Generative AI strategy should begin with the data and decisions that make a business distinctive, according to Ikigai Labs president Kamal Ahluwalia. In an interview with AI News, he argued that the excitement around large language models can distract enterprises from the structured, tabular information in systems such as SAP and Salesforce. That data drives planning and forecasting but requires different techniques from models trained primarily on internet text.
Ikigai Labs focuses on large graphical models developed from MIT research. Its platform is designed to turn sparse, siloed enterprise data into predictive and actionable insights, particularly across four time-series areas: sales, products, employees and capital or cash. Understanding how these streams interact during a product launch, geographic expansion or supply-chain change can help a company make decisions tied directly to its operations.
Structured Data Offers a Different Path to Value
Large language models excel at unstructured material such as text, but much of the information used to run an enterprise is stored in rows, columns and sequences over time. Forecasting demand or workforce needs requires models that recognise relationships in those records. Ikigai’s argument is that organisations should not force every business problem into a conversational interface when the underlying question is numerical and temporal.
This distinction also affects differentiation. Generic chatbot, writing and summarisation features are becoming widely available. If competitors can access the same foundation model and reproduce the same use case, the result may improve productivity without creating a durable advantage. A company’s own operational history, constraints and plans are harder to copy. Applying AI to that specific information can reveal optimisation opportunities that reflect the business rather than the general internet.
Enterprise Requirements Differ From Consumer AI
Ahluwalia described consumer-oriented language models as products of internet-scale data and expensive computing. Their cost can be spread across a huge user base, and imperfect answers may sometimes be tolerated. Enterprises operate under different conditions. Hallucinations can affect financial plans or regulated decisions, confidential data must be protected and most organisations cannot build a complete foundation-model stack themselves.
Businesses still want the fluid experience associated with generative AI, but they need accuracy, predictable costs and control over private information. Ikigai presents large graphical models as a way to train on an organisation’s own data, operate on CPUs and provide greater transparency. This approach does not make deployment trivial, but it targets the constraints that often stop experiments from reaching production.
Imperfect Data Should Not Become an Excuse to Wait
Nearly every organisation works with incomplete, sparse or inconsistent information. A common response is to postpone AI until a major clean-up programme is complete. Ahluwalia warned that this can become an endless cycle, especially with time-series data. By the time several quarters have been spent cleaning historical records, new information has arrived and may contain a fresh set of problems.
The alternative is to choose methods designed to learn from limited data and account for change. That does not mean ignoring quality. Teams should understand missing values, document assumptions and evaluate how uncertainty affects an output. It means making progress with the information available rather than treating perfection as a prerequisite. A bounded forecasting project can also reveal which data improvements would create the greatest practical benefit.
Human Expertise and Culture Determine Adoption
In regulated or consequential settings, Ahluwalia emphasised expert oversight, validation and reinforcement. An expert in the loop can challenge an output, supply context and prevent the model from making decisions in isolation. This partnership is particularly important when historical patterns do not reflect a new event or when the organisation must explain why a recommendation was made.
Successful adoption also requires executive commitment and continuous education. Leaders need to understand what the technology can do, support experimentation and communicate how work will change. Ahluwalia’s view was not that AI would simply eliminate jobs, but that roles across the organisation would evolve. Teams need time to learn new workflows and must see that executives are accountable for the transition.
The interview offers a useful corrective to AI strategies built around fashionable tools. Technology should be selected after the organisation identifies a valuable, specific decision. For a manufacturer reconsidering its geography or a company designing more sustainable products, the important variables may include logistics, labour, materials, vendors and costs. Modelling those relationships can be more strategically valuable than adding another generic content generator.
Ikigai Labs is proposing a narrower but potentially more defensible form of enterprise generative AI: models grounded in structured company data, aligned with time-sensitive planning and supervised by people who understand the domain. Whether its approach fits a particular organisation depends on the problem and data. The broader lesson is sound: competitive advantage comes from applying AI to what is unique about the business, not from reproducing the same capabilities available to everyone else.
That principle gives executives a useful filter for investment. A project should be able to identify the decision it improves, the proprietary information it uses and the metric by which success will be judged. If none of those elements is clear, the organisation may be adopting an interface rather than building a capability. Specificity makes both value and risk easier to evaluate.
