The expectation people bring to WhatsApp comes from messaging their friends, not from emailing companies. That sets a bar most support teams have not consciously decided whether to meet.
Measure the right thing
Three distinct numbers get muddled constantly:
- First response time — customer message to first human reply. The one customers feel most.
- Subsequent response time — the gap between replies mid-conversation. Often worse than the first and rarely measured.
- Resolution time — start to resolved. Matters for the business, felt less acutely.
Teams that report only first response time can look excellent while making customers wait forty minutes between every subsequent reply — which is the experience people actually complain about.
Use the median and a high percentile, never the mean. One conversation answered after nine hours drags an average into meaninglessness. Median plus 90th percentile tells you what typical feels like and how bad the bad cases are.
Setting a realistic target
Start from your data, not from an aspiration. Measure your current median and 90th percentile over a full month including your busiest days. Then set a target modestly better than the current median, and work on the 90th percentile separately — that tail is usually where the complaints originate.
A target you miss regularly is worse than a looser one you keep, because you will publish it in your auto-reply and customers will hold you to it.
Tier by urgency
- Payment failed, order about to ship wrongly, service down — minutes.
- Order status, delivery questions, returns — an hour or two.
- Product questions, general enquiries — same day.
- Feedback, suggestions — next business day.
A single blanket target forces you to treat a broken payment like a product question. Routing by keyword at intake makes this practical rather than theoretical.
Publish only what you will hit out of hours too
If your auto-reply promises two hours, someone messaging at 8pm expects a reply by 10pm. Either state working hours explicitly in the same message, or set the promise to something that holds overnight. Ambiguity here generates more annoyance than a slower stated time would.
Fixing a slow tail
- Look at when the 90th percentile happens — it is usually a specific hour or day, not a general problem.
- Check whether slow conversations share a topic. A single unclear policy can generate long stalls.
- Check assignment. Conversations that sat unassigned account for a large share of slow first responses.
- Check for the handoff gap between automation and a human.
Speed is not the only goal
A fast wrong answer creates a second conversation. Track repeat-contact rate alongside response time, and if speed improves while repeats climb, the team is being pushed to answer before understanding. That is a target set too aggressively, not a team performing badly.