Everyone is talking AI. Everyone is building AI.

The fact is, we are now living in an AI-ready and AI-fueled world. Some businesses are just starting to dip their toes into AI, while others are already using it to drive innovation and shape strategy.

But here’s a shocking truth: Over 85% of AI projects fail.

And as business leaders, what could you do to avoid the 85%?

AI Projects Crash And Burn

AI isn’t the magic bullet it’s sometimes made out to be. A report from Gartner highlights that a staggering 85% of AI projects fail to meet expectations, leaving companies with little to show for their efforts – and sometimes, a lot less in the bank.

New research from data management platform Qlik reveals that 11% of UK businesses have more than 50 AI projects stuck in the planning stage. Meanwhile, 20% have had up to 50 projects progress beyond the planning phase, only to be halted or cancelled due to various setbacks. 

“AI has the potential to impact nearly every industry and department, but it’s not universally applicable,” says James Fisher, Chief Strategy Officer at Qlik.

The key factors? 

Poor data quality, inadequate data availability, and a lack of understanding about AI’s requirements are at the core of many failures. AI models, without access to clean, relevant, and accurate data, simply cannot perform. According to a 2023 McKinsey report, 70% of AI projects fail because of issues with data quality and integration.

Data, however, is not the only case. 

Define Your Mistakes - And Own Up To It

The Data Kryptonite

• The recent public debacle involving law firm Levidow, Levidow & Oberman serves as a cautionary tale. The firm used ChatGPT to generate legal opinions, which contained fake quotes and citations. The results were disastrous: the firm faced legal fines and a PR nightmare. The firm and its lawyers “abandoned their responsibilities when they submitted non-existent judicial opinions, then continued to stand by the fake opinions after judicial orders called their existence into question,” a judge said in a June ruling, which also levied a $5,000 fine.

If there’s one thing I learned from my time integrating generative AI chatbots like ChatGPT, Gemini, or Claude, even the free and the upgraded, premium, newest version, one problem always stuck with me - their data is never up to date. AI models can only work with the information they are trained on, and if that data isn’t accurate or current, the results will be unreliable. “AI applications are only as good as the data they are trained on,” says Troy Demmer, co-founder of Gecko Robotics. “Trustworthy AI requires trustworthy data inputs.”

Worldwide data creation is projected to grow to more than 180 zettabytes by the end of 2025. With so much data available, possessing quality data starts with having a complete picture of the information generated by your organization. These issues also highlight the need for meticulous data management and strategic planning, including the integration of cloud-based models and large language models (LLMs).

Rising Cost - Insufficient Funding

Generative AI tools might appear cheap and accessible at first. But when companies move from pilot projects to full-scale deployments, the costs quickly spiral out of control. Gartner estimates that a retrieval-augmented generation (RAG) AI document search project can cost up to $1 million to deploy, with recurring costs of up to $11,000 per user annually. In more specialized fields like medical or financial AI models, costs can exceed $20 million.

Not to mention, 42% of companies face setbacks due to inadequate funding or resource allocation. 

AI isn’t cheap, and pilot projects that produce no value can be money pits. You want to integrate feature A in your AI bots, while still maintaining feature B, and oh, let’s add in feature C. And that will be, say, another $30,000 to $300,000 more. 

And we don’t even know if it will work that well. 

Chasing Shiny And Unrealistic Expectations

Thanks to the crazy hype, every business leaders are now seeing AI as a magic bullet. If it’s not “AI-powered” or ‘using AI”, they don’t want it. 

Expectations often exceed what AI can deliver, leading to frustration when the technology fails to meet the hype. As Ajgaonkar, CTO of product innovation at Insight, points out, some leaders expect AI to magically predict things like pricing without considering the complex data preparation and training required.

The key to machine learning success is constant tuning. “In AI engineering, teams often expect too much too soon,” explains Shreya Shankar, a machine learning engineer at Viaduct. “They don’t build the infrastructure needed to continually test and improve the system.” 

Business leaders often expect AI to effortlessly analyze historical data, pull relevant insights, and apply them to new customer requests, such as predicting purchasing behavior based on past transactions. Instead of doing the necessary groundwork – cleaning data, testing, and retraining models to ensure accurate results – they’re eager to jump straight to the end goal, bypassing the critical steps that drive success.

This, in turn, leads to unrealistic expectations.

The real key to machine learning success is something that is mostly missing from generative AI: the constant tuning of the model. It’s all the work that happens before and after the prompt, in other words, that delivers success.

