When we started looking for a pilot partner for Sinan AI, the obvious choice would have been another software company. Software companies have clean data, technical founders who understand what an AI agent is doing under the hood, and traffic patterns that are easy to model. We picked Biscotto instead, a small homemade cookies and tiramisu business, and we want to explain why.
Small businesses don't have the luxury of a funnel
A lot of what gets written about AI sales agents assumes a business with a marketing team, a CRM, and a defined funnel someone can point an AI at. Biscotto doesn't have any of that. It's a small operation selling a physical, perishable product in limited batches. There's no marketing department deciding what the AI should say. There's just a person who bakes and someone who has to answer messages about orders, and until now, not much time to do both well.
That's closer to what most of the businesses we're building Sinan AI for actually look like. If Sinan AI only works when a business already has clean systems and a person dedicated to managing it, it isn't useful for the businesses we say we're building it for. Testing it somewhere without any of that scaffolding is a more honest test of whether it holds up.
Homemade and limited-batch products create real constraints
A cookie business selling small batches has a problem a lot of e-commerce advice doesn't account for: you can run out. Inventory isn't infinite, and it isn't restocked on demand the way a dropshipped product would be. That means any AI agent talking to customers or driving traffic has to be aware of what's actually available, not just what's technically listed on the store. We're paying attention to how that plays out in practice, because it's the kind of constraint that's easy to overlook when you're building for a hypothetical business rather than a real one.
What we're actually tracking
We've already written about the specific traffic and conversion numbers we're watching in the early days of this pilot. What we want to flag here is the limits of what one pilot can tell us. Biscotto is one business, in one category, with one owner's judgment about tone and pricing baked into every decision. Anything we learn from this pilot is a data point, not a conclusion. We're not going to claim Sinan AI works for small food businesses generally based on one run with one bakery, and we'd be skeptical of anyone who did make that claim from a single pilot.
What we can say is that running the pilot with a business this small and this unstructured forces Sinan AI to work without shortcuts. There's no existing funnel to plug into, no marketing calendar to follow, no team to catch mistakes before a customer sees them. If something breaks, it breaks in front of a real customer trying to order tiramisu, not in a demo environment.
Why this matters for how we build
We'd rather find the gaps in Sinan AI on a business this small, where the stakes of a bad message or a missed order are limited, than discover them later with a customer who has more on the line. That's the actual reason we chose Biscotto. Not because it's a flattering case study, but because it's a harder and more honest one.