Out of the PoC Quagmire: How to Make AI in Business Stick

Ashling Unfiltered is our quick-read POV series that cuts through buzzwords to separate signal from noise in AI and automation. Each post breaks a trending claim into what’s said, what we’ve seen in the field, and what leaders should do next.

In this edition, we put SSON's 2025 State of Intelligent Automation under the microscope. We pull the headline claim, check it against field reality, and end with one move you can make this quarter. Let's dive in.

The hype cycle and interest levels for AI in business, and more specifically Agentic AI, are at peak levels. This level of exposure brings pressure for organizations to use artificial intelligence and be seen to be “doing something with AI” which results in rushed pilots, unclear outcomes and are often seen as failures. Organizations that acknowledge the power of AI while approaching it with the same vision, structure & governance as their existing IA programs will have the highest odds of getting out of the PoC quagmire we’re seeing in the market. 

This is a well written piece and from our experience in the field it’s representative of what we’re seeing with clients. The areas of focus for GBS and key points around strategic alignment, governance and value are exactly what our most successful clients are doing.

There are two points asserted we’re yet to see proven: 

1. “Agentic AI is highly scalable”: Seems true in theory but when it comes to cost, consistency and maintenance effort we’re just not seeing the ability to scale that we already have with more traditional IA tools. 
2. Citizen Development Benefits: Accessibility of these AI tools means it’s never been easier for Citizen Developers to automate part of their day-to-day work. The challenges with this approach remain: how do you track these improvements as hard benefits/value back to the organization; and, how do you govern and maintain these automations over time?
   

1. Several use cases we’ve worked on with clients leveraging AI or Agentic AI have had their business use cases become non-viable because of AI consumption costs. Complex, multi-technology or multi-agent systems require multiple LLM calls and thousands of tokens per case run which can run costs up quickly for high volume use cases.
2. In the same vein we’ve found the interconnected nature of multi-technology and multi-agent systems combined with non-deterministic outcomes makes break/fix and/or process updates time consuming once in production. The actual dev work is relatively simple but the regression testing and prompt iteration is extensive and costly.
3. When it comes to Citizen Development outside of the contact center, almost no clients have been able to capture tangible benefits provided by the productivity aids typically built in a Citizen Development scenario, whether the technology is AI or traditional IA. Interestingly this seems to point to a future where Process & Task mining become essential to organizations trying to quantify the productivity uplift achieved through Citizen Development.

If you’re just dipping your toe into Agentic Automation, the first step is educating yourself on the value it can unlock in your organization. Ask yourself: “What can we do now that we couldn’t do before?” and use that as a north star, along with strategic objective alignment as a guide on where to begin. A simple, but fruitful, approach we’ve seen with clients is going back to their IA pipelines for the past few years and looking at high value processes that were disqualified due to complexity or unstructured inputs and assessing those candidates for Agentic AI. 

If you’re looking to learn more about AI and Agentic processes that have been deployed to production and the realities of what it takes to get them there—bring Ashling in for an education and knowledge sharing session. We also offer hands-on workshops and rapid PoCs to help build your understanding and uncover valuable, but considered, opportunities to begin your AI & Agentic journey.