Skip to main content

2025 Gartner® Report: The 3 Business Cases of Generative AI Value | Download the Complementary Report!

Webinar: Scaling GenAI for Maximum Impact – 3 Takeaways

Jan Overney
Post By Jan Overney April 1, 2025

Enterprise AI. The term is everywhere – a buzzword tossed around in corporate boardrooms and media headlines. By now, we've all seen the demos, the promises of transformation, and the rush to implement AI solutions. But behind the hype lurks a stark reality: to this day, scaling AI from pilot project to enterprise-wide deployment that delivers actual value is a challenge few have mastered. 

In our recent webinar, "Scaling AI for Maximum Impact," we brought together Gartner analyst Darin Stewart and Squirro CEO and co-founder, Dorian Selz, to cut through the hype and provide clear, actionable guidance into navigating this challenging terrain. They shared precious insights, real-world examples, and practical strategies for tackling the complexities of scaling enterprise AI. 

Here, we break down some of the most salient points covered during the webinar (which, if you missed it, is available on-demand here.)

The Harsh Reality of Scaling Enterprise AI 

To state the obvious, GenAI has been developing at breakneck speed. As Dorian Selz put it, before 2022, AI was a “minority sport.” Then came ChatGPT, and by 2023, AI  went mainstream. In 2024 it was “pilots galore,” with companies experimenting with chatbots, co-pilots, and generative text tools. And here we are in 2025, and AI's honeymoon phase is over. As companies seek to transition from pilot to production, they are waking up to the harsh reality that the real work of scaling AI within the enterprise is only just beginning.

Gartner’s Darin Stewart noted that this transition isn't just a technical upgrade; it's a strategic imperative: “Scaling generative AI to the enterprise is critical because it's no longer optional. If you don't have an adequate generative AI deployment, you're viewed as lagging and risking becoming irrelevant.” And it’s about overcoming another widespread phenomenon: AI sprawl, the emergence of pockets of often conflicting AI systems across organizations. Instead, he said: “real value comes from a coherent strategy and deployment plan. Getting to that point is a major struggle.” 

David Hannibal, CPO and Head of Corporate Development at Squirro, who moderated the webinar, echoed Darin Stewart’s experiences seeing too much AI sprawl going around organizations and not enough ROI. Staying stuck in pilot mode means your AI investments are failing to deliver their full potential. You're spending time, energy, and resources – including precious GPU power – on “creating new models that are similar to others,” and running the “same SharePoint data against all these new models, wasting time, headcount, and GPU energy,” he pointed out. 

As the era of AI experimentation draws to a close, unlocking the transformative potential of AI will require businesses to move beyond pilots and embrace a strategic, production-focused approach. This means creating robust, secure, and integrated AI systems that are aligned with overall business goals and deliver measurable ROI.

Data Governance and Security in Age of Enterprise AI 

One of the most significant takeaways from the discussion was that AI acts as a powerful magnifier, revealing and exacerbating any existing weaknesses in your data management practices. “Many companies say they don't need governance anymore, that the AI will take care of it. But the AI will expose all the issues you haven't addressed,” said Stewart. Far from fading away, the increasing reliance on generative AI is making good content governance, appropriate security measures, and solid permissions and access controls more essential than ever. 

Permission-enablement – the ability to ensure that sensitive information is only seen by those authorized to view it – is a particularly pernicious challenge in this context, explained Dorian Selz: “It's solved in traditional enterprise search, but it becomes complicated in vectorized spaces. But without proper permission enablement, you won't be able to scale enterprise-level generative AI.” The shift from document-based to vector-based data representation brought about by AI introduces new complexities in enforcing role-specific access controls.

Why is this so critical? Businesses are dealing with increasingly sensitive information, especially in regulated industries such as financial services and manufacturing, for example, in the form of personally identifiable information (PII) data. As a result, the ability to handle data privacy, PII data, and security have become a prerequisite for being able to scale their AI experiments to production. Prioritizing data governance and security provides an essential foundation for AI initiatives to deliver real business value while mitigating potential risks.

The Rise of GraphRAG and Agentic Intelligence

Traditional retrieval augmented generation (RAG) has been a game-changer in improving the accuracy and relevance of AI responses. By grounding language models in external data, RAG reduces hallucinations and provides more contextually rich answers. Dorian Selz outlined one of the core challenges behind its remaining limitations: the probabilistic nature of its two enabling components: search, on the one hand, and the foundational large language model, on the other. Relying solely on probabilistic methods is hardly the most effective way to eliminate inaccuracies. 

This is where knowledge graphs come to the rescue. “Integrating graph at ingestion, retrieval, and result resolution provides a reference set to guard and improve quality,” he explained. GraphRAG, which incorporates knowledge graphs into the RAG process, adds a layer of deterministic precision. Their integration enables AI systems to understand the nuances of data, connect disparate pieces of information, and provide more accurate, reliable, and traceable GenAI outputs

Gartner’s Darin Stewart emphasized, “Incorporating a knowledge graph addresses many issues. It helps get around generic training, focuses queries, and aligns with business processes. On the back end, it validates responses and provides traceability. In regulated environments, that's essential.” 

This increased reliability is critical in view of the growing excitement around Agentic AI. As Darin Stewart pointed out, people want “systems to be proactive and semi-autonomous rather than coming up with detailed multi-stage prompts. They want an agent to take the problem, break it down, recruit resources, and present the answer.” This represents a shift from reactive AI systems that respond to specific prompts to more proactive and intelligent agents that can take initiative and solve complex problems on their own.

Finally, amid the hype around surrounding agentic AI, both speakers cautioned against oversimplifying the concept and rebranding traditional workflow automation as agentic. True agentic AI involves systems capable of understanding and reasoning about problems rather than merely following pre-defined steps. The real power, Dorian Selz added, lies in dealing with 'enterprise data ambiguity,' where AI can navigate complex and messy data landscapes to drive business processes effectively.

Key Takeaways

By moving towards more proactive and intelligent systems, businesses can advance GenAI beyond RAG and unlock new levels of efficiency, innovation, and value. The transition from experimenting with AI to harnessing its full potential as part of a coherent enterprise-wide AI strategy is not a simple one, but as Dorian Selz stated, “it's possible to succeed.”

If you're ready to take on the AI scaling challenge in your organization, we encourage you to dive deeper into the insights shared during our webinar, "Scaling AI for Maximum Impact.” Additionally, we've created a collection of resources that provide further guidance and practical steps to help you navigate the complexities of scaling AI effectively. 

Whether you're just starting your AI journey or looking to optimize existing deployments, we're here to help you every step of the way. Get in touch today to find out more about how the Squirro Enterprise GenAI Platform can power your AI ambitions.

 

Discover More from Squirro

Check out the latest of the Squirro Blog for everything on AI for business

Webinar: Scaling GenAI for Maximum Impact – 3 Takeaways
Webinar: Scaling GenAI for Maximum Impact – 3 Takeaways
Future-Proofing Enterprise AI with LLM Agnostic RAG
Future-Proofing Enterprise AI with LLM Agnostic RAG
Leveraging GenAI for Effective Compliance Risk Management
Leveraging GenAI for Effective Compliance Risk Management