Coverage of AI in business tends towards transformation. What small companies actually do with it is narrower and genuinely useful.
The largest category
Writing nobody wants to do. Product descriptions, listing copy, routine emails, first drafts of documents that get edited anyway.
This accounts for more usage than everything else combined.
Extraction and sorting
Pulling structured information out of unstructured input – invoices, forms, enquiries, receipts – and putting it somewhere useful.
Unglamorous, immediately valuable, and easy to verify, which is why it works.
Customer enquiries
Drafting responses to the twenty questions that account for most of the volume, reviewed before sending.
The businesses doing this well keep a human in the loop. The ones that let it respond directly generate the failures that make the news.
Translation
For anyone selling across borders, where hiring a translator for every update was never affordable.
What does not get adopted
Anything requiring integration work, because there is nobody whose job that is.
Anything where the output cannot be checked quickly, because a confident wrong answer costs more than no answer.
The economics now
Prices fell sharply through 2026, which moved the threshold. Tasks not worth automating at previous rates are worth automating now, at volumes that are small by enterprise standards.
The practical approach
Pick the single task consuming the most time that requires the least judgement. Automate that one thing. Measure whether it saved time including the checking.
Most failed adoptions started with something ambitious rather than something boring.