Many firms fail to see AI ROI because they pick the wrong problems. Successful early adopters focus on internal, employee-facing use cases where success is easily measured against existing performance metrics. This approach lowers data security risks and provides a clear, simple path to calculating return on investment.
Early AI adoption metrics focused on usage, like tokens consumed by engineers ('token maxing'). This incentivized wasteful activity. Mature organizations now measure AI's impact on core business metrics, such as the speed of shipping code from idea to production, which provides a true measure of value.
LLMs can fail to follow critical instructions even when explicitly prompted, making them unsuitable for business decisions with 'hard constraints' like environmental regulations or budget limits. For high-stakes problems, mathematical optimization provides a defensible framework that guarantees constraints are never violated.
An ideal workflow separates probabilistic and deterministic tasks. Use an LLM agent for the creative front-end: helping users identify business constraints, research regulations, and formulate the problem. The agent then calls a dedicated mathematical optimization engine to generate a guaranteed, reliable, and explainable solution.
Merely buying AI tools (10% of budget) and managing execution (20%) is insufficient for ROI. Former Microsoft and Google exec Priyanka Vergadia advises dedicating 70% of the budget to upskilling employees. This focus on education is critical for building a true 'AI habit' and moving beyond experimentation to production.
Without a semantic layer, AI agents querying raw data must re-derive business logic for every question. This is slow, expensive due to high token usage, and prone to errors. A semantic layer encodes this logic, ensuring agents can quickly and accurately retrieve answers that align with agreed-upon company metrics.
Unlike previous tech waves, AI trends have an incredibly short lifecycle. Concepts like 'token maxing' became obsolete in months. This rapid churn makes it extremely challenging for engineers to know which skills to invest in for the long term and when to abandon a technology and move on to the next thing.