To understand the challenge, look at how Large Language Models work. They are trained on massive datasets, adjusting billions of internal parameters to build a baseline knowledge graph. While an out-of-the-box model is excellent at leveraging its general knowledge base, anything relevant to your specific business must be introduced directly to the model so it can run within your unique business context.
The Reality Check: This is roughly the same dynamic we have historically dealt with in prior SaaS tools. When you purchase a new CRM or marketing campaign tool, you have to configure it to reflect your specific business context and data model. AI is no different.
Experiencing some of this organizational change firsthand, I can say that truly achieving the "AI technical leap" hyped by media outlets is a challenging endeavor. My experience—and likely yours—boils down to three distinct roadblocks:
Many companies are attempting to leverage 1st-party tools, like the various Microsoft Copilot offerings, packaged as a standard SaaS solution. This means the amount of flexibility for the tool to fit your unique business needs is inherently limited by the versioning and roadmaps of the provider.
This locks a business down within certain constraints. It’s the same historical problem we saw with SaaS tools packaged to provide standard CRM capabilities; they ultimately required heavy integration with tools like Salesforce CPQ to truly fit actual business workflows.
For larger organizations, data hygiene is a massive hurdle. The larger the enterprise, the more likely it is that critical operating data is spread across entirely different systems. Most importantly, that data is often represented by different methods in each iteration of what is defined as a "customer".
This is where tools like Microsoft Fabric and Salesforce Data 360 come into play. They are explicitly designed to help standardize, structure, and connect data from various storage silos across an organization.
Organizations frequently stumble when deciding which high-impact AI tools to implement first. Naturally, leadership gravitates toward sales motions like prospecting or automating customer service.
For the sales motion, we quickly hit internal data constraints. We might have valuable customer data, but we lack a unified view across the entire organization due to identity resolution issues. A billing specialist might hold a critical piece of customer knowledge that remains completely invisible to the mechanism driving the automated sales prospecting engine.
Similarly, when AI implementations are wrapped in a tight, restrictive harness within customer service, it creates immense friction. The AI restricts the agent's ability to truly manage the customer experience without escalating to a human. And because organizations have aggressively shrunk their labor force around the promise of AI automation, that inevitable transfer to a person takes longer, leaving customers more frustrated than before.
None of these problems are uniquely native to the AI era; they existed long before the current boom. They are foundational data and operational challenges that require deep attention before any true AI implementation of value can take shape and impact an organization's upside.
Furthermore, we need to shift away from the media's fear narrative around mass labor force reductions. Instead, the story should focus on how the labor force can be liberated to drive positive, high-value change. Perhaps this looks like creating a more community-centric workforce—one that gets out into the market to genuinely deepen a brand's relationship with its community.
