Why most AI projects still fail to show real business returns

AI value often fails to extend beyond pilot projects

Most large enterprises still measure their AI success through pilots and proofs of concept instead of actual profit and loss impact. The excitement around early results fades when it’s time to scale these projects into daily business operations. This is because many companies have not built the foundations to support sustained AI use at scale. Complex data environments, legacy systems, and disconnected pipelines make moving from pilot to production slow, costly, and frustrating.

What usually happens is that teams run small experiments that look great in controlled conditions. The models work; the demos impress. But when integrated into real-world systems that carry decades of technical debt, the projects stall. They fail not because the idea was bad but because the enterprise lacked the architecture to make AI a permanent part of its decision-making process.

For leaders, the message is simple. Stop taking “proof of concept” as proof of value. Until AI becomes part of your operating and financial cycle, you haven’t yet won. Move beyond pilots by focusing equal attention on the infrastructure, data flow, and governance that support AI at scale.

According to MIT Sloan research, between 70% and 90% of AI projects never make it past the pilot stage. That failure rate highlights how common the gap remains between technical potential and real business performance.

Dr. Pallab Deb, Chief Data and AI Officer at Carrier, has highlighted this exact issue within large industrial organizations. He notes that most enterprises underestimate how deeply systemic and cross-functional the AI value chain truly is. His experience shows that the real work begins after the first demo, when leadership turns hype into measurable, recurring gains.

Data infrastructure fragmentation and vendor sprawl significantly undermine scalability

AI depends on data, but most organizations have data systems that don’t talk to each other. Multiple legacy ERPs, incompatible CRMs, and outdated manufacturing systems still dominate in large enterprises. They prevent clean data flow and create confusion about which system owns which data source. This fragmentation slows down execution and stops otherwise capable AI models from being used consistently across the company.

Another layer of complexity comes from vendors flooding the market with tools and “copilots.” Each provider offers different AI assistants, often bundled into existing contracts at little or no extra cost. The result seems positive at first, more options, less expense, but quickly turns into chaos. Companies end up with overlapping tools performing similar tasks, each requiring training, compliance checks, and separate maintenance. This is the same pattern many executives saw during the early SaaS boom, when redundant software multiplied costs instead of reducing them.

To scale AI effectively, companies must standardize their data architecture and choose their platforms with precision. Executives need a harmonized, secure, and governed data system before they push for advanced AI integration. This ensures models perform consistently, data quality is reliable, and investments aren’t wasted across duplicate systems.

The broader trend is clear. Integration and coherence matter more than the number of tools in use. Leaders who reduce complexity and align their data ecosystems will not only accelerate AI deployment but also see far greater financial returns.

Organizational misalignment is the main cause of AI failures

Many companies believe better models will fix their AI problems. In reality, most failures happen because the organization itself isn’t designed to use AI effectively. The issue is managerial and structural. Major projects collapse when leaders underestimate how much change management, process redesign, and incentive alignment influence success.

The ratio often quoted by implementation experts is telling: 70% of the effort in AI transformation is about people and process, 20% about tools, and only 10% about models. Yet, most budgets flow in the exact opposite direction. Too much time is spent refining algorithms or benchmarking performance, while the business fundamentals that allow those models to deliver impact remain underdeveloped.

Teams often lack clear ownership and accountability. Change agents are not empowered, and cross-functional collaboration becomes slow or inconsistent. Meanwhile, contractor-heavy teams, sometimes forming over 80% of delivery capacity, erode institutional knowledge and momentum. Leadership turnover further resets priorities before transformation can take root. These hidden inefficiencies are where AI value evaporates.

Executives should pay closer attention to organizational foundations. Success depends on stable leadership, process integration, and measured follow-through far more than on acquiring cutting-edge models. MIT Sloan research confirms this, showing that 70–90% of AI initiatives never scale due to organizational gaps.. High conviction from leadership, supported by disciplined operating models and consistent incentives, is the only way to close that gap.

Proofs of concept often hide the structural weaknesses that block AI at scale

Proofs of Concept, or POCs, have become a routine step in AI adoption, but they often give leaders a false sense of progress. A POC usually runs on static, controlled data and ignores most of the difficult work needed to bring a system into live production. There’s little or no focus on data quality, integration with existing systems, or compliance with corporate governance. This narrow scope makes early results look strong while concealing the operational friction that appears later.

