I messaged a shop at 11:40 on a Tuesday night. One question: does the boot come in a 43. The answer arrived at 9:12 the next morning. By then I had bought a different pair from someone else.
That is not a story about rude service. The seller was asleep, which is a very reasonable thing to be doing at midnight. It is a story about timing, and timing can be counted. So we counted it. We scanned ~7.7 million messages across ~2 million conversations and 15,000 channels, looking at one number: how long a customer waits for a reply.
What the data says
| Who replies | Median reply time | Average reply time |
|---|---|---|
| AI | 9 seconds | 13 seconds |
| Human | ~60 minutes | ~7.5 hours |
Put the two rows side by side. The median AI reply lands roughly 400 times faster than the median human one. Not 400 percent. 400 times.
The average tells the story the median hides
Look at the human row again. The median is 60 minutes, but the average is 7.5 hours. A gap that wide is a long tail of very late replies: messages that land at night, messages buried under a busy afternoon, messages nobody sees until tomorrow. To a customer who was ready to buy an hour ago, 7.5 hours means never. We have written before about how little patience DM customers actually have, and this is the same number seen from the other side.
AI is not only fast, it is steady. Median 9 seconds, average 13 seconds, almost no tail. A customer who writes at 3pm and a customer who writes at 1am get the same experience. That steadiness is worth as much as the raw speed, because the only thing worse than a slow reply is an unpredictable one.
There is a third thing in this data, and it is the uncomfortable one. A large share of the fast replies in DMs are no longer written by people. Automated answers, keyword rules and AI together now carry most of the instant response load. That is the new floor. A seller still answering everything by hand is not competing with another seller. They are competing with a machine that answers in nine seconds and never sleeps.
What actually works
1) Treat speed as your cheapest conversion lever. You do not need a discount or a bigger ad budget to sell more. You need to answer in seconds instead of minutes.
2) Cover the hours you are offline. That 7.5-hour tail mostly comes from night and early morning. Our look at the busiest DM hours found a peak that lands squarely while most sellers are asleep. An automated first reply covers exactly those hours.
3) Split the work: AI first, human for the hard part. Let the assistant take the first second and keep the conversation alive. You step in where real judgement is needed, which is a much smaller share of messages than most sellers assume.
4) Measure your consistency, not just your average. A human team can be quick on a good day and invisible on a bad one. That swing costs customers. The target is a fast reply every time.
The takeaway
DM sales are decided in seconds. A 60-minute median is already too slow for a passing moment of curiosity, and a 7.5-hour average means opportunities quietly burning every night. AI changes the arithmetic: nine seconds, steady, around the clock.
That is what Vardast does. It answers in seconds at any hour and keeps the conversation going until the customer is ready to buy, so you stop losing the race you did not know you were in. See how Vardast works.
Based on our own analysis of ~7.7 million messages across ~2 million conversations and 15,000 channels over a 14-day window; times are the median and average gap between a customer message and the reply that follows it.