Enterprise AI: Why 30% of projects will fail by 2025
What enterprises get wrong about AI
Currently, only 4% of all AI initiatives have created meaningful value for businesses. According to Gartner, by 2025, at least 30% of generative AI projects are expected to be abandoned after proof of concept. A reason for this bold prediction: poor data quality.
Why does data quality play such a huge role in the success of AI initiatives?
AI delivers on its value when it’s working with proprietary business data—like customer behaviours, transactions, pricing, and more. For example, if you’ve got a list of customers who are up for renewal, AI identifies those at risk of churning through specific interactions (website visits, app usage) and then sends sales a list of the most at-risk customers with personalized pre-written emails about each customer's specific account details.
AI and data can deliver on the promises so many enterprise AI initiatives tout. But there are a few roadblocks setting up enterprise projects to be in that failing 30%.
Two enterprise AI anti-patterns to avoid
Some AI vendors are pushing a "just get started" narrative, claiming you don't need comprehensive data infrastructure or that you can achieve success with limited integration — but that couldn’t be further from the truth.
Here’s why all-guns-blazing AI initiatives can cause problems:
1. The data disconnect
Enterprise AI projects are launching without the necessary data plumbing to deliver any specific business value. Companies are investing millions in advanced AI capabilities — with 8% of UK and 7% of U.S. decision-makers planning to spend over $25 million on AI initiatives this year — despite their data remaining scattered across disparate systems, formats, and departments.
Yes, modern AI systems can work with imperfect data. But should they? Gartner's prediction isn't happening in a vacuum — poor data quality is one of the primary culprits. When organizations build on shaky data foundations, they're likely to crumble when scaled beyond limited proofs of concept.
This isn't just about data quality – it's about data architecture, accessibility, and governance.
Without addressing these foundational data issues, you won’t get the results you want from AI. Take customer service for example. If your systems aren’t set up to give human agents access to real-time customer data, don’t expect an AI agent to swoop in and suddenly solve that problem.
2. The implementation oversight
This might be the most expensive mistake. Organizations treat AI implementations as standalone projects rather than integral parts of the business ecosystem.
The pattern is predictable: Companies launch isolated pilot projects that show promising results in controlled environments. Leadership gets excited. But when it's time to scale, everything falls apart. Why? Because these pilots weren't designed with enterprise-wide integration in mind from the start.
We're seeing this with the current wave of AI agent implementations. Companies deploy sophisticated AI agents that can analyze and make recommendations — but these agents exist in their own bubble – unable to execute changes or integrate with core business processes.
Take a Virtual Agent. While it can analyze access patterns and recommend the appropriate software permissions, it's bound by the underlying data infrastructure. Without clean, standardized data about roles, systems, and access policies, the AI can only make basic suggestions that still require manual verification and implementation. Even a sophisticated AI system becomes a glorified ticketing tool if it can't seamlessly integrate with your permission management systems and security protocols.
So, if your team is already overworked and your systems lack access to all the relevant data, it’s unlikely AI will be the silver bullet.
The same rings true for DIY AI systems. You want customer service success, but retraining your LLM model, handling scaling, hiring engineers, and managing rapid change can be an ongoing burden. It’s in these cases that adopting an existing model like Salesforce’s Einstein or Copilot can help you reduce your time-to-market so you can get your AI project out the door much faster.
How enterprises succeed with AI
A strong data foundation
According to IDC, 90% of business data is unstructured — customer contracts, product specs, employee handbooks, and more. This represents massive untapped value, but it's usually scattered across systems.
Organizing your data isn't just about cleaning up – it's about making all that information accessible and usable in real time. You need a flexible data platform that can break down silos and connect information across departments.
Data is the fuel that powers AI, so ensuring it’s accessible, accurate, and available from a single source means your AI projects will be armed with the information it needs to function properly.
An adaptable AI engine
Your AI needs clean, consistent data to learn and adapt effectively. This isn't about having one model that does everything; it's about having the right tools for specific jobs. Different tasks require different models, and you need the flexibility to deploy them where they make sense.
This means building a data infrastructure that can flex and scale for different use cases while maintaining consistent data governance and security standards. When an AI system can adapt to specific business needs while staying integrated with core processes, you move from interesting experiments to real business use cases.
An ecosystem of innovators
Technology alone isn't enough. Success means empowering your team to work effectively with both AI systems and the data that powers them.
It’s important to understand when to let AI handle tasks and when human intervention is needed. This means engineers who understand your customer pain points, analysts who can translate AI insights into business strategy, and domain experts who know when to trust AI and when to apply human judgment.
Without this human layer, you risk building powerful tools that never quite connect with your actual business needs.
What really works…
Instead of rushing to implement AI with whatever you have, focus on building the right foundation:
- Migrate from legacy architecture to a fit-for-purpose cloud environment so your AI systems have a single source of truth with consistently high-quality, accessible data
- Develop clear governance frameworks to guide AI usage, model deployment, and decision-making
- Create integrated workflows — as AI delivers real value only when seamlessly connected to your business processes
- Build cross-functional teams combining technical and business expertise to troubleshoot problems and find AI solutions
Success will come to those who take the time to build proper data infrastructure, establish clear governance, and create truly integrated systems. The question isn't whether you should implement AI — it's whether you've built the foundation to make it succeed.
Let's talk about how we can help you unlock the full potential of AI. Get in touch with us today to get started!
__________
By Barry Sheehan, CCO @ Showoff