I currently invest roughly a quarter of my professional time working on applying AI in B2B markets. Given today’s hype cycle, that hardly makes me unique.
Fear of missing out (FOMO) has pushed AI to the top of nearly every board agenda. The current AI FOMO is primarily due to Baby Boomer executives being slow to adopt e-commerce and not wanting to miss another big disruption. The billions of silly monies being invested just add more anxiety.
The fear is accelerated when an executive has a competitor who goes all in on AI. How would you compete against your archrival if their seamless and friction-free customer experience earns them a small gross margin premium and their SG&A expenses are 20% less than yours? They can increase their net profit while starting a major price war with you. You would then have a knife in a gun fight.
My first AI project was in 1970 for the U.S. Army, using Monte Carlo simulation and LISP arrays, and I’m now engaged in a significant orchestration project for a large industrial distributor. One lesson has remained consistent throughout the decades: When a field is advancing this rapidly, no one understands more than a slice of it.
But the slice I do understand tells me commercial businesses are paying too much attention to the battle among LLM providers OpenAI, Anthropic, and Meta; the AI rivalry between the U.S. and China; and artificial general intelligence (AGI). It’s an expensive and dangerous distraction.
The Economist has recently published multiple articles highlighting the contrast between U.S. and Chinese approaches to AI. China emphasizes engineering talent and rapid deployment, while the U.S. is characterized by legal caution and ambitious aims toward AGI. The magazine pointed out that China is now being run by engineers while the U.S. is run by lawyers. The implication was that in a period of disruptive change, engineers will get through it faster and better than lawyers. Their statement is certainly provocative, but if true, it will have a significant impact on China’s AI adoption speed over the U.S. in the medium term.
The AGI debate is largely irrelevant for commercial executives. Even if AGI emerges in the next several years, its immediate impact will be concentrated among the largest technology companies that control the infrastructure, data, and ecosystems required to exploit it. What is true: Businesses protect their shareholders and will not give up power if it adversely impacts them, beneficial corporation greenwashing notwithstanding.
Meanwhile, distributors have real problems to solve today.
WHAT ACTUALLY MATTERS TO COMMERCIAL BUSINESSES
Strip away the noise, and the AI arms race, and the commercial opportunity becomes clear. AI, properly deployed, can:
- Reduce SG&A expenses by double digits
- Eliminate friction across internal handoffs
- Improve customer responsiveness and consistency
- Increase organizational resilience in volatile markets
These outcomes do not depend on AGI. They depend on redesigning how work gets done. This is where most companies go wrong.
WHY 95% OF BUSINESS AI PROJECTS FAIL
Every day, new AI tools, point solutions, agents, and apps are announced. At the annual Distributor Strategy Group (DSG) AI conference earlier this year, 41 software providers presented AI point solutions aimed at distributor challenges.
With this much activity, it’s easy to believe that AI success is simply a matter of picking the right tool. The data tells a very different story.
MIT recently published a study on business AI projects, and 95% have failed to meet original expectations. There are many insights into what the failures have in common, but in my view, they missed the largest one.
Almost all these projects were treated as IT projects.
IT plays a critical role in AI projects, but IT can’t own AI transformation. While IT professionals are typically fascinated with AI, tuned into new tools, and eager to learn more, their efforts only address a fraction of the problem.
Embedding AI into a commercial business isn’t just about technology; it is about using the new AI tools to redesign work processes and make organizational changes to take friction out of the system.
AI is not a tool that automates existing tasks in isolation. It changes how decisions are made, who makes them, and how information flows across the organization.
The typical pattern looks like this:
- The board pressures management to “have an AI strategy.”
- The CEO authorizes a pilot.
- IT selects a use case and builds a demo.
- Early results look promising.
- Complexity explodes.
- Progress stalls.
- The project quietly dies.
The failure point is not technology. It is complexity.
COMPLEXITY THEORY AND THE WALL EVERY AI PROJECT HITS
Any business is a complex combination of processes designed to optimize certain characteristics, often in conflict with each other. Consider sales vs. credit, or customer fill rates and inventory turnover, plus dozens more. Most organizations already struggle to manage this complexity.
Now, along comes AI — fun to play with, easy to approach, and it can do some cool stuff.
That’s why there are many successes applying AI to a single process. For instance, we shared an application at the National Association of Wholesaler-Distributors (NAW) Executive Summit in 2023, where Infor, the ERP folks, deployed a full process for Grosfillex. The punchline in our presentation was that improved profits in the first 90 days paid for the entire deployment.
NAW has created petabytes of research, economic data, and intellectual property over decades of work. This year, NAW, working with ProfitOptics, created an AI tool called Nucleus so their members can find what they seek both quickly and richly.
What kills most initiatives is what complexity theorists call Ashby’s Law of Requisite Variety: “To manage a system effectively, the controlling system must be as complex as the system being managed. When the number of variables or relationships exceeds managerial capacity, control fails.”
Applying AI to one workflow can work. Applying it to dozens of interconnected workflows without coordination quickly becomes unmanageable.
This is why single-use-case success stories are common — and why enterprise-wide impact is rare.
WORKFLOW REDESIGN IS THE REAL OPPORTUNITY
The full commercial impacts from AI do not occur until AI tools have redesigned every workflow process in the firm. That is why prioritizing new use cases starts with the weakest workflows.
When CRM company Salesforce laid off 4,000 of their 9,000-member customer service department to improve their customer experience with AI, the CEO, Marc Benioff, was trashed-talked by the social media mob. However, I am confident that he understood his customer friction points and designed new processes and solutions made possible by AI that did not exist before. What the critics missed is that he was driven to improve the customer experience, not to cut heads.
A strategic AI deployment is having the accountable executive review every workflow process within their portfolio and applying the AI tools to streamline their own processes, taking friction out while improving responsiveness and resiliency.
