Understanding the high failure rate of AI projects
Causes of failure
Many AI projects collapse due to misunderstood complexity, poor project management, and unanticipated requirements. Technical complexity is just one dimension of risk. Equally important are organizational factors like alignment between technical teams and business stakeholders. When business leaders lack understanding of AI's capabilities and limitations, projects often suffer from unrealistic expectations and poorly defined success metrics. Similarly, technical teams focused exclusively on model performance metrics may lose sight of the business outcomes that justify the investment.
Infrastructure limitations pose another significant challenge. Legacy systems can create substantial friction for AI implementation, lacking standardized APIs, robust data pipelines, and scalable computing resources. The resulting integration challenges can dramatically increase implementation costs and timelines.
MxD's 2025 Digital Manufacturing Adoption Survey found that 71% of foundational manufacturing organizations don't have the data systems necessary to deploy generative AI. This gap between ambition and infrastructure reality helps explain why so many AI initiatives fail to achieve their intended outcomes.
Barriers like fragmented systems and poor data governance are major drivers of failure. In our research, we've seen mid-market manufacturers struggle particularly with connecting operational technology systems that were never designed to share data. Production equipment, quality control systems, and inventory management platforms often exist as information silos, making it exceptionally difficult to gather the comprehensive, consistent data sets that effective AI requires.
Cultural resistance represents another significant failure factor. Without thoughtful change management strategies, even technically sound AI implementations may fail to gain necessary user adoption. Employees may fear automation will eliminate their roles, leading to passive resistance or active opposition.
Learning from failure
Mark Rickmeier, CEO of TXI, emphasizes that "success isn't about project completion—it's about outcomes." This outcomes-focused mindset helps maintain focus on delivering measurable business value and helps organizations avoid the technical rabbit holes that often derail AI projects.
Successful mid-market AI implementers tend to favor incremental adoption over "big bang" approaches. By starting with narrowly defined, high-impact use cases, these companies establish early wins that build organizational confidence and technical capabilities simultaneously. This measured approach creates a foundation for more ambitious AI projects while minimizing financial risk.
Cross-functional collaboration represents another critical success factor. By involving stakeholders from across the organization—from IT and operations to finance and customer-facing teams—companies create both broader support for AI initiatives and more comprehensive views of implementation challenges. This diverse input helps identify potential obstacles earlier, when they're less costly to address.
Rigorous testing frameworks with clearly defined acceptance criteria help ensure AI systems deliver their intended business value before full production deployment. Testing should evaluate not just technical performance measures like accuracy and response time, but also business outcomes and user experience. For operational use cases, this might include side-by-side comparisons between AI-driven and human decision-making to validate improvements.
The importance of cost understanding and ROI
The ROI of AI Investments
For mid-market companies, AI investments must be justified through tangible business returns. This requires establishing clear connections between AI capabilities and specific business outcomes before significant resources are committed. Industry research demonstrates potential for substantial returns: McKinsey estimates a 3–5% revenue lift via personalization; Bain reports 7,000 hours saved monthly with AI-driven procurement.
TXI's approach to AI ROI assessment follows the Data-Information-Knowledge-Wisdom (DIKW) framework. This progressive model helps organizations understand how value compounds as they move from basic data collection to true actionable wisdom. Each step up the DIKW hierarchy corresponds to increased business value and more sophisticated AI applications.
When constructing ROI models, organizations should consider both direct and indirect benefits. Direct benefits might include labor savings through automation, reduced waste through improved quality control, or increased sales through enhanced customer personalization. Indirect benefits, while harder to quantify, can be equally valuable—improved employee satisfaction through elimination of tedious tasks, enhanced decision quality through better analytics, or increased organizational agility through more accurate forecasting.
ROI calculations should account for the full lifecycle of AI systems, not just initial implementation. Ongoing costs for infrastructure, model maintenance, continuous learning, and user support can substantially impact long-term returns. Similarly, benefits often increase over time as systems improve through additional training and organizational processes adapt to leverage AI capabilities more effectively.
Calculating ROI
Effective ROI calculation begins with establishing appropriate baseline metrics before implementation. Without this foundation, organizations struggle to accurately measure improvements attributable to AI. For predictive maintenance implementations, relevant metrics might include historical data on unplanned downtime, maintenance costs, and equipment lifespan. With these baselines established, companies can more accurately measure AI's impact on reducing downtime, optimizing maintenance resource allocation, and extending asset life.
Time horizons represent another critical consideration in AI ROI calculations. While some benefits—like automation of repetitive tasks—may generate immediate returns, others—like improved decision quality or enhanced innovation—may take longer to materialize. Organizations should structure their ROI expectations accordingly, with appropriate timeframes for different categories of benefits.
Phased implementation approaches enable more accurate ROI tracking by creating multiple measurement points throughout the project lifecycle. This incremental approach allows organizations to validate assumptions earlier and adjust strategies if expected returns aren't materializing. It also helps build organizational support by demonstrating value continuously rather than requiring stakeholders to wait for a distant project completion.
