When we started building the assistant behind Lameda, the early temptation was to make it as conversational and clever as possible. We deliberately pulled back from that, and it was the right call.
The tradeoff we chose
A more open-ended assistant is more fun to demo. It answers a wider range of questions, improvises, and feels impressively humanlike in a short conversation. It is also unpredictable in exactly the ways that matter for commerce: it can misquote a price, imply a policy that does not exist, or answer a question about a product it has never actually seen in your Catalogue. None of that is acceptable when the conversation ends with a customer paying real money.
So we built the opposite: an assistant that only answers from your Catalogue and your Knowledge FAQs, and that hands off anything outside that boundary instead of guessing. It is less impressive in a demo. It is far more trustworthy in an actual sale.
Why WhatsApp specifically makes this harder
WhatsApp conversations move fast, are often informal, and carry real expectation of a quick, correct reply. A customer asking about a price at 11pm is not going to tolerate ambiguity. Building for that meant designing the assistant to be fast on what it knows and honest about what it does not, rather than filling every gap with a plausible-sounding guess.
- Answers are grounded in your Catalogue and Knowledge, never invented.
- Anything uncertain routes to a Handoff instead of being answered speculatively.
- Payment confirmation depends on Paystack verification, never on what the assistant infers from the conversation.
What this looks like from a merchant’s side
In practice, it means your assistant will sometimes say, in effect, “let me get someone to help with that,” rather than answering everything on its own. That is not a limitation we are working to remove. It is a deliberate design choice, and it is the reason merchants can hand real customer conversations and real payments to it with confidence.