What a small business is actually deciding
The question is not whether AI is impressive. It is whether any of it changes what happens between nine and six in a business where the same person answers the phone, sends the quote and does the chasing. Almost every AI CRM demonstration is built around a company with thousands of leads and a dedicated operations team. You are not that company, and the features that shine in that demonstration are frequently the ones that do nothing for you.
The useful frame is triage and recall. A small business does not have a targeting problem, it has a time problem: more enquiries than hours, and no way to know which hour was well spent. AI that orders your day and remembers your conversations addresses that directly. AI that predicts quarterly revenue from a hundred data points does not.
Small pipelines are bad training data, and that is fine
A model learns patterns from outcomes. With a modest number of closed deals there is not much to learn, so predictions carry wide error bars whatever the interface implies. This is not a reason to avoid AI; it is a reason to prefer the features that work on the content of a conversation rather than on the statistics of your history. Summarising a thread does not require a large dataset. Forecasting does.
The real bottleneck is usually capture, not intelligence
In most small businesses a meaningful share of enquiries never enters any system. They arrive on someone's personal WhatsApp, as a missed call, or as a message on a social account nobody owns. Adding AI on top of a half-complete record set produces confident rankings of the wrong list. Fix the capture first; it is less exciting and worth considerably more.