When is AI or automation truly worthwhile?

Problem first, tool second

AI and automation are instruments, not goals in themselves. They only make sense when you can name a repeatable bottleneck: too much manual work, error-prone steps, waiting time or scale you can no longer manage with people alone. You need to be able to explain what needs to change and where the boundaries lie (data, quality, compliance). Without that, "something with AI" is mostly an expensive experiment without an owner.

When to push forward (and in what order)

Clean up process and data: what happens now, step by step? Where does time or confidence leak? Is the data you need complete, current and explainable? Without that foundation, any model or script remains guesswork.
Lock down what should always run the same way (workflow): if you can describe what should happen in which case (conditions and next steps), you can often automate it with fixed rules and integrations: think approvals that route automatically, forms that feed the right system, or data that moves between applications at set moments. This is not machine learning: the system predicts nothing, it executes agreed steps. That approach is often faster to build, cheaper and more predictable than immediately looking for an AI model.
AI where patterns, text or prediction are worth it: for example classification, summaries, recommendations or signals across large volumes. Only deploy this if the business case and quality bar (bias, explainability, error impact) are met. Human in the loop remains necessary where the consequences of an error are too significant to hand over entirely to software.

How Codana approaches this

AI & smart automation sits within the same line as Imagine and Create for us: not a standalone hype, but part of a product and integration strategy you can measure and manage. The bar is measurable improvement (time, quality, cost) and risks you can understand and contain, not a demo that merely impresses.