Siloed Teams, Failed Collaboration: The Blind Leading The Blind

No one really noticed this,

But the common cause of AI failure isn’t really about the technology, sometimes, it’s about the people. 

This starts with the people at the top - and their view on AI. 

Business leaders frequently misinterpret the problems AI is supposed to solve. Many executives also have inflated expectations, fueled by the hype around AI from sales pitches and flashy demos. They underestimate the time, resources, and careful planning needed for AI to succeed. Often, models are delivered at only 50% of their potential due to shifting priorities and unrealistic timelines."

Deloitte found that 40% of companies struggle because their technical and business teams aren’t aligned. Even if the AI model works technically, if these teams don’t work together, the project often fails to deliver tangible value to the business. Additionally, many engineers and data scientists are drawn to the latest technological trends, even when simpler solutions would suffice. 

Teams may spend time on cutting-edge technologies that don’t necessarily address the core issue.

Check The Boxes: The Five Phases Of AI Readiness

No matter the size of your business, don’t panic.

If you’re feeling uncertain about your AI product (still), there’s a simple way to check your progress. 

Just run through the five phases of AI readiness. If you’ve ticked all the boxes, you’re on the right path.

Awareness - The Knowledge Bases

At this stage, your goal is to build awareness of AI and how it can be applied to your industry. Educate leadership through workshops and seminars, research AI use cases for your organization, and identify where AI can solve real business problems. Research shows that 60% of organizations are still in this phase, with no formal AI initiatives in place.

• A manufacturing company exploring AI might find that predictive maintenance could reduce downtime by 20-30%, saving millions annually. But first, they need to understand the basics of how AI works.

Exploration - Start Small

In this phase, businesses experiment with small-scale, low-risk AI projects to demonstrate its potential. These pilot projects are often low-cost and involve small teams (e.g., one data scientist and one engineer). Gartner reports that 25% of companies in this phase see measurable returns within six months of starting AI pilots.

A focused, straightforward pilot helps secure stakeholder buy-in, provides early insights to refine your strategy, and sets the stage for more complex AI applications in the future.

Operationalization – Building Scalable Infrastructure

Once you’ve moved beyond pilots, it’s time to build the infrastructure needed for scalable AI adoption. This includes setting up governance structures, ensuring data privacy, and deploying AI in real-world use cases. 

Establish an AI Center of Excellence (CoE), create scalable data platforms like data lakes, and develop policies for compliance. McKinsey reports that companies in this phase see a 20% improvement in operational efficiency.

•  Use AI-powered routing to escalate critical cases, such as VIP churn risks or sensitive issues, while allowing AI to handle routine queries. By setting clear business rules, AI can make accurate distinctions between scenarios and smoothly hand off more complex cases to human agents, ensuring the right support at the right time. Liberty London uses AI to direct customer service inquiries based on agent skillset and customer intent, streamlining the process. This approach resulted in a 73% reduction in first reply time and a 9% boost in customer satisfaction.

Proficient - Making AI A Part Of Everything

AI becomes part of everyday operations. Businesses establish systems to monitor the performance and fairness of AI models while training employees to use AI tools effectively. AI solutions are scaled across departments, and employees are trained to integrate AI into their daily roles. 

The crucial element of AI readiness here is human involvement. By analyzing both AI-resolved and human-assisted issues, you can gain a comprehensive view of performance. Track key metrics, such as automated resolution rates, human escalation frequency, and customer satisfaction, to refine and improve the process.

Leader - An “AI-first” culture

The final phase is where businesses fully integrate AI into their core strategy, operations, and innovation. Companies at this level use cutting-edge techniques like generative AI and autonomous systems to drive competitive advantage. They foster an AI-first culture through continuous employee upskilling.

Only 10% of organisations are at this stage, but they account for 70% of all economic gains from AI.

Be Part of the 15%, Not the 85%

This isn’t a one-size-fits-all solution for every business leader, nor is it the ultimate guide to creating a perfect AI model or product for your company.

But there is one thing you, as a business leader, can learn from.

Success doesn’t hinge on avoiding failure—it’s about learning from it and adapting. 

If your business is struggling with AI, the problem may not lie with the technology itself, but with how it’s integrated into your organization. Take a step back, and check the boxes. The key to AI success starts with a solid foundation: ensuring alignment between your teams, setting realistic expectations, and creating the right infrastructure to support growth.