Companies spend billions on AI but see no returns: How Amazon cut routine time by 92% and what holds others back

Companies worldwide invest billions annually in automation and AI, yet the expected returns often don't materialize. A Bain & Company study of nearly a thousand global companies revealed that while 37% aimed for 11–20% cost reductions, almost 40% of those who measured outcomes achieved only 0–10% savings. Despite this, 90% of these same companies are increasing their AI agent budgets again. How can this be, and why do some, like Amazon, achieve exponential efficiency gains?

AI investments are not a magic bullet. If business processes are inherently inefficient, an AI agent will only accelerate and scale that inefficiency, making it even more costly. The problem is not the technology itself, but the company's readiness to implement it. Without a systemic approach, without rethinking processes and data management, an AI agent will not deliver the expected results, but merely create an illusion of progress.

The Problem: Why AI Agents Fail to Deliver Expectations

The primary reason AI investments don't yield desired returns lies not in the technology itself. The issue is the gap between expectations and reality, and the underestimation of organizational changes required for successful implementation.

According to the study, only 7% of companies are running fully autonomous AI agents in production today. The majority (38%) require human approval, and another 32% operate with guardrails, where a human intervenes if the agent encounters an uncertain situation. This means that business cases for AI agents are often built on the assumption of full autonomy, whereas in practice, a significant portion of decisions still require human involvement. This creates a financial gap: the finance department approves one set of numbers, while the organization operates with another.

Funding "Air": The Next Wave of AI on Unmet Hopes

Another risk is the approach to funding new AI initiatives. 44% of companies plan to fund generative AI and AI agent investments with savings from previous automation programs. At first glance, this seems like a disciplined approach. However, the problem is that previous programs often fell short of their planned savings targets. Thus, the next wave of investment is funded by "air"—assumed but unrealized savings.

This creates a vicious cycle where risks accumulate rather than being managed. Companies that do not verify the actual returns from previous automations before investing in new ones significantly increase their financial risks.

Data: The Wall That Cannot Be Bypassed

The biggest barrier to AI progress remains data access and integration, cited by 41% of respondents. This ranks above compliance concerns, budget, skills gaps, or executive buy-in. Despite decades of investment in data modernization, companies still cannot reliably access their own data, which is a primary reason for underperforming AI programs.

Interestingly, companies that met their targets cite data as a bigger barrier (44% vs. 40% for those who missed), because they are deploying solutions at scale. Underperformers, in contrast, cite more organizational obstacles: insufficient budget, lack of a Center of Excellence, and competing priorities. These are not technology problems, but signals that AI has not been given the necessary mandate and executive attention.

How to Turn the Tide: Lessons from Leaders

Companies that successfully implement AI and achieve real returns make several key decisions:

1. Pay down your "workflow debt" before deploying AI. Automating an inefficient process will only entrench its flaws. Before implementing AI, it is necessary to rethink the process from scratch, eliminate unnecessary steps, and optimize it. AI should not perpetuate business process "debts" but rather build on a clean slate.
2. Validate the investment case and name a governance owner. CFOs should audit actual returns from prior automation programs, rather than relying on projections. Furthermore, it is essential to determine in advance who is accountable if an AI agent makes an erroneous decision. Accountability cannot be improvised.
3. Use AI to solve the data problem; don't wait for it to be solved. Imperfect data infrastructure is the least valid reason to defer AI investment. It's better to start in areas where data is already constrained and accessible, and use AI itself to improve data flows within the organization. For example, Amazon's Finance Technology team implemented a solution for tracking VAT regulatory updates worldwide. What used to take 26 minutes per update now takes 2 minutes, and 80% of AI-generated summaries are accepted by experts without modification. This is not a "moonshot"; it's a targeted automation of a specific workflow where the data was already available.
4. Redesign the operating model, not just the process. Deploying AI agents without changing how people work with them guarantees underperformance against business goals. In an AI agent-led model, employees become orchestrators and supervisors, making decisions that agents cannot. This requires investment in role redesign, new ways of working, and change management.
5. Measure outcomes at the enterprise level, not the program level. Programs will always optimize for what they are designed to measure, typically costs and hours saved. For the company, what matters more is whether AI produces better decisions, faster responses, and improved customer outcomes. If these metrics are not on the CEO's dashboard, programs will continue to efficiently deliver "the wrong things," and the value gap will persist.

The turning point for most companies is not finding the best AI technology, but the realization of leadership's personal responsibility to create the organizational conditions for AI success. The window of opportunity to make this decision is narrowing faster than many executive teams realize.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager