The numbers are contradictory, and that is the point. Nearly every enterprise has an AI strategy. Most have a budget. Many have tools already in production. And almost none of them can point to the financial return with any confidence.
BCG’s 2026 AI in Logistics Executive Survey, which polled 30 leading global logistics players, found that 97% of executives rank AI as a strategic priority, 70% have an AI strategy, and 67% have a dedicated AI budget. Only 13% say AI is delivering measurable financial impact. The tools are going in. The results are not coming out.
This is not a logistics problem. It is an enterprise-wide problem that spans industries, geographies, and company sizes. And the reason it persists is not that AI does not work. It is that most companies are measuring against the wrong timeline.
The two timelines companies are actually running on
Peter Bendor-Samuel, CEO of Everest Group, wrote this week that enterprises are pursuing two distinct AI strategies simultaneously, and the confusion between them is causing most of the frustration.
The first is evolutionary. Take software development as an example. Everest Group’s research indicates organizations can achieve productivity improvements approaching 30% by applying AI to today’s software development lifecycle. But those gains require changing team structures, development processes, management practices, and more. Simply adding AI coding assistants will not deliver the full benefit. This evolution is likely to unfold over 18 months to two years, with further gains accumulating over a three to five year horizon.
The second journey is transformative. It asks a different question entirely: what would this business look like if we built it around AI from the ground up? Answering that question requires integrating customer service with sales, finance, fulfillment, marketing, and other adjacent functions. AI becomes the platform that enables a fundamentally different operating model rather than simply automating existing tasks. This second journey is significantly more ambitious and will likely take five years or more to mature.
Most companies are conflating these two timelines. They announce AI transformation programs with 18-month targets, measure them against five-year outcomes, and conclude that AI is not working. The mismatch between expectations and reality is not a failure of the technology. It is a failure of planning.
The confusion is understandable. Vendors sell AI as a quick win. Consulting firms frame transformation as a 24-month engagement. Board members read headlines about companies that deployed AI and saw immediate results. None of this matches the operational reality of restructuring how a large, complex organization actually works on a daily basis. The gap between the pitch and the process is where most AI initiatives go to die — not because the technology fails, but because the timeline was never realistic to begin with.
Why the ROI is real but slow
The economics underlying AI deployment are compelling on paper. BCG estimates that the global logistics market, which generates $5.5 trillion in annual revenues, could see end-to-end AI transformation boost EBITDA margins by roughly 5 percentage points. About half of that gain would come from the top line, with leading players achieving revenue lifts of up to 10% through AI-enabled dynamic pricing, cross-selling, and cost reductions. The rest would come from cost-side improvements of 3 to 4 percentage points through procurement optimization, automation of customer service, and operational efficiency gains.
Those numbers are real. But they assume end-to-end transformation, which means every process in the value chain has been rethought around AI capabilities. That does not happen in 18 months. It does not happen in two years. For most enterprises, it is a five-year project at minimum. The compounding effect is real — each AI-enabled process improvement makes the next one easier and more valuable — but the compounding requires a foundation that most companies are still building.
Flexport offers an early glimpse of what this compounding looks like in practice. The company has broken ocean freight forwarding down into more than 100 discrete operational steps, a level of detail that enables it to deploy AI agents at the task level rather than at the process level. The results compound. Ocean operations are now over 70% automated, cost-to-serve is on track to fall by around 30% within a year, and customs audit error rates run at one-tenth of the industry standard.
But Flexport spent years building the data infrastructure and process decomposition that made this possible. The AI agents are the visible part. The invisible part is the thousands of hours of process mapping, data cleaning, and organizational restructuring that came before them. When companies look at Flexport’s results and ask why they are not seeing similar returns, the answer is usually that they have not done the foundational work yet. The AI is not the bottleneck. The readiness is.
