The cost is in the engineering around the AI, not the AI. The first lever is integration: an automation that touches one tool is cheap, while one that spans your CRM, email, accounting and WhatsApp costs more because it does more real work across your systems. The second is logic: the more decisions the automation makes on its own, the more there is to build and test. The third is volume, the amount of work it runs through once it is live. None of these three is the model itself, which sits in the background as a low, per-task cost.
This is also why scope, not the technology, decides the bill, and why the honest framing matters: cost is the most cited barrier to AI adoption for SA small businesses, named by roughly 58% of them (industry survey, 2026, directional). The way past that barrier is not a cheaper bot, it is a tighter scope. A focused piece of AI automation aimed at one expensive process earns its price quickly, while a sprawling platform bought before anything is proven is where the money leaks.
