Where AI actually saves time in a smaller company
The useful applications are unglamorous. They are also the ones that survive past the pilot.
The pattern to look for
AI pays off where work is high volume, text-shaped, judgement-light and currently done by a person who would rather be doing something else. Where all four hold, it works. Where any is missing, be sceptical.
What we see working
Document extraction. Supplier invoices, delivery notes, contracts. Pulling structured fields out of unstructured documents was hard for decades and is now routine. The saving is immediate and measurable.
First-draft correspondence. Replies to common enquiries, drafted for a human to check and send. The draft is worth eighty percent of the time; the human keeps the accountability.
Search across your own material. Ten years of proposals, specifications and notes, actually searchable by meaning. Most companies have this archive and no way in.
Meeting notes and actions. Transcription with a summary and a list of commitments. Small per meeting, substantial per month.
Triage and routing. Classifying incoming email or tickets by topic and urgency, so the right person sees it first.
Translation. For a company operating across languages, competent first-pass translation of internal material removes a real bottleneck.
What does not work yet
- Anything where a confident wrong answer is expensive and unverified
- Final financial or legal output without review
- Replacing judgement that depends on context the model cannot see
- Tasks done twice a month — the setup will never pay back
How to choose the first one
Ask where people spend hours on something repetitive that they describe as annoying. Measure the current time honestly. Pick the task with the highest hours and the lowest cost of being wrong.
Then keep a person in the loop. The reliable pattern is not automation — it is a good first draft, checked quickly by someone who knows the business.
We scope and build these under AI implementation.