TL;DR
Rob Hanna of Precision Content says enterprise AI underperforms because organisations treat language like structured data. The real bottleneck is ungoverned documentation, and technical publications teams already have the skills to fix it.
Rob Hanna observes that many enterprise AI initiatives may be losing momentum because organizations continue to treat language like structured data while overlooking the systems that make knowledge reliable. The co-founder and CEO of Precision Content, a technical communications consultancy, says, “Longstanding technical publications teams already possess many of the capabilities needed to establish a scalable content supply chain that supports AI, although those teams aren’t always included in strategic AI discussions.”
Conversations surrounding enterprise AI often focus on increasingly sophisticated models, expanding infrastructure, and new platform capabilities. Hanna observes a different pattern emerging inside organizations. “I’ve seen AI copilots produce inconsistent responses, enterprise search programs struggle to meet expectations, and customer service assistants deliver experiences that leave users wanting greater confidence in the information they receive,” he shares. From his perspective, these outcomes invite a broader discussion about the quality of the knowledge supporting AI instead of focusing exclusively on the technology itself.
That perspective can be viewed alongside an earlier technology cycle. During the chatbot surge Hanna witnessed between 2016 and 2018, expectations expanded rapidly. In 2018, it was projected that 25% of customer service operations would integrate virtual customer assistants by 2020, reflecting widespread confidence in conversational technologies.
“The technology itself showed considerable promise, but many organizations discovered that existing documentation couldn’t consistently support meaningful customer interactions,” Hanna states. He believes today’s enterprise AI landscape echoes many of those earlier lessons because the underlying knowledge environment has often changed far more slowly than the technology designed to use it.
Several studies reinforce this broader pattern. A 2019 paper noted that while hundreds of thousands of task-focused chatbots were developed, successful deployments beyond relatively simple scenarios proved considerably more challenging than many anticipated. A 2021 study examining 103 real-world chatbots likewise identified outdated, incomplete, and poorly maintained knowledge as recurring contributors to implementation difficulties.
According to Hanna, these findings suggest that conversational technologies often depend as much on trusted source material as on advances in software. He believes this history also highlights an important distinction between simply possessing documentation versus possessing usable knowledge.
Hanna notes that many organizations maintain extensive collections of manuals, policies, knowledge bases, training materials, and support resources. Yet those assets may exist across disconnected systems, follow inconsistent standards, or contain overlapping versions of the same information. AI systems drawing from those environments may therefore inherit uncertainty that already existed within the organization’s content ecosystem. “Hallucinations rarely begin inside the model,” Hanna explains. “They often begin much earlier, when nobody has established which information represents the trusted source of truth.”
An example from the earlier chatbot era illustrates this principle. Hanna points to a food brand whose seasonal virtual assistant succeeded by focusing on a narrowly defined subject supported by decades of carefully maintained expertise. Instead of attempting to answer every conceivable question, the experience focused on authoritative guidance for a specific customer need. Hanna reflects, “It’s notably ironic that Butterball’s chatbot should still be held as the gold standard for successful conversational AI, when so many larger organizations have invested and have been unsuccessful. This demonstrates how success begins with carefully governed knowledge that reflects genuine expertise within a clearly defined domain.”
This leads Hanna to another concern that he argues deserves greater executive attention. AI initiatives frequently originate within IT organizations or data science teams whose expertise emphasizes structured datasets, analytics, and computational models. While those capabilities remain essential, he suggests they sometimes encourage organizations to treat written knowledge as another data management exercise. Hanna draws a distinction between the two.
“Data is typically organized into structured fields for computation, while content consists of language, documentation, procedures, policies, and instructions designed to communicate meaning,” he explains. Since large language models learn patterns from written language, he emphasizes that they perform effectively when organizations prepare content in ways that reflect how language is created, governed, and maintained.
Precision Content has built its work around that philosophy by helping organizations turn fragmented documentation into structured, reusable content that serves both people and AI systems. Through structured authoring, metadata, reusable content components, governance frameworks, and component content management strategies, the company aims to help enterprises establish a reliable content supply chain capable of supporting evolving AI initiatives.
Hanna views this as an opportunity to elevate content operations from a publishing function into an important element of enterprise knowledge infrastructure. “Content deserves the same discipline organizations already apply to software development and data management,” he says. “Knowledge should be maintained as infrastructure, so AI can gain a much stronger foundation.”
For Hanna, this conversation also points toward a resource many organizations already possess. Technical publications teams routinely manage version control, structured authoring, taxonomy, metadata, reusable components, review workflows, and content lifecycle management. Those capabilities, he notes, align with what enterprise AI increasingly requires to retrieve reliable information at scale. Instead of searching exclusively for additional technology, Hanna stresses that organizations could begin by recognizing the expertise already available within their own documentation teams.
Ultimately, Hanna encourages leadership teams to broaden the questions guiding their strategies as enterprise AI continues to mature. He says, “We should ask: Which content is authoritative? Who owns enterprise knowledge? Where does organizational truth reside? Can documentation be interpreted consistently by both people and AI? And are technical publications participating in conversations about AI strategy from the beginning?” In Hanna’s view, thoughtful answers to those questions may contribute as much to long-term AI success as continued advances in the models themselves.
Stephan is the sports journalist for the Maple Grove Report.
