Mbalas
AI WhatsApp sales assistant that closes deals
11 followers
AI WhatsApp sales assistant that closes deals
11 followers
Mbalas is an AI-powered WhatsApp sales assistant that doesn't just answer FAQs—it actively closes deals. It handles customer chats 24/7 with human-like responses, calculates order totals and down payments, and automatically follows up with 'ghosting' prospects to recover lost sales. Designed for SMEs, it takes under 10 minutes to set up and learns directly from your past chat history. No rigid keywords, just a smart AI admin that sells.
How does it handle handoffs when a customer asks something the AI isn't sure about, like a complex pricing exception or a complaint that needs a real human touch?
@serapdzte Good question! Short answer: it knows when to tap out 😄 Every message goes through a quick triage — does the answer actually exist in the business’s knowledge base? If not, but it’s a casual question, the AI just says “let me get back to you on that” and quietly logs it for the owner. But if the customer is stuck without that answer (or asked twice), it fully hands off: tells the customer the team is taking over, pings the owner urgently with a one-line “here’s what they need,” and then goes completely silent until a human steps in. Complaints and “I already paid” claims skip straight to human, always. No awkward AI-pretending-to-be-helpful-while-a-human-types situation.
Does it hand off to a real person smoothly when a customer asks something tricky, or does the AI try to handle everything itself?
@basogulgon30432 It hands off — and honestly that was the part I obsessed over most. The AI literally can’t make up answers (it’s locked to only what’s in the merchant’s knowledge base), so when it hits something tricky it has two moves: defer politely and log it for the owner, or — if the customer actually needs that answer to move forward — do a full handoff and stop talking until a human takes over. Payment stuff always goes to a human, no exceptions. My biggest fear building this was a confidently wrong AI, so the whole thing is designed around “when in doubt, don’t wing it.”
How does the AI handle situations where a customer asks something super specific that isn't in the past chat history, like a custom modification or a complaint that needs real human judgment?
@sraaf2l Ah, good catch — let me fix that. The knowledge base isn't just something the owner types in manually:
it also grows from imported WhatsApp chat history (I can feed in old conversations and the system extracts useful Q&A patterns from them), and every time a human CS rep replies to a customer, that reply gets evaluated and folded into the KB too. If it's genuinely reusable info, not just a one-off personal answer. So the KB is really built from real conversations over time, not just manual entry. But the core rule still holds: whatever's actually IN that KB, however it got there, is the only thing the AI treats as fact. So a genuinely novel request, like a one-off custom modification nobody's asked before, still won't be answerable yet. The AI defers politely and flags it for me rather than guessing.
And complaints aren't an automatic "call a human" either the AI actually tries to resolve them first, using the same de-escalation instincts as its sales side: acknowledge the frustration for real, apologize once (not overdo it), then offer a solution if the KB has one. It only pulls in a human when things cross a real line — legal or "I'll go public" threats, or still upset after two genuine apology attempts. So it's "AI tries in good faith first, hands off when it's genuinely out of its depth" , not "any complaint = human."
The follow-up feature for ghosted chats actually works really well, I was surprised how natural the tone sounded in my test conversation. Setup took about 8 minutes from connecting my WhatsApp to having it ready, and it picked up on my old chat patterns fast.
@aleynanfyx Ahh this made my day, thank you! 🙏 The follow-up tone was the thing I rewrote the most times — it’s banned from saying stuff like “just following up!” and instead has to pick up the actual conversation thread (“about that product we talked about…”). It also checks whether the convo is genuinely done so it doesn’t nag people who already got their answer. And really glad the 8-minute setup wasn’t just marketing math 😅 If anything felt weird in your test, tell me — I read everything at this stage.
Setup was genuinely under 10 minutes and the follow-up messages to ghosted leads feel surprisingly natural. Wish I had this months ago for my small shop.
@naznegi1edi Thank you! 🙏 "Wish I had this months ago" is basically the sentence I'm building for — most small shop owners lose leads not because customers said no, but because nobody circled back. The under-10-minute setup was a hard target from day one since most of my users aren't technical at all. If the follow-ups ever start feeling less natural on certain types of chats, let me know — that's the part I keep tuning.
Finally got around to testing this and the follow-up on ghosted chats actually works, brought back two lukewarm leads I had written off last week. Setup was genuinely painless, just forwarded old chat history and it picked up the tone fast.
@beratbbdj Two revived leads from chats you'd already written off — okay that's the best kind of comment to wake up to 😄 That's exactly the use case: the leads that aren't dead, just... drifting. The follow-ups are designed to reopen the actual topic you were discussing instead of sending a "hey, still interested?" nudge, and it caps out after a graceful final message so nobody gets spammed into blocking you. Glad the tone matching worked on your history too. Curious — did the two that came back reply to the first follow-up or the second?