Artificial intelligence has moved rapidly from experimentation to enterprise boardrooms, with organizations investing heavily in models, platforms, and AI agents. Yet, despite successful demonstrations and proofs of concept, many AI initiatives struggle to translate into measurable business value when deployed at scale.
In the latest episode of the Analytics Insight Podcast, host Priya Dialani speaks with Ranjan Kumar, Founder and CEO of DecisionX, about the challenges enterprises face while scaling AI. The conversation explores causal AI, explainability, organizational context, AI-driven decision-making, performance measurement, and the infrastructure required to move AI from pilots to operational impact. Here are the excerpts:
According to Ranjan Kumar, the enterprise AI challenge is not simply about the availability of powerful models. Over the past two decades, companies have invested heavily in digitizing information through systems of record, CRMs, ERPs, and other enterprise platforms.
The arrival of large language models added another layer of intelligence to this information. However, organizations initially attempted to connect enterprise data directly to LLMs and use them for business applications.
While these systems performed well in demonstrations, problems emerged when companies tried to apply them in complex organizational environments. Ranjan points to four major challenges: lack of organizational context, the inability to answer “why” questions through causal reasoning, rising costs, and security concerns.
LLMs can summarize documents or answer straightforward questions, but enterprise decision-making often requires understanding why something happened and what could happen if a particular action is taken.
For enterprises, getting the correct answer is only one part of the decision-making process. Ranjan explains that organizations also need to understand how an AI system arrived at its answer.
This becomes particularly important when AI is used in industries such as financial services, healthcare, and pharmaceuticals, where incorrect decisions can create compliance, operational, and financial risks.
Enterprise AI systems therefore need to be more deterministic and reproducible. If organizations ask the same question multiple times, they need confidence that the system can provide consistent reasoning rather than unpredictable outputs.
Ranjan also highlights the importance of traceability. AI-generated answers should show the information and files used to conclude. For regulated industries, sources, citations, and an audit trail can help users validate AI-generated decisions and build trust in the system.
Ranjan describes causal AI as an approach focused on understanding cause-and-effect relationships within an organization.
A conventional data-driven system can answer a question such as how much a company sold in a particular month. However, enterprise decision-making becomes more complex when leaders need to understand why sales declined or determine how a particular action could affect future performance.
For example, if a company wants to increase sales by 20 percent over six months, an AI system needs to understand the factors influencing sales. These could include marketing expenditure, the number and capacity of sales employees, and other business variables.
Causal AI attempts to understand these relationships and their potential implications. This can support use cases such as scenario planning, forecasting, stress testing, and evaluating “what-if” situations.
Ranjan argues that AI cannot remain a technology experiment. Enterprises need to connect AI investments with measurable business outcomes.
He broadly identifies three forms of value.
The first is top-line growth, where AI can help unlock additional revenue. For example, AI can help sales teams prioritize leads and opportunities, potentially improving conversion rates.
The second is bottom-line efficiency, which can involve optimizing marketing expenditure, reducing operational costs, saving employee time, and improving productivity.
The third is strategic value. As organizations become increasingly AI-native, the way employees use information and make decisions can become a source of differentiation.
Ranjan emphasizes that these outcomes need to translate into measurable KPIs across organizational functions. A marketing leader, for example, should understand what efficiency improvement AI is expected to generate, while a sales leader should have a defined top-line objective.
One challenge with enterprise AI is attempting to use the technology for everything at once. Ranjan argues that organizations should instead define specific objectives and bounded use cases.
Once a use case is clearly defined, enterprises can determine the context required, establish the relevant organizational ontology, improve trust and verification, and develop evaluation mechanisms for consistency and explainability.
The level of AI investment should also depend on the use case. Strategy teams involved in scenario planning, forecasting, and stress testing may need deeper causal AI capabilities, while other teams may primarily need systems that can answer straightforward, information-based questions.
Ranjan identifies enterprise infrastructure as an important part of the AI scaling challenge. He explains that enterprise AI involves multiple layers, including data, models, and workflows.
Enterprise data can include structured and unstructured information such as call recordings, PDFs, plant manuals, standard operating procedures, and numerical tables.
A common mistake, according to Ranjan, is connecting organizational data directly to large language models and then building automated agents on top of those models.
The conversation highlights the need to consider how data is prepared, how organizational context is represented, how models are used, and how workflows are automated rather than treating the model itself as the complete AI infrastructure.
The discussion points toward a shift from AI experimentation to AI systems designed specifically around business decisions and measurable outcomes.
For enterprises, the focus is increasingly shifting from simply deploying AI models to building systems that understand organizational context, explain their reasoning, trace the information behind decisions, and identify cause-and-effect relationships.
As AI becomes more deeply integrated into enterprise operations, the ability to connect technology investments with tangible business outcomes will remain central to determining how organizations adopt and scale AI.
The full discussion can be heard on the Analytics Insight Podcast.