In 2025, the Massachusetts Institute of Technology (MIT) released a report that jarred many in the business and investment world. This much-discussed report states that, despite investments of between $30 billion and $40 billion in generative artificial intelligence (AI), 95% of organizations receive “zero return.”
While debate has surrounded the methodology and conclusions, the report itself focuses on the two broad typologies of generative AI applications: horizontal and vertical. “Horizontal” AI includes Chatbots, co-pilots that are meant to make me, as an individual contributor, quicker, better, and faster,” said Doug Croy, Data Advisory Lead for 1898 & Co. “The value driven by horizontal AI is going to reflect on only me and how I do my work.” They are not integrated with work processes, nor are the processes changed or optimized to take advantage of generative AI.
The transformative value and measurable ROI of AI, however, come from the implementation of vertical AI, process-aligned and industry-aware solutions that apply AI to workflows and patterns across an organization. Whereas horizontal AI has broad knowledge that can help you write recipes, vertical AI solutions are focused on specific industries, domains, with knowledge of terminology, relationships, processes, logic and accuracy.
For example, generative AI can identify text in technical drawings, photos and scanned documents, as well as identify objects such as pumps, valves and other equipment found in a power plant. But reading text or identifying an object is not the same as understanding context, the business process and industry-specific language and challenges.
That’s what distinguishes vertical AI; it not only identifies the object with a name or type but also understands the object’s function, behavior and purpose in relation to other objects or assets in a diagram that represents a system, including equipment hierarchies, asset relationships, and safety considerations.
“What you need is a semantic ability to understand what’s on the page, not just to extract the text,” said Chris Wiles, AI Solutions Architect at 1898 & Co. “You have to understand parent-child relationships, the location of equipment, and how valves work together. Extracting some labels isn’t good enough.”
Vertical AI in action
To understand the organizational value that industry-specific vertical AI can deliver, consider the case of a large Midwestern investor-owned utility. The utility sought to migrate data from multiple power plants into the enterprise asset management system (EAM). The EAM serves as a foundation for utility operations by tracking work orders, maintenance schedules, parts inventory, equipment condition, and compliance requirements. With the right data, the system can provide instant access to essential information that plant operators need to make decisions, helping to avoid unplanned outages and reduce operations and maintenance (O&M) spending.
But the utility faced a daunting challenge. Because the plants were built in different eras with varying documentation standards and naming conventions, data consistency was nonexistent, and quality varied wildly — all stored in a mix of old databases, paper documents, and the minds of experienced personnel. The utility expected it would take three years to manually extract and organize data into a format EAM could use.
To provide the structured, relational data the utility needed to get the most out of EAM, the company worked with 1898 & Co. to deploy a custom vertical AI solution that understands power plant systems and engineering.
The process combined decades of power plant experience from data scientists and engineers, resulting in a transformative outcome. Not only was the data extraction completed in three months, rather than the estimated three years, but the equipment location accuracy also jumped from 50% to close to 100%.
The vertical AI solution also unlocked other beneficial applications. A semantic search tool now makes all documentation searchable in real time using plain language. That means maintenance technicians can search for equipment and quickly retrieve maintenance recommendations, manufacturer manuals, and work order history. For the utility, reduced downtime and more efficient maintenance planning translate to tens of millions of dollars in annual operational savings.
“When a plant engineer has a question about a work order on an asset,” Croy said, “users can access P&ID, onelines, PM recommendations, and anything else they might find pertinent to their task immediately at hand for a given asset.”
Similar vertical AI applications are emerging across utility operations. For example, one 1898 & Co. client is exploring a model to coordinate substation outages. By analyzing work orders, system topology, reliability standards, and project schedules, vertical AI can optimize the efficiency of work during planned shutdowns. In another example, geographic information system (GIS) teams are utilizing vertical AI to automate mapping tasks that typically require a significant amount of time to complete manually, thereby freeing up staff for more strategic work.
Do you really need pristine data before you begin?
Vertical AI solutions need access to your data and systems. Lacking enterprise data management practices can make this a challenge to access, understand data and trust its quality. AI can be part of the solution. AI can help identify quality issues with the context of the data and apply algorithmic approaches to substitute or mitigate insufficient data quality. A better outcome is to use the logic of Gen AI and adjacent data sources to address data quality. For instance, AI is being used to identify and address data gaps in a database of distribution pole data containing age, type, attachments, etc. An AI solution is able to extract sufficient data from a variety of mixed, multi-model sources to address data quality.
AI presents an opportunity for improved data quality and, in many cases, scenarios where improved data quality might have been technically and cost-prohibitive.
Your critics are your friends
Devoting time and resources to support vertical AI is a key step in securing meaningful ROI. But successfully implementing AI is not purely about the technology. People matter, and navigating skepticism, resistance to change, and legitimate concerns is as important as choosing the optimal AI strategy.
Counterintuitively, the most effective change management strategy is the one that many organizations avoid: engaging the most outspoken critics of change. “When you’re working on a vertical solution, sometimes leaning into that person who says ‘I’m not going to use AI’ and letting them be your first person to provide feedback can be powerful,” Croy noted. “Let them look under the hood and be the first to provide feedback on what’s wrong — and grow from that. When that 40-year veteran in the plant says ‘go use the AI tool,’ that’s a pretty powerful message to everyone else.”
There is no doubt that Horizontal AI is increasing the productivity of individuals, to the obvious benefit of both them and their respective organizations. But ROI for a utility comes from deploying vertical AI that understands where there is significant value in optimizing a utility-focused process or removing complexity with a new capability.
Contact 1898 & Co. to learn how your organization can get more ROI out of AI.
