Let me tell you the BIGGEST strategic mistake most GenAI products make: they are chasing use cases where precision is non-negotiable.

70% accuracy in building a financial model is not “better than nothing”, it is actually useless!

If an LLM generates a model with “hallucinations” (a fancy word for “random formulas” or “invented numbers”), I cannot predict where the errors lie. That forces me to re-check every cell.

Fix one error, another appears. At that point, what am I accelerating?! I am simply shifting my human effort from building the model to auditing it, while my modelling muscles slowly atrophy.

The machine is good at scaffolding work, so maybe I save 10% on structure. OK. But let us be honest: that is a far cry from the 50%+ productivity gains people like to claim.

This obsession with quantitative precision use cases reveals a misunderstanding of where GenAI creates real leverage.

GenAI is a probabilistic machine. It thrives in domains where probability is good enough: brainstorming options, reframing narratives, summarizing messy threads, exposing blind spots, stuff like that.

When you apply it to domains that require deterministic accuracy, you get a more-or-less shiny shortcut that collapses under scrutiny.

Consultants, bankers, lawyers… we all live in a world where a single wrong number can kill credibility (on my LinkedIn profile I recently posted the case of a $439K Deloitte artefact to an Australian government agency that contained LLM-generated numbers, and that created “some issues” for the firm…….)

In these areas, a human-built model that is 100% right is infinitely more valuable than an AI-assisted one that is “almost there.”

My view is that the smart play wouldn't be jamming GenAI into deterministic work, but redeploying it into ambiguity, creativity, and speed of iteration.

These are things where being “70% right” is actually a catalyst, not a liability.