THE MISSING LAYER: GOVERNANCE AND ORCHESTRATION
Hope for AGI and the seductive impact of single-point solutions are distractions to the strategic deployment of AI in a business. Something has been missing so far from most AI projects. No one had a solution to manage the complexity challenge.
The missing piece from many of the projects in the MIT study was governance. It is correct to prioritize use cases; it is best to move quickly, constantly testing, training, and iterating. But how can speed be maintained when dozens of important decisions are being modified on the fly by independent functions within an organization?
That’s where a governance layer capable of absorbing complexity while maintaining momentum comes in. A governance layer provides guardrails, oversight, and the ability to maintain visibility of the entire process from the CEO on down. One of the tenets of complexity theory is that only variation can absorb variation.
This governance layer can be built or bought. My view is that 100 times out of 100, it is better to buy it than build it. Those who choose to build it themselves typically call it agent operations or AgentOps.
But what you do not know costs money. There is a software category called Integrated Platforms as a Service (iPaaS). This category is the only solution today to manage complexity with a mature and robust governance layer. Google lists 21 of these providers, and Gartner has an IPAAS Magic Quadrant.
Gartner created the iPaaS term in 2011 and defined the category as “a cloud-based platform that offers application and data integration capabilities delivered as a service, enabling the standardized, scalable, and streamlined integration of diverse business systems across both cloud and on-premises environments.”
Big words, but the dream was to seamlessly integrate disparate systems and data sources in real time so friction was removed, and decisions could be made with complete information and fewer dropped balls.
What has changed is that AI — particularly no-code and low-code capabilities powered by LLMs and small language models (SLMs) — has finally made this vision practical and scalable.
iPaaS platforms provide:
- Centralized orchestration of workflows
- Built-in governance and monitoring
- Integration across cloud and on-premise systems
- The ability to scale AI-enabled processes safely
AI has supercharged all the iPaaS providers by dramatically lowering the cost of integration and customization.
FROM PILOTS TO ENTERPRISE IMPACT
Reading the MIT study, my takeaway is that the projects shared a common set of assumptions that worked when deploying software. The flaw was assuming that this was a software project.
Any AI deployment is an organizational change management process that reshapes workflows and decision-making. Decision policies that were once defined and constrained by existing work practices are re-engineered, with many data-dependent decisions shifting from people to AI systems.
A typical successful deployment flows like this:
- Senior leadership commits to executive ownership of AI transformation.
- IT partners with an iPaaS provider to set up the private cloud, establish governance and integration infrastructure. The IT team, working with their iPaaS partner, decides which LLM or SLM to use. Even global multibillion-dollar firms can often find that an SLM is more than sufficient for managing mega petabytes of their own data.
- Organizational change planning occurs in parallel with technical setup. The plans are resilient but flexible.
- A simple, visible use case is deployed to build executive understanding and confidence.
- Functional leaders redesign workflows within their domains using AI tools directly.
- Governance ensures alignment, consistency, and control as scale increases.
This approach allows speed without chaos.
AI lowers the costs for a manager to create incremental changes in a solution that can be tested and revised in real time. A major change in a customer, or a supplier, or integrating a new acquisition can be addressed on the fly. It is not an IT project. It is led by senior line and staff executives with strong responsive support from IT. When an entire organization has an opportunity to leverage the tools, the level of complexity requires a sophisticated governance model, so individual contributors are not held back by the slowest members in their organization or a confused IT department.
THE BOTTOM LINE FOR EXECUTIVES
Three conclusions matter most:
1. AI deployment in a commercial business is not an IT-led project.
It is an executive management-led transformation, with IT in a critical support and enablement role.
Why this matters:
AI changes decision rights, workflows, and accountability. Only senior leaders can redesign how work gets done across functions. When AI is owned by IT, it inevitably optimizes tools instead of outcomes and stalls at scale.
2. Real value comes from redesigning processes from the ground up – not from automating existing processes.
Modern AI-enabled processes eliminate unnecessary handoffs between functions and reduce friction across internal operations, customers, and suppliers. The result is faster execution, clearer accountability, and a more seamless value chain. Peter Drucker, the famous management guru, said that there is nothing more pointless than making something more efficient that you should not be doing at all.
That’s why you don’t just bolt AI onto existing workflows. You start by questioning every process from scratch and then rebuild it, assuming AI is available. That requires reprioritizing processes, shifting decision authority, and restructuring teams.
Why this matters:
AI changes who decides, not just how fast tasks run. Salesforce, Domino’s Pizza, and firms highlighted by McKinsey all redesigned workflows and roles, not just systems.
3. Complexity requires governance, and governance must scale as fast as AI does.
Organizations committing to AI at this level should rely on a robust integration platform (iPaaS) to connect systems and workflows. Gartner would be my starting point to choose one.
Why this matters:
As AI proliferates across functions, independent decisions multiply faster than organizations can manage them. A robust governance layer provides the guardrails, visibility, and coordination required to maintain speed without losing control.
What this all means in practice is simple, but not easy.
If you are a CEO or president, your first AI investment should not be a model, a chatbot, or a pilot owned by IT. It should be your own time spent understanding where decisions stall, where handoffs break down, and where organizational friction silently taxes margins and customer experience.
AI will not create a competitive advantage by making existing processes marginally faster. It creates an advantage when it allows you to redesign how work gets done, who makes decisions, and how information moves across the enterprise.
Firms that treat AI as a collection of tools will generate impressive demos and disappointing results. Firms that treat AI as an operating model change will quietly reset cost structures, responsiveness, and competitive dynamics.
The window for this advantage will not remain open indefinitely. As AI-enabled workflows become standard, today’s early movers will become tomorrow’s marketplace incumbents, and the opportunity will shift again.
