AI for small business: where it actually pays back first
The pattern is consistent across sectors: the tasks that pay back are dull, repetitive and unglamorous. The ones that fail are the ones that sounded exciting.
Harriet Bishop, Founder, Archetype
Former Director of AI and strategy consultant. Founder of Archetype.
Small and mid-sized businesses tend to approach AI from the wrong end. They start with the technology, which model, which platform, which vendor, and then look for something to point it at. The businesses that get real returns start with a task that is already costing them money and work backwards.
Across the work we do, the returns cluster in a narrow and fairly unexciting set of places.
Where it reliably pays back
- Turning unstructured information into structured information. Invoices into spreadsheets, emails into tickets, notes into records. High volume, clear right answer, immediately checkable.
- First drafts of repetitive writing. Job specs, listing copy, standard responses, meeting summaries. A human still signs it off, but starting from 80 per cent rather than blank is worth hours a week.
- Forecasting against your own history. Demand, staffing, stock. Most businesses have years of usable data sitting in a system nobody queries.
- Search across your own documents. Being able to ask a question of ten years of contracts or specifications, instead of remembering which folder it is in.
Where it usually does not
- Anything customer-facing and unsupervised, early on. The failure mode is public and expensive, and the saving rarely justifies it in year one.
- Replacing judgement in your core commercial decision. If it is the thing you are actually good at, AI will help you do it faster, it will not do it for you.
- Bespoke builds before you have proven the workflow manually. If it does not work as a documented process with a person doing it, automating it just makes the mess faster.
- Anything chosen because a competitor announced it.
A test worth applying
Before committing to anything, ask three questions about the task. Does it happen at least weekly? Is there a clearly correct answer that someone could check in under a minute? Would the same person be doing it either way?
Three yeses is a strong candidate. Two is worth a trial. One usually means you have found something interesting rather than something valuable.
The uncomfortable part
The most common finding when we audit a business is not that they need new AI tools. It is that they are already paying for capable tools nobody was trained to use, and the returns available from that alone exceed anything a new purchase would deliver in the same period.
That is a cheaper problem to fix than most people expect, and a more annoying one to discover.