Where AI Actually Belongs in Business Software
Past the hype, AI is genuinely transforming a narrow set of business tasks. Here's what's real, what's not, and how to tell the difference.
Every business owner I talk to now asks about AI, usually in one of two modes: anxiety ('are we falling behind?') or scepticism ('is any of this real?'). Both are reasonable. The truthful answer is that AI is transformative for a specific shape of task and a distraction elsewhere — and knowing the difference is worth real money.
The shape of a good AI task
AI earns its keep on tasks that are high-volume, language-heavy, and tolerant of review: answering the same fifty customer questions, extracting fields from documents, drafting first versions of content, qualifying and routing incoming leads, summarising long records for humans who decide.
Notice what these share: a human currently spends hours on them, the cost of an occasional imperfect draft is low, and there's a clear checkpoint before anything irreversible happens.
Where AI doesn't belong yet
Anything where a wrong answer is expensive and unreviewable: final pricing decisions, medical or legal conclusions, irreversible customer commitments. The technology can assist in all of these — drafting, flagging, summarising — but a system that lets a model act unsupervised in high-stakes flows is a system designed by someone who won't be answering the phone when it goes wrong.
What implementation actually looks like
The successful AI projects I've built are unglamorous. A support assistant strictly limited to the client's documented content, that hands off to a human the moment it's unsure. A pipeline that reads incoming documents and pre-fills a form a person confirms. Each scoped to one task, measured against the hours it saves.
The failed projects I've seen share a pattern too: they started with 'we need AI' instead of 'this task wastes twenty hours a week.'
A sober way to start
List the tasks your team repeats most. Pick the one that's language-heavy, high-volume and low-stakes. Automate that one task, measure the hours saved, and only then decide on the next. AI adoption that works looks like compound interest, not a moonshot.
Written by Abhinav Saxena — founder of Kodinav, an independent software studio.