Exploring AI implementation costs
Overview of AI development costs
When considering AI implementation, mid-market companies face distinct challenges compared to their enterprise counterparts. While large corporations might allocate millions to experimental AI research and development, mid-market firms must be more strategic about resource allocation. The Manufacturing Leadership Council's 2025 Talent Outlook report notes that 61% of industry leaders find emerging tech like AI is making it harder to recruit top technical talent.
This talent constraint isn't merely about hiring challenges. Even existing technical teams frequently lack the specialized knowledge needed to effectively deploy and manage AI systems. According to our research, this skills gap extends beyond technical roles to leadership positions, creating a knowledge vacuum that hampers strategic decision-making around AI investments.
We're seeing a fundamental shift in how mid-market companies approach AI investments—treating it as a capability-building journey requiring thoughtful resource allocation across people, process, and technology.
Build-vs-buy becomes a pivotal financial consideration when talent constraints are factored in. While building custom AI solutions offers greater control and potential competitive differentiation, it requires specialized expertise that may be difficult to acquire and retain. Conversely, buying pre-built AI solutions from vendors may offer faster implementation but potentially at the cost of customization and strategic alignment.
Estimating AI costs
Estimating AI development cost involves more than just technology—it includes data collection, model training, infrastructure, testing, and ongoing maintenance. For mid-market organizations, developing accurate cost estimates presents a particular challenge. Unlike established technologies with predictable implementation patterns, AI projects frequently encounter unforeseen complexities that drive up costs.
Initial data cleaning and preparation typically consumes 60-80% of a project's effort. Model training and continuous learning represent another significant cost center that mid-market companies frequently underestimate. Depending on the complexity of the problem and the volume of data, training sophisticated AI models can require substantial computing resources over extended periods. Without proper planning, these costs can escalate quickly, particularly when multiple training iterations are needed to achieve acceptable performance.
While proof-of-concept projects might demonstrate exciting possibilities with minimal investment, transitioning these prototypes into robust, production-ready systems demands significantly more resources. This disconnect between experimentation and production implementation is where many mid-market firms encounter unexpected expenses.
"Companies are moving past the honeymoon phase to embrace the work that matters most: creating value from this tantalizing technology," notes McKinsey's 2025 report "A data leader's operating guide to scaling AI."
Strategic frameworks for AI cost estimation should account for scaling challenges by incorporating phased approaches with clear decision points. By breaking AI initiatives into discrete stages with defined outcomes and costs, mid-market companies can mitigate financial risk while methodically building their AI capabilities.
Factors influencing AI development costs
Cost depends on factors like datasets, access to skilled data scientists, development team structure, and labeling requirements. Other variables include cloud-based infrastructure and algorithm complexity.
Data availability, quality, and structure represent fundamental cost drivers. Organizations with clean, well-organized data assets generally experience smoother, less expensive AI deployments compared to those requiring extensive data preparation. For mid-market companies with legacy systems and inconsistent data governance, this preparation phase can represent a substantial portion of overall project costs.
Algorithm complexity significantly impacts both development and operational costs. While sophisticated deep learning models may offer superior performance for certain applications, they typically demand more computational resources, specialized expertise, and extensive training data compared to simpler approaches. Organizations should carefully evaluate whether advanced techniques deliver sufficient incremental value to justify their additional complexity and cost.
Infrastructure requirements—including computing power, storage, and networking—represent another significant cost category. Cloud services offer flexibility but can become expensive for compute-intensive workloads at scale. On-premises infrastructure provides more predictable costs but requires upfront capital investment and ongoing maintenance. Hybrid approaches often deliver the best balance for mid-market companies.
Managing influential factors
Strategic partnerships with experienced AI consultancies like TXI can help mid-market companies navigate the complex landscape of AI development while controlling costs. These partnerships provide access to specialized expertise without the expense and challenge of building large internal AI teams. The right partner brings not just technical capabilities but also strategic guidance on use case selection, implementation approach, and organizational change management.
Patrick DiMichele, VP of Strategy at TXI, notes that "we help clients build and scale digital products and platforms that drive meaningful business outcomes." This outcome-oriented approach ensures AI investments deliver sustainable value rather than becoming expensive technological experiments.
Open-source tools and frameworks offer another avenue for cost control, eliminating significant licensing fees. Libraries like TensorFlow, PyTorch, and scikit-learn provide sophisticated AI capabilities without licensing fees. While these tools still require technical expertise to implement effectively, they eliminate one significant cost category from the equation.
Custom AI solutions that address specific business challenges will almost always deliver better ROI than attempting to implement broad, generic AI capabilities. For mid-market companies, these targeted implementations offer opportunities to create meaningful competitive advantages in their specific market niches.
Making AI affordable for small businesses
Affordable AI strategies
Mid-sized and small firms can leverage AI-as-a-Service models from platforms like AWS, Azure, and Microsoft. These reduce upfront costs and provide access to proven AI tools without managing your own infrastructure. The consumption-based pricing models of these platforms align costs more closely with value realization, reducing financial risk for mid-market companies. Rather than making large upfront investments in infrastructure and specialized talent, organizations can scale their AI usage—and costs—as business benefits materialize.
