Abhijit Kakhandiki is senior vice president and general manager of Digital Business Automation at BMC Software, Inc.
Enterprises aren’t struggling to build AI. They’re struggling to make it matter.
I recently spoke with the CIO of a large global enterprise whose data science teams had built more than 200 AI models, yet fewer than 10 were in production.
That gap is the defining pattern of enterprise AI today. Organizations are rich in experimentation but poor in production. McKinsey’s 2025 global AI survey found that while AI use is widespread, most organizations remain in experimentation phases, with only about one-third reporting they have begun scaling AI programs across the enterprise. It’s tempting to blame talent, tooling or model performance, but the issue is that enterprises treat AI as a technology problem when it’s an operational one. The hard part is running AI reliably, governing it and integrating it into the workflows the business already depends on.
Big data and machine learning followed the same trajectory, where early innovation outpaced enterprise readiness, and later operational discipline caught up. AI is now at that inflection point. What distinguishes the organizations moving forward is not better models, but an understanding of what it takes to make AI part of how the business runs.
Three persistent breakdowns explain why AI stalls before production.
1. Starting With Technology Instead Of Business Outcomes
Thousands of vendors populate the AI and data landscape with an excess of tools, and enterprises get caught up in assembling technically sophisticated solutions without establishing what success looks like.
A team I met with built a complex generative AI solution with RAG pipelines, vector databases that needed regular refresh and fine-tuned models. When asked which business KPI it was improving, there wasn’t a clear answer. Months later, despite significant investment, the system remained stuck in pilot.
“Improving productivity” is a frequent objective, but a model that improves accuracy by five percent is meaningless unless it reduces costs, accelerates processes or drives revenue.
Organizations that move to production flip the starting point. They define success in business terms first, such as cycle time reduction, cost savings or the avoided cost of a missed SLA, and then determine the technology required to achieve it.
Gartner research reinforces that organizations with higher AI maturity are more likely to select initiatives based on business value and technical feasibility, supported by governance structures, engineering practices and measurable outcomes. The presence of executive-level KPIs creates alignment between business and technology teams and establishes outcome ownership.
2. Treating Trust As A Feature Instead Of A Requirement
Even when AI works technically, it often fails organizationally, because enterprises don’t trust what they can’t explain.
Consider a model denying a loan application. If customers ask why, the enterprise must be able to answer with what data was used, which version of the model was running, how the decision was made, and who approved it. Without transparency, the risk isn’t just technical failure; it’s regulatory exposure and reputational damage. Frameworks such as the EU AI Act and the NIST AI Risk Management Framework are putting greater emphasis on transparency, traceability and accountability in AI systems.
Many organizations approach governance as an enhancement to be added later, but lineage and explainability cannot live only in a policy document. Governance is operationalized in workflows that run the models and every execution leaves an audit trail.
Gartner has identified trust as a differentiator between AI success and failure, with high-maturity organizations more likely to report that business units trust and are ready to use AI solutions.
Trust must be engineered into the AI lifecycle initially, and every model must be traceable to its training data and development process. Monitoring must detect drift and bias in real time, alongside the production inference pipelines, before issues translate into business impact.
A healthcare provider we worked with chose a less complex model because it could clearly explain recommendations to clinicians. That decision accelerated adoption and enabled full deployment, while more accurate but opaque models remained in pilot.
AI that can’t be trusted won’t be used and won’t reach production.
3. Failing To Integrate AI Into How The Business Runs
Many organizations assume that once a model is built and tested, the hardest work is done, but that’s when the challenge begins.
AI is often introduced as a standalone capability rather than integrated into the broader operational fabric of the enterprise.
A financial services organization implemented an AI model within its transaction-processing workflow. The model performed as designed, but the AI service wasn’t incorporated into the company’s SLA framework, and business SLAs were breached. The model succeeded, the system failed.
Leaders begin asking operational questions:
• How do we orchestrate AI alongside the batch processes and business systems that already run the enterprise?
• How do we ensure reliability, including observability, failure recovery and clear ownership, across increasingly complex, multi-step processes?
• How do we manage cost, performance and scale when inference and agents run around the clock?
The answer: AI in production must be treated like any other mission-critical capability subject to the same disciplines of orchestration, SLA management, governance, monitoring and cost control.
This becomes more urgent as AI systems evolve from passive models to active agents that can recommend, initiate and execute actions. As autonomy increases, so do requirements for oversight, auditability and control.
AI fails because it isn’t embedded into the systems and processes that keep the business running.
From Experimentation To Execution
The enterprises that succeed with AI are not the ones with the most advanced models. They are the ones that operationalize AI as part of their core business infrastructure, treating AI orchestration as an extension of the automation backbone they already trust.
They start with business accountability, not technical possibility. They build trust into every stage of the lifecycle. They integrate AI into the operational backbone of the enterprise, rather than treating it as something separate.
AI will reach production at scale; it’s only a matter of time. For most organizations, the breakthrough will come from recognizing that the real work of AI isn’t building it, but making it real.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?