SAP data captures a pattern that appears repeatedly in IT implementations, not only those involving artificial intelligence. Companies are quick to adopt solutions that promise fast results with relatively little effort, but they often struggle to turn isolated successes into lasting business value. As a result, the most commonly measured benefits are time saved (41%) and shorter processes (39%), while overall return on investment is measured least often (28%). Time is simply easier to quantify than ROI, which requires companies to clearly attribute costs and benefits to specific activities — something that becomes difficult when processes themselves are poorly structured.
This exposes a fundamental divide between two approaches to using AI in business. The first can be described as fix at output: technology is applied at the end of the process to work around chaotic data or inefficient workflows. It can produce an immediate result, but that result is difficult to scale because the underlying problem remains unresolved. A well-known example of what can happen when rules and processes are not clearly defined was the 2024 Air Canada case. The airline was held responsible after its chatbot provided a customer with incorrect information about a bereavement fare that conflicted with the company’s actual policy.
The second approach is fix at source. Instead of using AI to compensate for problems at the end of a workflow, the company redesigns the process itself. AI then operates within a clearly defined objective, context, boundaries and success criteria. This requires more conceptual work at the beginning, but the result is more predictable, mature and scalable.
Some experts argue that AI transformation should always begin with rebuilding processes at the source. That view is too rigid. Starting with AI at the output stage can be a pragmatic and sometimes necessary first step. Quick automation can free up resources and demonstrate where technology creates immediate value. Those gains can then finance or support a broader redesign of the underlying process.
The real problem highlighted by the SAP findings is therefore not that companies begin with simple automation and quick wins. The problem starts when they stop there. The next stage of AI adoption will show whether today’s focus on time savings is merely a transitional phase as organizations learn how to measure value, or whether many companies will remain stuck with isolated efficiency improvements without ever translating them into measurable returns.
