TopSales doesn't rush to quote. It asks existing customers what they care about, what they fear, and where they're stuck — writes the answers into a mind card — and lets that card decide whether to push, when, and who. Then it sends the message or makes the call itself.
2–5% of customers held out at random. We're paid only on incremental net GMV.
“Milk tube, or the tray?”
“mostly the milk tube”
Confirm the trade-off first
Card and next step rewritten
One customer record, played through in 15 seconds. All they said was "the last one needed daily descaling." TopSales didn't quote. It asked what exactly was painful, and let the answer decide whether this round sends a link, makes a call, or waits.
The last one needed daily descaling. We stopped using it. Message me after 8pm.
Which part was the hassle — the milk tube, or the grounds and drip tray?
Mostly the milk tube. I only drink black coffee.
For black coffee you'd just empty the grounds and rinse the tray; the milk line only needs a clean when you use it. Is that acceptable day to day?
That's fine. Call me.
Asking the price isn't enough to tell you why they hesitated. TopSales turns what they said, what they've been through, and how their tone shifted into a mind card that keeps updating — separating what's known, what's inferred, and what's still unknown.
How much is this one?
Let me think. The last one needed daily descaling. We stopped using it. Message me after 8pm.
1 machine, 2 years ago · no repeat order since
That's all. Two messages, plus their past orders.
| Field | Current reading | Evidence |
|---|---|---|
| Cares about | Whether daily use is a hassleinferred | "daily descaling" |
| Afraid of | Buying something that sits unusedinferred | "we stopped using it" |
| Stuck on | Hasn't accepted this model's upkeep | Upkeep is the live topic; the worry is unresolved |
| Current mood | Cautious, holding backinferred | "let me think" + a past abandoned machine |
| Shift | From asking price to explaining why the last one failed | Change between the two messages |
| Still unknown | Which cleaning step bothers them most; whether price is also an obstacle | Confirm on the next contact |
The next thing to confirm is the last row: which cleaning step they least want to do.
One card decides what to do this round, when to do it, and who should. Each of the three calls carries its reason.
| Call | Decision | Why |
|---|---|---|
| Push or not | Keep talking, no order push | The upkeep worry is unresolved; a checkout link has nothing to stand on |
| When | Today, 20:10 | The customer said "after 8pm" |
| Who | TopSales AI | This round is about usage habits and upkeep, not price negotiation |
20:10 — AI sends one message: which cleaning step bothers you most?
A new reply from the customer can overturn any of the above.
At the agreed time TopSales contacts the customer directly: settle the worry first, then answer from the product documentation. The purchase link goes out only once they've said the trade-off is acceptable.
"When you say upkeep — is it the milk tube, or the grounds and drip tray at the end of the day?"
"Mainly the milk tube. I only drink black coffee. Just call me."
Confirmed step by step from the documentation — no glossing over:
AI"For black coffee you'd empty the grounds and rinse the drip tray; if you use the milk line you'd also clean that. Is that acceptable day to day?"
Customer"Yes. I just don't want to take the milk tube apart every day. Send me the link."
"Here's the model we just went through, with the cleaning notes so you can check them yourself."
Product facts sourced from the brand's confirmed cleaning documentation (illustrative). Four things happened before the sale, each one on the record: the customer named the specific worry → got accurate information → accepted the trade-off → asked for the link.
This conversation settled a specific question: the customer can live with the grounds and drip tray, but not with taking the milk tube apart daily. Replies, purchases, refusals and refunds all go back onto the card and trigger the next decision.
| Field | Before | After |
|---|---|---|
| Worry | Upkeep, exact steps unknown | Milk tube is the dealbreaker — confirmed |
| Habit | Unknown | Drinks black coffee only |
| Acceptable | Unknown | Accepts grounds and drip tray |
| Intent | Cautiousinferred | Asked for the exact model link |
Card updated to "purchased"; refunds and cancellations stay tracked.
Next step: stop chasing this sale.
Customer says no: record "upkeep unacceptable for this model" and stop recommending it.
Customer goes quiet: record "worry unresolved" and re-judge the timing — silence is not the same as no interest.
The objective is always whether this sale closes. Replies are a signal along the way, not the goal.
From an agreed customer list, 2–5% are held out at random and keep running as before; the rest are followed up with TopSales. Over the same period, on the same refund and cancellation basis, we compare net GMV per customer between the two groups.
No TopSales. Net GMV per customer recorded as-is.
AI decides whether to push, when, and who. Net GMV per customer recorded.
=(net GMV per customer, TopSales − net GMV per customer, holdout) × customers in TopSales group
We take an agreed share of positive increment only. No increment, no cut.
The sale above only explains how the follow-up works. Whether the whole base sold more is a question only the randomised holdout can answer. With a small base or wide order values, results will swing — so the measurement terms go into the contract up front.
We agree the eligible customers, the holdout, and the settlement terms before anything starts. No system access needed — an export of past conversations is enough.
Built by renlab · [email protected]