BoteamChoosing the first role: pick the work that repeats
Most small businesses have one or two tasks that eat the week in a predictable pattern. That pattern is where the first managed role belongs.
Find the pattern, not the pain
The instinct is to automate whatever annoys you most. That is usually the wrong choice, because the most annoying work is often the most variable. Instead, look at the last month of incoming work and ask: does the same kind of request arrive in the same shape, and does it get answered the same way each time? A quote for a standard job. A booking confirmation. A supplier order. A monthly report. Those are candidates. Anything that needs a fresh judgement every time — a pricing exception, a difficult client, a scope change — leave it alone for now.
Put a number on the dropped threads
The real cost of repeatable work is almost never the work itself. It is the threads that drop when the week gets full. The quote that went out three days late and lost to a competitor who replied in an hour. The payment reminder nobody sent. The Google review nobody replied to. The research on response speed is unusually consistent here: 78 per cent of customers buy from whoever responds first, and lead value decays about 50 per cent within 48 hours. Naming that cost is what turns “we should look at AI” into a decision with a number attached.
One role. Not five.
The temptation is to hand over everything at once, because the possibilities are obvious once you start looking. Resist it. One well-defined, well-maintained role proves the model, surfaces the problems you did not anticipate, and gives you a real standard to judge the next one against. The second role is much easier than the first, but only if the first actually settled.
Judge it on the follow-through, not the demo
The question is not whether the role can produce a good quote in a demo. It is whether it produced them every day for a month, including the Friday afternoon ones, without you checking each line. Give it four to six weeks before you decide. The biggest time gains in the research come in that early window, and then they plateau unless you deliberately expand what the role covers.