AI for small business, realistically.

Where it reliably pays

The pattern is consistent: high-volume, low-stakes, language-shaped work where a mistake is cheap and obvious.

• First-line support. A large share of enquiries are the same dozen questions. Answering them instantly at 2am is genuine value; escalating the rest to a human is the part that must work properly.
• Drafting. Proposals, follow-ups, product descriptions, job adverts. A decent first draft in seconds turns a 40-minute task into a 10-minute edit.
• Summarising. Calls, long threads, documents. Reliable and immediately useful.
• Structuring messy data. Turning free text into consistent fields — a task that is tedious for people and easy for a model.
• Triage. Reading an inbound enquiry and routing it correctly.

Where it does not

Be equally clear about the other side:

• Anything where a confident wrong answer is expensive. Legal, medical, tax and financial specifics need a human who is accountable.
• Final numbers. Use a calculator for arithmetic that matters. That is what our free tools are for — deterministic and checkable.
• Work requiring real context you have never written down. A model cannot infer the history of a client relationship that exists only in your head.
• Tasks that happen twice a year. The setup and supervision cost more than doing it manually.
• Anything you would not sign your name to unread. If it needs full review every time, the saving is smaller than it appears.

How to start without wasting a quarter

1. Pick one task. Daily, boring, and cheap to get wrong.
2. Time it honestly first. Without a baseline you cannot tell whether anything improved.
3. Run it alongside the human version for two weeks. Compare outputs rather than trusting impressions.
4. Count the supervision. If checking takes as long as doing, it has not worked.
5. Only then expand. One proven use beats five half-configured ones.

The questions worth asking a vendor

• Where does my data go, and is it used for training? Get it in writing.
• What happens when it does not know? "Escalates to a human" is the right answer. "Answers anyway" is disqualifying.
• Can I see what it did? Transcripts and logs, or you cannot audit anything.
• What is the true monthly cost at my volume? Usage pricing has a habit of surprising people.
• What happens if I leave? Export, and whether the configuration is portable.

This is the standard we hold ourselves to: AWGCOR AI ships with an explicit rule that the assistant says when it is not configured or cannot see something, rather than inventing an answer about your business. A confident fabrication about your own numbers is worse than no answer at all.

What to expect, honestly

For a small business, realistic outcomes look like: several hours a week returned on drafting and admin, faster first responses to enquiries, and less time spent assembling reports.

What it does not look like is replacing a team member outright. The reliable win is compressing the mechanical portion of knowledge work so the people you already have spend their time on the parts that need judgement. Price it against that, and it usually pays. Price it against replacing a salary, and it usually disappoints.