Despite enormous hype around the productivity gains businesses will realize from agentic AI automation, organizations are expressing increasing doubt about their ability to scale the technology past the pilot stage or reap much ROI—at least in the near term.
For every evangelist like Marc Benioff or Andy Jassey touting the benefits of investing in enterprise AI technology there is a new report warning of AI projects being scrapped. A recent Gartner report estimated that more than 40 percent of agentic AI projects will be canceled by the end of 2027. Another report noted more than twice as many companies have abandoned most of their AI initiatives compared to the year before.
But, according to Johann Hallim, CRO, CMO and founder of Greenlight Consulting, successful AI implementations are within reach of every business. It’s simply a matter of organizational focus and a systematic approach.
“When it comes to AI, a lot of organizations dabble,” Hallim tells Automation Today. “There’s too much focus on individual productivity and they’re running their ideas through generative AI on an ad hoc basis. That makes it very hard to get consistent execution. It’s such a powerful tool, but if you’re not focused on outcome-based solutions, it’s hard to get value out of it.”
Without a process-first mindset, he says, AI initiatives will inevitably lose momentum.
“If you’re not looking at orchestrating your overall process and using it to guide how you implement a solution, I’m not sure you can track the value,” Hallim explains. “And without being able to demonstrate the value, those projects tend to fizzle.”
Process First, Technology Second
From Hallim’s perspective, the organizations that are making meaningful progress with AI are those that start with deep process analysis, then gradually layer in automation and agents. He points to work Greenlight is currently doing with a major bank as an example.
The well-known financial institution wanted to start by automating its lending process but understood a successful proof-of-concept could scale to multiple lines of business within the organization. Hallim says a focus on thoroughly defining the lending process, and any subsequent process the client wanted to automate, was the most important step.
“What we’ve done is we’ve mapped out their whole lending process,” he notes. “We are taking parts of that process and automating, then adding AI into it, then adding agents into it. Over time the bottlenecks reveal themselves and you see where improvements need to be.”
By grounding the AI agents’ decision making in the bank’s own historical data and operating procedures, the team is ensuring those decisions reflect institutional knowledge, not generic outputs.
“The thing that’s critical is using your data,” Hallim says. “If you’re not training the models based on your history, anyone can pretend to be you. You need your specific information—your historic performance, your procedures—to feed the models and tell them how to behave.”
This tailored approach has enabled the bank to cut loan cycle times and deliver faster service to customers, while still operating within its own risk frameworks. Hallim believes that’s only possible when organizations recognize their processes as the foundation.
Agentic Orchestration: A Different Discipline
As AI and, more recently, agentic AI have become enmeshed in process automation solutions, it has become evident that they have changed the capabilities of the technology and even the notion of the kinds of projects that are possible and how business problems are attacked. This, he says, is where orchestration becomes paramount.
“Traditional automation projects have a very deterministic mindset,” Hallim says, referring to a time when RPA was the most advanced automation technology. “With agentic AI, the teams are expecting ambiguity. They know the results aren’t going to be exactly what they want and they’re working with agents in an iterative way.”
This requires new skills, particularly in prompt engineering and testing, which has resulted in a far more agile development approach, Hallim explains.
“There’s a lot more collaboration between AI engineers and SMEs,” he notes. “And the idea that you have to retrain models is something that traditional automation teams may not be used to.”
Platform-level orchestration also requires effective governance to produce results. Without consistent guardrails, organizations risk introducing variability that undermines enterprise value.
“If I have ten pricing analysts and one gives AI access to everything and another only gives it access to a subset, you’re not executing consistently,” Hallim notes. “Putting something into an enterprise-governed platform allows you to monitor and enable the technology to perform on a consistent basis.”
‘Not Trying is Not an Option’
Hallim acknowledges some of the difficulties organizations have faced trying to scale successful AI pilots into full rollouts. He also understands that skepticism persists inside many companies. Concerns about poor data quality, unclear objectives, or limited infrastructure can stall projects before they start. But he views these as challenges to be addressed head-on, not reasons to opt out.
“If those issues are legit, they have to be addressed in a concerted fashion,” he says. “Not trying is not an option in this world.”
He also believes that AI has shifted from being a stopgap solution to a driver of long-term value. “People used to think of automation as a short-term bridge until their ERP or CRM got better,” Hallim notes. “What we’re actually seeing is a shift. By going from automation to orchestration, you’re able to provide more value to the organization on a sustained basis without being held hostage waiting for your core systems to improve.”
For practitioners who have already proven the value of RPA, Hallim sees agentic orchestration as the next frontier. “It’s the systematic execution of your proprietary process, leveraging your information, the compute speed of AI and LLMs, and the systematic execution of automation to deliver value as efficiently as possible,” he says.
As enterprises continue to experiment with agents and orchestration platforms, Hallim predicts that agility will become the differentiator.
“The way to approach these projects is not to try and solve everything all at once,” he says. “Build your process and implement technology solutions that make your process better over time.”
That philosophy opens the door to a future where organizations flexibly swap humans, bots, and agents in and out of workflows as business conditions change. For Hallim, that’s the real promise of agentic orchestration: not a one-off project, but an evolving system that allows companies to continuously reshape how work gets done.
