When your customers actually write
Maybe most of your questions arrive in the evening, maybe during the lunch break. You will only know from your own data. Take the last full month and lay the order times from your shop panel next to the message times from your inbox. Then put your real schedule on top of both, the one where you pack parcels in the morning and sit down to email around eleven. Count weekends separately. Also make sure your analytics and your inbox use the same time zone, otherwise evening questions will land in the afternoon column.
Look for the places where a question waits longest. Traffic at 10 pm tells you little by itself, because someone browsing your range on the sofa may not need any help. Next to each message, note the topic and the time to the first useful answer. An automatic “we received your message” does not count.
Topics tell you more than hours. If most of the backlog comes from one confusing size chart, fix the chart and see next month whether the questions went quiet. You will not solve it any cheaper. For questions that need individual advice, check whether the answer can be read from the product data or whether you need someone who knows the goods inside out.
How to work out whether after-hours support pays off
Take round numbers that you can swap for your own later. Say 40 questions a month arrive after hours from people who have not bought yet. Assume that after a faster answer four of them order, each for PLN 200. That makes PLN 800 in revenue, and this is the point where it is very easy to stop counting.
Far less than that reaches your pocket. After the cost of the goods and everything you pay per order, say you keep PLN 50 on an order, which is PLN 200 a month. That is the amount that has to pay for evening support, whether it is your own evening or a bot subscription. If support costs more, this scenario will not fund it.
Also work out the version with zero extra orders. What remains then is time, because in the morning you no longer answer the same delivery questions. Write down how many minutes a day repetitive questions take you now, and check the same figure after the change.
Do not count the value of every evening basket as a loss. You do not know which of them would have ended in a purchase even with an answer in one minute. Someone who put a pair of shoes in the basket at 11 pm may simply have been checking the delivery cost.
How do you handle customers outside working hours?
An on-call rota, yours or someone else's on shifts, makes sense when the evening questions need knowledge that is not in the product descriptions. Matching parts to a specific car or a made-to-measure order is best handled by a person. But the calculation above has to come out clearly above PLN 200, because you pay for every hour at the computer, including the evenings when nobody writes.
Live chat with no human behind it changes nothing in the evening. The customer types in the window and gets the same “we will reply soon” they would have got by email. If you run a chat staffed by people, show the hours when someone is actually there, and after closing give a deadline you will keep even on a day with thirty parcels to pack. Something like “Support is back tomorrow morning. Leave your question and email and we will reply by noon” is enough, as long as your schedule really allows it.
An AI bot answers in the evening the questions it has data for. CartBot searches your catalog and the shop's knowledge base and shows the products it finds in the conversation with a photo and a price. When it does not know something, it asks the customer for contact details and leaves the case for a human, so in the morning you see the whole conversation in the panel. It can do nothing about a dimension that is missing from the catalog. If the description of a monitor mount does not list the VESA hole pattern, you have to add it.
Test an evening with nobody at the computer
Before you launch anything, play the customer at 10 pm. Ask a question that the descriptions do not answer and see whether you can leave an email or a phone number without digging through a form. Then check where that contact lands and who is responsible for getting back to it in the morning. In a one-person shop this is simple. With two people it is easy to end up with a morning where each of them thinks the other one replied.
Compare the numbers after a month
Collect the enquiries from the month after the start and set them against the ones from before the change. Compare the time to the first useful answer and the number of cases you had to come back to in the morning. If a promotion started in that period or you changed your ads, allow for it.
Once measurement is set up, CartBot shows the orders linked to conversations. That is a good starting point, but you do not know how many of those people would have bought without the chat, so do not enter that amount as extra profit. The calculator on our home page works the same way, because it is a simulation built on fixed assumptions. Next to its result, put the real number of after-hours questions from your inbox.
You can take the first step this evening. Go through last month's messages and add the hour and the reply time to each one. After an hour of that work you will know whether this problem exists in your shop at all.