Every AI assistant for a small business has two costs: the build, and a monthly model usage bill that grows with every request. Choosing the model for each task, instead of sending everything to the most expensive one, is what keeps that second bill under control.
TheMarker reports that successful AI products are squeezing the margins of the companies that sell them. The more customers use them, the bigger the API bill. Companies in Israel and abroad are cutting their dependence on OpenAI and Anthropic models and moving to open models, both to save money and to control their own product. One industry manager says companies don't want to stop, and the costs are blowing up on many of them from the inside.
What this means for a small business
A small business doesn't make millions of calls like a SaaS product, but the same logic applies at a smaller scale. In a lead-handling or invoice pipeline, most steps are simple: classify a message, pull fields from a document, route it to the right person. A small, cheap model handles those well. Claude is worth paying for where reasoning matters, such as contract review or a tricky customer reply. Splitting the work this way is one of the main cost drivers I break down for projects. Open models still cost money, though. You pay for hosting or a provider, and their quality in Hebrew or on messy scans has to be tested on your own data before you switch.
Why it matters: Before launch, ask what the assistant will cost per month at your real request volume, and which steps could run on a cheaper model.
Source: TheMarker Technation