What’s Really Holding AI Back in 2026

You've launched an AI project, but is it delivering results?  For most companies heading into 2026, the answer is no. So, how's it going for you? Is everything running smoothly, or is it still a work in progress? Maybe you're told it's “Almost ready." Either way, only a small fraction of AI pilots generates measurable ROI or scale beyond the initial phase.

The data backs this up. McKinsey reports that while 88% of organizations use AI in some capacity, only one-third are scaling it across their business. That's a significant gap between experimentation and impact.

What's holding companies back? It's rarely the model itself. The real obstacles are people, processes, data quality, trust, and governance - all the infrastructure surrounding the technology.

Enlighten Designs examined what's working in successful AI deployments, identified the biggest barriers to enterprise adoption, and developed a practical framework for turning AI pilots into scalable solutions.

What Is Working Right Now

Despite the slow progress across the market, some organisations are getting impressive results. They don’t necessarily chase the latest model drop, but they’re ruthlessly practical about workflow and business value.

A few examples that have been working recently:

• Creating Communities Australia (CCA) optimise an AI powered solution developed by Enlighten Designs for accurate survey data analysis. CCA manages complex data sets, and the AI system generates powerful, high value insights, previously impossible to achieve at that speed. The attention to detail in the analysis streamline processes and positively impacts crucial community initiatives, demonstrating how AI can scale to support meaningful social impact and strategic decision making.

• Enlighten Designs Recruitment Copilot transformed the internal hiring process. This AI solution reduced a process that would require 290 hours of manual work down into just 16 hours for review, eliminating significant administrative overhead tasks. This innovation supercharges recruiter’s workload so they can focus on high value strategic tasks, while also eliminating the inherent bias of first impression reviews to ensure a more equitable and objective candidate experience.

• Bank of America’s very own AI-powered assistant and chatbot, Erica, has already handled more than 2 billion interactions since its launch back in 2018 and continues to serve their customers daily. And this is just the start, as an estimated 80% of bank routine tasks can be handled now by AI by way of automation. Where AI chatbots used to just answer simple questions and were confined to FAQ functions, they can now process more complicated tasks like blocking cards or process loans in seconds.

• Smart factories are now using AI for predictive maintenance and reducing downtime. Siemens has increased their plant availability by up to 30% through reduced production downtime, while Nissan’s unscheduled machine downtime dropped by 30%, improving output consistency across critical assembly lines. These examples indicate that when AI is applied in an industry where downtimes are crucial, it makes a difference in the long run.

• Then there’s the time and money savings through AI-assisted document review. As an example, law firm & technology consultancy, icourts, integrated Relativity’s active learning to save a client 28,000 hours of review time and more than $5 million, where they delivered findings in just 11 days. This is a real value that AI has delivered and saved up valuable resources for both the law firm and its client.

Across all these wins, the same themes show up

• Specific, high-value problems: Smoother customer service, fewer equipment failures, faster document review.

• Worked with what they had: Existing data, established workflows and tools already in use.

• Kept people in the loop: Human oversight guided accuracy, built trust and kept outputs on track.

• Focused on measurable outcomes: Reduced downtime, hours saved, money protected, and customers served faster.

• Built on repeatable processes: Clear owners, defined KPIs and continuous improvement baked into delivery.

The takeaway here is that AI delivers real value when it’s anchored in practical goals. It’s always supported by the same people using it as a tool and embedded in the work culture.

The 10 Blockers in the Way of AI Adoption

So, if AI is already delivering real results, why is there still hesitation to adopt? It boils down to these 10 blocks.

1. Poor Integration with Core Systems: Many organisations still rely on ageing, rigid systems that make AI integration slow or impossible. When AI can’t connect to core workflows, it never becomes part of daily work. CIO research consistently names legacy systems as a top barrier to AI success.

2. Trust & Reliability Issues: Teams quickly lose confidence when AI makes confident but incorrect claims. High-profile cases like the Avianca tech hallucination incident show how AI hallucinations erode trust in professional settings. Simply put, this incident involved a personal injury lawsuit against Avianca Airlines, where the plaintiff's attorney utilised ChatGPT to assist in preparing legal filings to be more efficient. And that’s just one of the possible issues that may deter potential adopters, so until reliability improves, many organisations will hesitate to scale AI.

3. No Clear Purpose for Users: AI adoption drops when people don’t understand why the tool exists or how it helps them. Unclear goals and poor communication are major reasons users resist change, so without a clear purpose, AI becomes another “extra step” rather than a helpful one.

4. Data Infrastructure Gaps: AI depends on clean, connected, and well-governed data. However, many enterprises simply aren’t there yet. Siloed data and outdated architecture stop AI from delivering consistent results. Financial-services leaders repeatedly cite this as a core adoption hurdle. The Stanford AI Index Report 2024 also notes that poor data quality and fragmented infrastructure remain two of the biggest barriers to AI scaling across enterprises. This reinforces that data issues are not anecdotal but a global and well-documented one.

5. Budget Blow outs in Production: Pilots look great, and then the invoice lands. In production, inference costs, bloated prompts, vector storage, and orchestration overhead can turn each interaction into a margin hit. Costs swing with usage spikes and multimodal requests, so budgets get blown and momentum stalls.

6. Weak Workflow Guardrails: Without human-in-the-loop checks, AI transparency or audit trails, AI feels unsafe to use. Governance frameworks emphasise the need for oversight, but many organisations haven’t built these steps yet. As a result, users revert to manual processes.

7. Perception & Confidence Gap: People often struggle to trust AI, even when it performs well. The “AI trust paradox” shows how human-sounding output can confuse users about what is true. Surveys reveal tension between IT and business teams when confidence isn’t built from the start.

8. Lack of Human Connection in AI Interfaces: AI that feels cold or mechanical turns users off, so empathetic and human-centred design dramatically improves adoption. We’re still far away from people comfortably talking to AI, so unless that empathy is present, hesitation to rely on AI for customer-facing or advisory tasks will be present.

9. Poor Change Management & Culture: AI adoption fails when organisations ignore the people side of the transition. Fear of job loss, unclear expectations and lack of training slow progress more than the technology itself. Culture, communication and leadership are often bigger barriers than the model.

10. Unclear Ownership & Governance: AI cannot scale without clear accountability, policies and oversight. Many leadership teams adopt AI faster than they can govern it, creating risk and confusion. Without defined owners and rules, projects stall before they ever reach production.

Once the Blocks Get Cleared Out, What’s Next?

There are far more blocks that can hinder any organisation from adopting AI into reality, but these 10 are the most pressing as of now. Now, if you’re confident these 10 are under control in your organisation, you’re probably itching to get to development and production.

Understandably, you’d want to jump from moving that pilot into something real and then scale it. There’s the assumption that it will be messy from start to finish, but we say ‘Absolutely not’ because there’s a better way to launch your AI ambitions. 

If you’re ready to stop wading through the mess and implement a streamlined, profitable strategy, keep an eye out for Part 2 where we’ll lay out a roadmap to a Simple 90-Day Plan of Moving from AI Pilot to Production.