations either don't have or aren't sure whether they have the data management practices AI requires.
Gartner found that 63% of organizations either don't have or aren't sure whether they have the data management practices AI requires. In concrete terms: an accounts payable tool reading vendor records with three different spellings of the same company generates duplicate payments. A demand forecasting model trained on inventory data with duplicate SKUs produces garbage predictions.
Before evaluating any AI tool, run a data quality assessment on your three highest-volume ERP modules. Count duplicates, measure completeness rates, and identify fields that haven't been updated in 12 months. This typically takes two to four weeks and costs a fraction of a failed AI deployment.
Map The Process Before You Automate It §
When companies talk about AI readiness, what they're really confronting is the gradual dissipation of a vital operational resource: documented institutional knowledge. Many assume the solution is simple: buy an AI platform and workflow problems will solve themselves.
Yet while purchased AI can automate surface-level tasks, it can't reverse structural gaps in how work actually gets done. Most implementation challenges (roughly 70%, according to BCG) stem from people and process issues, not technology.BCG, 2025. 70% of AI implementation challenges stem from people and process issues, not from the technology itself. Exception-handling for accounts payable lives in the heads of three people who've done it for a decade. Procurement approvals differ by department in ways the ERP doesn't reflect.
Prepared organizations don't just deploy AI tools; they reimagine how work should be done. They don't just participate in digital transformation; they reshape it. Map the actual workflow — not the documentation version, but the one people follow. The AI's job is to augment documented decision points, not replace processes nobody has written down.
Start With One Workflow, Not A Transformation §
The central challenge for any enterprise is increasing AI capability while preserving operational stability. That requires practical approaches that separate meaningful AI performance from its historical prerequisite: expensive, multi-year platform overhauls. The solution isn't to reverse existing ERP investments but to reimagine AI deployment within a functioning enterprise.
Your ERP vendor offers what seems like a blueprint for AI adoption: SAP's Joule agents now ship in two-thirds of cloud deals, Oracle embedded AI agents at no additional cost, Microsoft's Copilot works across Dynamics 365. Yet the results are sobering. McKinsey's enterprise AI data sits at 40% EBIT impact, well below the transformation vendors promise.McKinsey, enterprise AI data. 40% EBIT impact reported — well below the transformation that vendor marketing promises. While these embedded features may have prevented even steeper adoption gaps, they haven't reversed the trend.
A workflow-first approach offers a more promising model. Platforms like CogniAgent let you build AI-powered workflows on top of existing ERP data, connecting conversational AI and process automation to business systems without replacing the ERP itself. Pick a tool matched to your specific use case, not the one with the most impressive demo.
How Urgent Is This? §
Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027 due to escalating costs and unclear value.Gartner, 2025. Predicts over 40% of agentic AI projects will be canceled by end of 2027 due to escalating costs and unclear value. The problems run deeper. Edinburgh City Council's ERP modernization started at £39 million and ballooned past £130 million, leaving them without an adequate financial system for more than two years.Edinburgh City Council ERP overrun. What began as a £39M modernization ballooned past £130M, leaving the council without an adequate financial system for over two years.
The mathematics of delayed preparation is unforgiving — quarters postponed now compound into years of competitive disadvantage later. Every planning cycle companies defer the unglamorous work, the challenge grows steeper.
Yet none of this calls for panic. AI readiness is a serious challenge, but it's not the transformation emergency that vendor pitches suggest. This reality calls for a measured approach.
What To Do Before Next Quarter §
First, undertake data and process interventions that make sense independently of their impact on AI performance. Run thorough data quality audits on your top ERP modules and document how decisions actually get made, including the informal steps no system captures.
Second, test one measurable workflow — from invoice processing to demand forecasting — and pilot it for 90 days with clear success criteria before expanding.
Too much depends on finding practical solutions. The enterprises that succeed with AI will be the ones that build capability while preserving the operational stability and institutional knowledge they already run on. The future of your AI investment, and of your next quarter's operating results, depends on striking this balance.