Enterprises are pouring billions into artificial intelligence initiatives across every sector. Yet McKinsey reports 80% of companies see no material contribution to earnings from their GenAI investments.
This paradox is huge. And crosses every industry.
Healthcare systems pilot tools that never reach critical workflows. Retail executives deploy recommendation engines that barely move conversion metrics. Every vendor showcase is a chatbot dressed up as an “AI agent.”
The pattern repeats: massive investment, minimal transformation.
In our recent webinar, polling of technology leaders reveals the complexity of challenges ahead. While 38% cite “knowing where to start” is still their biggest agentic AI hurdle, another 25% struggle with “complex systems and integrations,” and 13% identify “lack of trust in AI agents” as the primary barrier.
Meanwhile, organizational readiness data shows 49% remain at ad-hoc and isolated AI implementation levels, with only 8% reaching governed and scalable approaches. Outliers to the norm are few and far between.
These numbers expose the core problem: organizations are investing heavily in AI technology while lacking the operational foundation to scale effectively.
The copilot trap
Bernhard Schaffrik, principal analyst at Forrester Consulting, sees enterprises making the same mistake repeatedly. They deploy copilots broadly, expecting transformation to follow adoption.
“So just throwing the technology, general-purpose LLM access, to everybody within a company or any type of organization doesn’t make you a leader in AI adoption,” Schaffrik observed. Returns from widespread deployments are “usually quite frustrating,” leading executives to question what went wrong.
The polling data confirms this challenge. Organizations struggle not just with basic AI confusion but with integration complexity and trust at operational scale. Customer-facing agents remain rare because companies adopt agentic AI “cautiously and incrementally for various reasons.”
This creates a scaling paradox: enterprises need transformational outcomes but approach implementation with fragmented, risk-averse strategies that keep them removed from where they’ll offer the biggest impact: Core, mission-critical processes. And for that, you need a new approach.
The orchestration breakthrough
The solution emerges when organizations treat agentic AI as an orchestration challenge rather than a limitation of the technology. Real-world business processes are not binary. Oftentimes they’ll have very defined, rule-based processes that also need dynamic, context based knowledge work.
This makes blending deterministic orchestration with dynamic orchestration essential to gain real returns from agentic AI.
Lily Wang, CIO for Wholesale Client Onboarding and Group Financial Crime at Barclays, discovered this while tackling financial crime detection. “The tools that we have, traditional tools that we have, are quite blunt, right? It’s literally just, you know, using those tools to say, you know, A plus B, is it likely? Is it very likely? I don’t know, but let’s look at it. So you have to throw a human into the equation.” Simple rule-based systems create a cascade of false positives that overwhelm human analysts.
Wang’s breakthrough came from recognizing that complex financial behaviors require explanation, not just flagging. “When you talk about special discipline in catching bad actors in financial crime…the complex customer relationship that we have is hugely complex. Trying to get a developer to create something that they can explain back to the people who are looking at the information in front of them. It’s really tough.” Business Process Model and Notation’s (BPMN) ability to construct “human-readable processes, and explain quite complex pieces of a jigsaw puzzle” solved what traditional development couldn’t achieve. And now that same power is applied to agentic AI, by designing trust into the agent itself using BPMN.
For complex customer onboarding involving thousands of regulatory permutations, agents can now analyze multiple paths and propose optimized solutions. “There is an optimization objective. The client clearly told us they want to get business running in three weeks’ time,” Wang explained. The system can propose: “If you are this type of business, and choose these products… we can onboard you in that time.”
This approach maintains human oversight for high-stakes decisions while enabling agents to handle complex knowledge work within defined parameters. Ultimately, this improves the lives for employees and the customers they serve.
The readiness problem
Schaffrik’s team developed an “AI quotient” measuring organizational readiness across understanding, soft skills, hard skills, and ethics awareness. The assessment reveals why scaling fails even when pilots succeed.
“Adoption doesn’t work without human subject matter expertise,” Schaffrik noted. Pilots work because experts contribute throughout projects. But without subject matter experts, scaling stalls because human knowledge remains inaccessible.
Wang confirmed this organizational challenge: “The most challenging part is actually trying to help people to understand, organizationally, we need to be ready when you do have this type of solution. The organizational shift is actually really, really difficult.”
Beyond organizational readiness lies the domain knowledge problem. “We don’t use just the general LLM and just, you know, kind of apply it to the banking industry. We actually need to anchor that with the knowledge that is banking specific,” Wang explained. This knowledge often exists “very much in people’s heads.”
“Doesn’t matter how smart machines are today, it can’t read your mind,” Wang emphasized. Making tacit organizational knowledge accessible to AI systems while maintaining human judgment becomes critical for scaling success.
The automation advantage
Schaffrik observed that organizations “already leading with automation” in terms of “scaling, governance, federation” are also “leading with AI adoption.” They’ve developed process orchestration as an “inner discipline or a capability” that transfers to agentic implementation.
This suggests successful AI scaling requires coordinating people, processes, and technology rather than just adding AI tools to existing operations. The 8% achieving enterprise scale aren’t deploying better technology alone. They’re leveraging orchestration capabilities to integrate intelligent agents into operational workflows and completely rethink their approach.
The investment-outcome paradox persists because most enterprises approach agentic AI as technology acquisition rather than operational transformation. The polling data confirms this reality: with 91% of organizations still in ad-hoc or piloting phases, most lack the governance and scalability frameworks needed for enterprise success.
The organizations breaking through this paradox follow a different AI readiness roadmap entirely. They recognize that AI success isn’t driven by models alone. It’s driven by maturity across five foundational pillars that must develop simultaneously.
While struggling organizations chase “emotional ad-hoc” purchases driven by fear of missing out, successful enterprises systematically address vision clarity, people readiness, architectural flexibility, delivery capability, and measurement rigor. The contrast is stark: ad hoc organizations measure success through anecdotes with no KPIs, while adaptive enterprises monitor business-level outcomes in real time to adjust strategy. One group builds cool prototypes they can’t operationalize; the other creates processes that “learn from themselves” and optimize continuously. This maturation depends on five key pillars developing simultaneously: vision, people, architecture, delivery, and measurement.
Success requires reimagining how work gets done when intelligent agents join teams, creating systems where human expertise amplifies artificial intelligence rather than competing with it. Organizations that master this orchestration move beyond the billion-dollar paradox toward measurable transformation.
Learn more
Watch the full discussion with Lily Wang, CIO at Barclays and Bernhard Schaffrik, Principal Analyst at Forrester to explore how enterprises are bridging the AI investment-outcome gap through orchestrated approaches that blend human expertise with artificial intelligence.