Pre-built AI services for common functions provide immediately accessible capabilities without custom development. While these services may lack the customization of bespoke solutions, they offer production-ready functionality with minimal implementation effort. For organizations new to AI, these services can provide valuable learning experiences and quick wins while building internal capabilities for more sophisticated future projects.
Start small with quick wins—the "think big, start small" approach accelerates learning and reduces waste. For example, a manufacturing company might begin with a focused computer vision application to identify product defects rather than attempting to implement AI across the entire production process at once. This targeted approach delivers faster time-to-value while building organizational confidence and capabilities.
Our work with Four Growers, a farming robotics startup, illustrates this principle. As John Quayle, their Chief of Staff, explains: "With the autonomy of a robotic harvester, an AI function detects the growing cycles of fruits and vegetables. When a tomato plant is ready for harvest, the AI detection allows the robotic arm to harvest it and put it in a container for packaging." This focused application creates immediate value while establishing a foundation for more advanced capabilities.
Cost-Effective AI Tools and Technologies
Prebuilt chatbots, recommendation engines, and generative AI tools offer affordable entry points. For customer service functions, AI-powered chatbots can handle routine inquiries, freeing human agents to address more complex issues. These systems typically offer straightforward configuration options that business users can manage without deep technical expertise. The resulting efficiency gains often deliver rapid ROI while enhancing customer experience through faster response times.
Recommendation engines represent another accessible AI application, analyzing customer behavior patterns to suggest relevant products, services, or content, driving increased engagement and sales. Cloud-based recommendation services provide this functionality with relatively simple integration requirements, making them suitable even for organizations early in their AI journey.
These off-the-shelf solutions can be particularly valuable for mid-market companies with limited in-house AI expertise, providing immediate functionality while building organizational understanding of AI capabilities. They also create opportunities for internal teams to gain hands-on experience with AI technologies in controlled, lower-risk contexts before tackling more complex custom implementations.
Budgeting for AI implementation
Budgeting basics
Effective budgeting involves breaking the cost of AI into phases. Assign line items for development time, technical risk, and support for future upgrades. For mid-market companies, phased budgeting approaches reduce financial risk while supporting more agile, iterative implementation strategies. Rather than committing large resources upfront based on limited information, organizations can make smaller initial investments to validate assumptions and prove value before expanding scope.
Risk-adjusted budgeting recognizes that AI projects face greater uncertainty than conventional IT implementations. By explicitly accounting for technical, data, and organizational risks in budget planning, companies develop more realistic financial expectations and appropriate contingency provisions. This approach helps prevent the resource shortfalls that often derail promising AI initiatives before they can deliver value.
Budget allocations should reflect the full AI project lifecycle, with operational costs typically accounting for 40-60% of total lifecycle expenses. In addition to implementation costs, organizations should plan for ongoing expenses like infrastructure, model maintenance, continuous learning, and user support. These operational costs make them critical components of comprehensive budget planning.
Financial planning and management
Talent development represents a critical long-term investment for sustainable AI capabilities. While external partners can accelerate initial implementation, building internal expertise ensures organizations can effectively maintain and enhance their AI systems over time. This capability development might include formal training programs, mentorship arrangements, or strategic hiring to address specific skill gaps.
Infrastructure management strategies should evolve as AI implementations mature. While cloud services often provide the most flexible starting point, organizations with stable, predictable AI workloads may achieve cost advantages through strategic insourcing of certain components. Hybrid approaches frequently deliver the optimal balance of flexibility and cost efficiency, particularly for mid-market companies with established IT operations.
Governance frameworks become increasingly important as AI deployments scale. Well-designed governance structures ensure appropriate oversight without creating unnecessary bureaucracy that slows innovation. Key governance considerations include data quality standards, model validation processes, ethical use guidelines, and clear responsibility assignments for system performance.
Stay aware of trends like Retrieval-Augmented Generation (RAG) and agentic AI that may alter future costs and capabilities. As our AI Readiness Briefing highlights, these emerging technologies are transforming what's possible with AI, particularly for mid-market companies looking to maximize the impact of limited resources. By monitoring these developments, organizations can identify opportunities to enhance existing AI investments and maintain competitive positioning in rapidly evolving markets.
Conclusion
Understanding the cost of implementing AI isn't just about dollars—it's about building systems that scale with your company, using the right AI models, and avoiding the mistakes that sink so many AI projects.
Final takeaway: AI is a once-in-a-generation opportunity. The companies that succeed will start small, plan wisely, and collaborate with the right partners to create sustainable competitive advantage through thoughtful AI implementation.
About the author
Katie Wolf is the Head of Design at TXI, where she leads the strategic direction and quality of all design initiatives, from systems thinking to human-centered interfaces. With over a decade of experience in design systems, operations, and data visualization, Katie brings a pragmatic lens to solving complex challenges, especially as clients explore emerging technologies like AI. She mentors and manages TXI’s growing design team, helping shape both careers and company-wide design strategy.
