Every board is asking for AI, every engineering team wants to build it, and every vendor is selling it. Most of these projects still do not deliver meaningful business value, because the teams focus on the technology and skip the valuation.
My position has not changed: Technology is easy. Valuation is hard.
For AI, this has never been truer. Using a large language model or a machine-learning API now costs almost nothing to start. A junior developer can prototype an impressive feature in a weekend. Working out whether it solves a high-value problem - or whether the running costs scale - is the hard part.
The trap of a solution looking for a problem
Most AI projects fail because they start at the wrong end. Companies see what modern AI can do and try to force it into existing processes. The result is isolated proofs of concept that shine in a demo and stumble in production.
Skip the honest valuation step and you end up with:
- Inference and operating costs that outrun the value they create.
- Small improvements to workflows that did not need changing.
- More data-privacy and security risk, with no matching business return.
A first-principles approach to Purposeful AI
The way out is first principles. Before writing a line of code or signing a vendor contract, step back and evaluate the problem space.
- Quantify the inefficiency: what exact problem are we solving? If it disappeared tomorrow, how much revenue is generated or saved?
- Determine the AI necessity: does this need a probabilistic AI model, or would a deterministic algorithm or better architecture do the job?
- Assess total cost of ownership: look past the API price to data-pipeline engineering, model-drift monitoring, and human-in-the-loop validation.
A Purposeful-AI strategy means having the courage to say no to good technology when the numbers do not work. Focus on the valuation, and the AI you do deploy has a better chance of paying for itself.