Median assistant reply
Across 2,216 replies. Not one waited longer than a minute.
Conversations closed with no advisor involved
251 of the 973 chats that received a message.
Messages delivered
53 of 13,398 failed, each with its reason recorded.
What changed
Six seconds, including the days nobody is working
The assistant replies in a median of six seconds, nine in ten inside eleven, and 99.9% of the time in under a minute. Across 2,216 replies in the period, not one took longer than five minutes. An advisor replying by hand during business hours takes a median of 5.4 minutes, which is a reasonable human speed and roughly fifty times slower.
The number that matters more is what happens when nobody is there. At night, on public holidays and across 523 out-of-hours messages, the median held at six seconds. On Sundays there are no advisors connected at all, so every single reply that day is automatic.
Nothing arrives to find the lights off. That is the part a bigger team cannot buy.
What the team stopped doing
One conversation in four never reaches a person
Of the 973 conversations that received a message in the period, 251 were resolved without a single advisor opening them. The assistant now produces 34% of everything the system sends, which is a third of the outbound volume that nine people no longer type.
The rest escalate on purpose rather than by default. There were 712 explicit handovers in the period and 674 of them, 94.7%, ended resolved. A chat that reaches an advisor arrives already identified, with its history on screen, so the first thing a person does is answer rather than work out who this is and what they asked for.
The deflection is not the whole point. The handover being clean is.
Whether it holds up
Volume that swings five-fold, and nothing dropped
A typical business day runs about 500 messages across 45 active conversations. The busiest day in the period hit 1,168 and the quietest business day 225, a five-fold swing the system absorbed without slowing down. Alongside the text it processed and filed 612 photos, voice notes and documents, all readable from inside the conversation they arrived in.
99.6% of messages were delivered. The 53 that were not all hit the same thing: WhatsApp requires an approved template once more than a day has passed since the customer last wrote. The system records the exact reason against the message and shows it in the conversation, so the advisor knows to retry rather than assuming it went out.
Nine people sharing one inbox needs memory, not just speed. The period carried 612 internal notes that the customer never sees, 19 labels applied 283 times, per-contact internal nicknames, chat assignment with reassignment, and full conversation export. None of it is visible to the recipient and none of it is read by the assistant.
How these numbers were counted
Every figure comes from direct queries against the production database for 21 September to 19 October 2026, with timestamps converted to Colombia time. The two go-live days before that window are excluded; they carried six and two messages. “Typical business day” is the median of Monday to Saturday rather than the mean, because several bulk-send days pull the average up. Sundays are reported separately since no advisor is connected. Response time is the interval between an incoming message and the next outgoing message in that same conversation, and a conversation is only counted once it has received at least one incoming message, so outbound sends with no reply are not counted as support. Flik Flak runs the messaging module only: there is no enrolment, signup or payment data in this system, and none is claimed here.