Many of these projects also lack role clarity and business accountability. Some have no named business owner or funding plan for long-term operations. Without clear ownership, it’s almost impossible to measure real ROI or justify continued investment. Governance then appears late in the process, often just before deployment, introducing delays and friction at the worst possible time.

Dr. Pallab Deb, Chief Data and AI Officer at Carrier, addresses this issue by requiring every AI project to have a defined business owner, value hypothesis, and ROI measurement plan before development begins. This rule forces teams to think about real impact from the start. It aligns AI work with financial and operational outcomes.

For C-suite leaders, the message is direct: treat POCs as validation steps. A POC only matters if it produces data pipelines that can scale, governance frameworks that hold up under scrutiny, and measurable outcomes that survive real business conditions. If those pieces aren’t in place early, even the best models won’t reach sustained performance or measurable ROI.

Early and integrated governance is critical for AI success

Most companies approach governance as an afterthought. They build an AI solution first and invite security, compliance, and risk teams to evaluate it at the end. By that point, the necessary changes often require rework, additional funding, or full redesign. The better approach is to move governance to the beginning of the planning process and make it continuous rather than reactive.

When governance operates early, it helps align technical, legal, and business goals before significant resources are committed. Carrier’s AI Review Board, led by Dr. Pallab Deb, Chief Data and AI Officer, applies this principle by reviewing project proposals upfront. Its function is not to block progress but to streamline it, combining governance, risk, compliance, performance, and commercial value reviews in a single early-stage process. This approach reduces delays later and ensures that every AI initiative begins with clear risk and reward visibility.

Executives should view governance not just as oversight but as an enabler. Integrated controls improve confidence among stakeholders, speed up approvals, and reduce hidden costs. By uniting governance and performance metrics, leadership can create consistency across teams and maintain accountability as projects evolve.

This system is particularly vital for regulated industries, where compliance can be as critical to ROI as technical execution. A structured, front-loaded review keeps AI trustworthy, repeatable, and aligned with business outcomes. Once governance becomes part of the initial design, it stops being a bottleneck and becomes a core component of operational scaling.

Leadership decisions on value linkage and operating models determine AI ROI

Technology alone doesn’t create business value, leadership decisions do. The biggest difference between successful AI enterprises and stalled ones is that their leaders draw a direct, measurable line between AI programs and financial outcomes. Without that financial linkage, AI remains a cost center instead of an earnings driver.

Dr. Pallab Deb at Carrier treats this linkage as non-negotiable. His team translates its corporate AI target of $200 million in projected 2026 value into $0.23 of earnings per share (EPS). That conversion forces every business unit to understand not just that AI matters, but precisely how it influences financial performance. This clarity drives accountability and helps prioritize projects that clearly support earnings growth.

Leadership also needs to define how AI work moves through the organization. Carrier operates through a hub-and-spoke model, with a central AI Center of Excellence supported by cross-functional business pods that handle discovery, delivery, and operations. Every project flows through a transparent intake and triage process with checkpoints for sponsorship, priority, funding, and measured value. This avoids the “science project” problem, where pilots multiply without linking back to the company’s strategic and financial goals.

For CEOs and CFOs, the lesson is straightforward. Make P&L impact the guiding metric, enforce structured intake and decision flows, and align operating models around outcomes rather than activities. A consistent leadership stance on these points prevents fragmentation, accelerates execution, and ensures that AI investments produce tangible, auditable returns.

Clear Build–Partner–Buy decisions are essential to preserve AI value

AI programs lose efficiency when organizations try to do everything themselves or rely entirely on external partners. The highest returns come from knowing exactly which capabilities to develop internally, which to accelerate through trusted partners, and which to buy as pre‑built solutions. This clarity prevents overlap and ensures that every investment contributes directly to measurable outcomes.

Dr. Pallab Deb, Chief Data and AI Officer at Carrier, applies this principle rigorously. His team builds internal capabilities in areas that define the company’s competitive advantage, such as pricing algorithms, forecasting engines, and product analytics, because these directly enhance intellectual property and create long-term differentiation. Where scale or speed matters more than ownership, Carrier partners with system integrators that can deliver proven accelerators. For widely available processes like accounts payable or receivable automation, the company purchases standardized tools instead of using internal resources to recreate them.

