Round-the-clock customer service with two people rests on three layers rather than on overtime: an AI agent that receives every conversation first, a human notified only on the cases it cannot handle, and a transparency reply for the rest. The team does not sit up at night. It decides in advance what the night handles on its own.
Why Half Your Sales Happen Outside Office Hours
Your customers write when they have time, not when you are open. Around 40% of the messages an online store receives land between 7 p.m. and 1 a.m., sent from a bed or the back of a taxi. A store that closes at 6 p.m. leaves that whole window unanswered until the next morning.
The problem is not the delay itself, it is the comparison. Someone who asks a question at 10 p.m. usually asks two or three sellers in the same breath. Whoever answers first takes the order, and not by being the cheapest: being there when the decision is made is enough on its own.
Office hours do not protect your day either. The night’s messages do not disappear, they pile up and wait for you at 8 a.m., mixed in with the morning’s. The next day’s queue eats the time you set aside for packing orders, and the backlog rolls straight into the day after.
Why Hiring Someone for the Night Shift Solves Nothing
Hiring someone for the night shift costs a full salary and still does not cover the need. One person handles messages one at a time, while the peaks arrive in parallel: fifty open conversations after a story do not clear any faster because somebody happens to be awake to read them.
Quality is the second problem. A reply written at 10 p.m. by someone ten hours into their day does not carry the precision of the same reply at 10 a.m. The customer cannot tell tiredness from carelessness: they read a vague price, and they go and check somewhere else.
Round-the-clock coverage therefore comes from stacked layers rather than from hours of presence. Each layer handles what it knows how to handle and passes the rest up, which keeps a two-person team on the only cases that genuinely require human judgement.
Layer 1: The AI Agent Takes Every Conversation First
The first layer is an AI agent connected to your catalogue, which receives every incoming conversation before you do. It handles roughly 80% of requests with no human involvement, because those requests concern information that is already written down somewhere: a price, a stock level, a delivery zone.
These requests look much the same from one store to the next, and they come in at 11 a.m. as readily as at 2 a.m. The agent answers with the same precision at both hours, since it reads the same product sheet and the same delivery table. Here is what it absorbs without waking anyone:
- Price and availability questions
- Delivery zones, their rates and their lead times
- Sending the payment link and confirming it
- Return and exchange conditions
- Anything your FAQ already covers
A conversation picked up within the minute stays alive: the customer replies, gives their size, sends their address. Picked up the next morning, it is usually dead, because it arrives after the decision. That is exactly what the effect of a reply sent in under 5 minutes measures.
Layer 2: Light On-Call Duty and the Transparency Reply
The second layer is a human notified only when the AI agent steps outside its remit. A complaint, a price negotiation, a request that is not in the catalogue: the agent recognises that it should not decide alone and sends a notification. The rest of the time, your phone stays quiet.
On-call duty alternates between the two people on the team, every other evening. The volume amounts to a handful of notifications per evening rather than a continuous watch: whoever is on call gets on with their evening and opens the app when it buzzes. That is the difference between being reachable and being on duty.
The third answer is an acknowledged wait. When a request lands at 2 a.m. and needs a human, the agent confirms what it does know — price, stock, lead time — then states that a team member will pick the conversation back up from 8 a.m. The customer leaves with useful information and a specific hour.
What to Document Before You Plug In the Agent
An AI agent only handles 80% of conversations if it has something to answer with. An incomplete catalogue produces an agent that escalates everything, and the on-call layer then starts ringing all night long over plain questions about prices and delivery zones.
Six blocks of information cover almost every request an online store receives. They are prepared once, in a single working session, and then serve indefinitely; keeping them current as new stock arrives becomes the only real upkeep your customer service needs:
- An up-to-date catalogue: photos, prices, variants, real stock
- Your delivery zones and what each one costs
- Delivery lead times, zone by zone
- Return and exchange conditions
- The payment methods you accept
- Current promotions and their end dates
This documentation pays off well beyond the AI agent. The same information feeds order tracking and follow-ups, and it is the starting point for the work described in handling the orders that arrive in your DMs. An hour of writing here is worth several weeks of improvised answers.
Measure Coverage, Not Effort
The right measure of round-the-clock customer service is the share of out-of-hours messages that get an answer before the next morning. Over one week, count the messages that arrived after 6 p.m. and before 8 a.m., then count those that received a reply within the hour. The ratio is your coverage rate.
That coverage rate reads without interpretation. Below 50%, most of your evening conversations wait out the whole night, and that is precisely the gap the first layer fills. Measure again four weeks later over the same time window: the comparison tells you what the agent actually changed.
Finally, count the time the team no longer spends in the inbox, on the same logic as the hours won back from your phone. Two people plus an agent do not replace a large team, but together they hold a volume of conversations that neither of them would get through alone, with nobody sitting up at night.