This is the pattern BCG found across its survey. Companies that jumped straight to model deployment without fixing their data foundations are stuck in what researchers call the “pilot purgatory” phase. They have successful proofs of concept that never scale to production. The models work on clean test data. They fail on the messy, inconsistent, incomplete data that actually runs the business every day. Fixing that data problem is not glamorous work, but it is prerequisite work, and it takes significant time and investment.
The patience problem
The organizations that benefit most from AI will not necessarily be those making the biggest announcements today. Bendor-Samuel argues they will be the organizations that set realistic expectations, invest in organizational change alongside technology, and pursue both evolutionary improvements and longer-term reinvention simultaneously.
This is hard for public companies. Quarterly earnings cycles reward visible action, not patient infrastructure building. CEOs who announce AI transformation programs expect to show results within their tenure, which typically runs three to five years. The transformative timeline lines up with that window, but only if the work starts on day one and continues without interruption.
The evolutionary timeline is more forgiving. Companies that focus on specific, measurable improvements in existing processes — code review automation, customer service response times, demand forecasting accuracy — can show results within 12 to 18 months. These gains are real but modest. They do not transform the business. They make it incrementally better.
The risk is that companies treat evolutionary gains as evidence that transformation is happening, when in fact they are just the first step. Getting 30% productivity improvement in software development is meaningful. It is not the same as rebuilding your product development process around AI-native workflows. The first is an upgrade. The second is a reinvention. Most companies are doing the first and calling it the second.
The manufacturing sector illustrates this dynamic clearly. AIoT World reported this week that the next phase of manufacturing AI depends on network infrastructure — the connectivity layer that allows edge devices, sensors, and cloud systems to share data in real time. Companies that skipped this step and jumped straight to deploying AI models on top of fragmented data are now hitting a wall. The models work in demos. They fail in production because the data they need is siloed, inconsistent, or simply not available at the speed the application requires.
This is the pattern across industries. The AI model is the easy part. The data infrastructure, the process redesign, the change management, the measurement framework — these are the hard parts, and they take years, not months. Companies that understand this distinction are the ones making progress. Companies that do not are the ones writing off AI as overhyped because their chatbot deployment did not transform the business in six months.
What the data actually says
BCG’s survey data tells a clear story. The gap between AI investment and AI return is not closing as fast as executives expected. But the gap is closing. Companies that have been working on AI for three or more years are showing measurably better results than those in their first year. The curve is real. It just does not bend as fast as the vendor pitch decks suggest.
The 13% of logistics companies reporting measurable financial impact share a few characteristics. They started with specific, high-value use cases rather than broad transformation programs. They invested in data infrastructure before deploying models. They treated AI as an operational change, not a technology project. And they measured progress in quarters, not months.
For companies still in their first year of AI deployment, the message is straightforward: the ROI is coming, but it is coming on a timeline that looks more like enterprise software adoption than like a startup product launch. Set expectations accordingly. Measure what you can measure. And resist the temptation to declare failure because the numbers are not what the vendor promised in the sales deck.
There is also a talent dimension that most transformation timelines ignore. Building an AI-capable organization requires hiring or developing people who understand both the technology and the business domain. This is not a staffing problem you solve with a job posting. It takes 12 to 18 months to recruit, hire, and onboard specialized AI talent. It takes another 12 months for those people to build the institutional knowledge needed to apply AI effectively to your specific business problems. By the time your AI team is fully operational, two to three years have passed. This is normal. It is not a sign that something is going wrong.
The practical takeaway is that companies should stop measuring AI transformation against startup timelines and start measuring it against enterprise software adoption curves. The first enterprise resource planning implementations took five to seven years. The first cloud migrations took three to five. AI transformation will follow a similar pattern, accelerated somewhat by better tooling and more mature ecosystems, but not fundamentally different in kind. The organizations that accept this reality and plan for it will outperform the ones that keep chasing the promise of quick returns.
The enterprises that get this right will have a significant competitive advantage in five years. The ones that do not will have spent a lot of money learning the same lesson everyone else already knew: real transformation takes time, and the patience to see it through is itself a strategic capability.