This disciplined separation between build, partner, and buy decisions maximizes ROI and reduces fragmentation. Enterprises that fail to define these boundaries tend to accumulate redundant systems that cost more to maintain and deliver less value per dollar spent.

For executives, this structure provides a simple rule: use internal resources for areas tied to strategic differentiation, leverage partners for time‑to‑market advantage, and buy only when functionality is truly commoditized. The clarity of ownership slows technical debt accumulation and strengthens the overall return from AI portfolios.

Change management and incentive alignment drive real AI adoption

Even the most advanced AI systems fail when people ignore them. Technology adoption only sticks when leaders make it part of performance expectations and link usage to measurable results. Without that alignment, employees treat new tools as optional, and productivity gains remain theoretical.

At Carrier, Dr. Pallab Deb’s team embeds adoption directly into business goals. When the company introduced AI assistants to tens of thousands of knowledge workers, progress wasn’t measured by availability or access, it was tracked through business‑unit productivity outcomes. Each leader was responsible for adoption targets tied to tangible improvements, such as faster analysis, reduced cycle time, or better customer outcomes. The data was reviewed monthly, and progress influenced broader evaluations and investment priorities.

Training, role redesign, and resource planning were funded upfront, rather than treated as afterthoughts. This ensured that teams had both the capability and motivation to use the tools effectively. When adoption metrics appear alongside traditional business dashboards, executives can see not just who deployed AI, but who turned it into actual value creation.

For C‑suite leaders, the takeaway is practical. Budget for organizational change as rigorously as technology itself, and make adoption measurable across all levels. A strategy that pairs incentives with accountability transforms AI from a technical experiment into a dependable contributor to ROI.

AI delivers real P&L impact through four distinct value pools

Executives often discuss AI’s potential at a conceptual level, but measurable financial results come from four specific value streams: productivity, cost savings, growth, and product differentiation. Each area plays a different role in translating technical capability into quantifiable business performance.

The first pool, productivity, is the most common starting point but also the hardest to measure effectively. Many companies record “notional” productivity gains, such as time saved or faster task completion, that have no material impact unless tied directly to changes in staffing levels, throughput, or customer service goals. For productivity to translate to financial impact, leadership must define measurable outcomes and adjust operating plans accordingly.

The second pool, non‑people cost savings, provides clearer proof of P&L value. At Carrier, an AI-driven policy enforcement tool saved approximately $25 million by identifying and preventing out‑of‑policy warranty reimbursements. This example demonstrates traceable financial value, backed by data and verified through standard accounting procedures.

The third pool, growth, represents AI initiatives that directly influence top-line revenue or margins. Examples include demand forecasting at the SKU level, AI‑assisted energy optimization, and generative systems for tailored sales recommendations. These capabilities are harder to build but can reshape revenue models once validated.

Finally, product differentiation anchors the most defensible form of ROI. Embedding AI directly into products, such as predictive features, adaptive services, or intelligent interfaces, creates unique advantages competitors cannot easily match with off‑the‑shelf tools.

For leaders, the key is prioritization. Each value pool requires different investment levels, validation methods, and ROI horizons. Executives who sequentially move from efficiency to differentiation build a balanced portfolio that drives sustainable and defensible financial returns.

A disciplined operating model ensures sustainable AI ROI

Delivering consistent AI impact requires more than innovation, it requires operational precision. At Carrier, the AI organization functions under a clearly defined framework that combines centralized coordination, agile execution, and financial discipline. This model ensures that initiatives move from conception to measurable value without drifting into unstructured experimentation.

The company runs a central AI Center of Excellence (COE) that sets standards, manages governance, and sustains core platforms. Around it operate cross‑functional pods responsible for discovery, build, and run phases. The structure ensures that team velocity remains high while maintaining accountability at every stage of production. Projects enter through a shared intake process, go through defined gating reviews, and are prioritized based on potential ROI, resource readiness, and strategic value.

Financial discipline is built into the process. A Value Office, working with FP&A, enforces standardized definitions for cost savings versus cost avoidance and uses risk‑adjusted multipliers and Net Present Value (NPV) models to calculate long-term benefits. These metrics feed directly into board-level dashboards, anchoring AI results to EBIT and earnings per share.

Carrier’s technology strategy complements this structure by employing a flexible multi‑model “model garden” that includes OpenAI, Anthropic, and Gemini models. Rather than deeply customizing every model, the team relies on Retrieval‑Augmented Generation (RAG) to create agility and scalability. Data architecture follows a producer-consumer alignment, supporting transparency, cataloging, and selective real-time pipelines where business-critical.

For executives, this disciplined model demonstrates how to make AI operational rather than experimental. Clear accountability, measurable financial reporting, agile delivery, and flexible architecture together guarantee that AI initiatives stay aligned with enterprise goals and deliver year‑over‑year returns.

A practical playbook turns AI from experiment into recurring financial impact

Momentum in AI transformation comes from structure, not spontaneity. A clear execution framework ensures that every new initiative begins with the end in mind, realized, measurable ROI. Carrier’s internal playbook outlines specific steps to minimize waste, keep accountability visible, and ensure AI acts as a business instrument rather than a lab exercise.

The first requirement is ROI at intake. No AI project starts at Carrier without a clear business owner, a measurable baseline, and a defined value hypothesis. This approach filters out experiments that cannot be financially evaluated. Each initiative must come with a quantifiable impact statement before resources are committed.

The second pillar is adoption accountability. AI tools are linked to specific business‑unit productivity, cost‑savings, or growth targets. Progress is measured monthly, and business leaders are held directly responsible. This ensures that adoption is not decorative but directly connected to operational metrics.

Third, change investments are funded upfront. This includes budgeting for process redesign, user training, and talent transitions. Funding change from the beginning protects projects from collapse during post‑POC implementation.

The playbook also calls for early governance alignment through formal review mechanisms; data product consolidation to reduce redundancy; and standardized contractor management to protect speed and continuity. Projects are planned in 12‑month horizons with flexible architecture, allowing quick adjustment to new models or technologies without rework.

Finally, results are reported with the discipline of finance, using EBIT and EPS impacts, distinguishing realized from projected value, and presenting growth and differentiation outcomes to the board every quarter.

For executives, this approach demonstrates how to operationalize discipline in AI strategy. The playbook’s simplicity comes from its precision: define ownership early, fund change visibly, and measure performance with financial accuracy. When these principles are applied consistently, AI becomes predictable, repeatable, and profitable.

Treating AI as an operational discipline ensures sustainable business impact

The companies that see measurable returns from AI are those that manage it with the same rigor applied to other capital‑intensive functions. The difference between activity and impact lies in treating AI as a long‑term operational capability.

Many enterprises still treat AI spending as discretionary. They measure success by the number of pilots or tools tested, not by direct profit‑and‑loss outcomes. To shift this, leaders must set AI performance metrics alongside traditional financial measures, linking projects to EBIT, EPS, and margin improvement. This shifts AI from a cost item into a contributor to enterprise value.

At Carrier, Dr. Pallab Deb, Chief Data and AI Officer, has proven that discipline produces results. His organization measures both realized and risk‑adjusted outcomes, limits discretionary investments, and uses standardized governance to ensure alignment between the technical, operational, and financial sides of the business. By doing so, AI operates under the same oversight as other key business functions.

For executives, the message is direct: success with AI is not determined by model accuracy or technology choice but by structure, continuity, and leadership discipline. Consistent reporting, cross‑team collaboration, and agile financial management are what allow AI to translate from innovation to earnings.

Legacy enterprises that embrace this view will see AI shift from a peripheral interest to a strategic driver of growth and efficiency. Treating AI as an operational discipline closes the gap between potential and P&L, and makes it part of the core operating rhythm of the company.

Concluding thoughts

AI will not transform a business simply because it’s advanced. It transforms a business when leaders treat it as a core operational and financial discipline. The strongest companies are already moving beyond pilots and experiments. They connect AI investments directly to earnings, productivity, and real, defensible growth.

For most enterprises, the barrier isn’t technology, it’s focus. The right data foundations, clear ownership, disciplined governance, and a financial reporting mindset are what separate activity from impact. Once those fundamentals are in place, AI stops being a side initiative and becomes part of the company’s operating system.

Executives who lead with this mindset will see AI value compound over time. The more disciplined the execution, the faster the gains become visible on the balance sheet. Treat AI as a managed capability, not a curiosity, and it will evolve into one of the most reliable engines for long-term enterprise performance.