The short answer
Indian e-commerce traffic peaks between 8pm and midnight, which is after every support desk has closed. Around 60% of inbound queries arrive outside business hours, most are the same fifteen repetitive questions, and conversations that wait more than an hour for a reply frequently go silent for good.
Most of your customers ask their questions after your support desk has closed. Those questions do not wait until morning - they go to whoever answers first. A bot that replies from your own documents at 11pm is not a luxury feature; it is the difference between an order and a statistic.
It's 10:14pm. A customer in Pune has your product open in one tab and a competitor's in another. She has one question - "do you ship to Wakad?" - and she sends it to both stores. One of them answers in four seconds. The other answers at 9:30 the next morning. You already know who got the order.
This is the midnight-query problem, and almost every Indian SMB we've spoken to has it. Your customers don't shop during your business hours. They shop after dinner, on their commute, in the gaps of their own workday. India's e-commerce traffic peaks between 8pm and midnight - exactly when founders finally put their phones down.
The real cost of "we'll reply tomorrow"
When we studied support inboxes across early customers, three numbers kept repeating. Around 60% of all inbound queries arrived outside business hours. Of those, the vast majority were one of the same fifteen questions - shipping, returns, sizing, availability. And when a reply took longer than an hour, roughly half the conversations went silent forever.
That silence is the expensive part. It doesn't show up in any dashboard as a lost sale - it just looks like a customer who "didn't convert." Multiply it across a month and the midnight inbox is quietly one of the biggest leaks in the funnel.
A question answered in seconds is a sale. The same question answered tomorrow is a statistic.
When the questions actually arrive
Support desks are staffed for the working day. Customers do not shop in the working day. The mismatch is the whole problem, and it looks like this across a typical week for an Indian D2C store.
| Time of day | What's happening | Who is answering |
|---|---|---|
| 9am - 6pm | Steady trickle of queries | Your team, fully staffed |
| 6pm - 8pm | Commute browsing picks up | Whoever is still online |
| 8pm - midnight | Peak shopping and peak questions | Usually nobody |
| midnight - 9am | Low volume, high intent | Nobody |
The two rows with nobody in them are where the buying decisions get made. That is not a scheduling inconvenience - it is the majority of your funnel running unattended. The same pattern shows up sharper during festival season, which our festival playbook covers separately.
Why hiring a night shift isn't the answer
The traditional fix is people: a support agent on the evening shift, or the founder just… never logging off. Both fail for the same reason - the queries are repetitive, but they're not schedulable. Some nights bring three questions, festival-sale nights bring three hundred. You can't staff for that variance, and you shouldn't burn a human on "what's your return policy?" at 11pm.
The queries that genuinely need a human - a damaged order, a payment dispute, a bulk enquiry - are maybe one in ten. The problem is they're buried under the other nine.
Then there is the arithmetic. A night-shift support hire is a monthly salary, every month, plus recruitment, training and the fact that one person cannot cover seven nights. A subscription is a fixed line item you can cancel. We do not publish a salary figure here because it varies too much by city and experience to be honest about - but you already know what you would have to pay, and you can compare it against our plans in about a minute.
- A hire covers roughly eight hours of one time zone, five or six days a week.
- A hire needs notice, training on your policies, and cover when they are ill.
- A hire answering "what's your return policy?" for the fortieth time is an expensive way to read a PDF aloud.
What an unanswered question actually costs
The reason after-hours silence is easy to ignore is that it never appears as a loss. Nothing gets refunded. Nothing shows up as a complaint. The customer simply doesn't come back, and the whole event is invisible in every dashboard you look at.
Follow one of them through. A customer finds you at 10pm, likes the product, and has one blocking question - does it ship to their pincode, is it in their size, can they return it if it doesn't fit. They ask. Nothing comes back. They are not going to sit with the tab open until morning, so they either buy from whoever answered or they lose interest entirely. By 9:30am, when your reply arrives, the decision was made eleven hours ago.
Multiply that by the number of after-hours questions you get in a month. That is the number worth knowing, and it is the one nobody measures - because the only trace it leaves is a conversation that stopped. The phone version is worse still: a missed call leaves no message at all.
What the always-on layer looks like
A grounded AI agent changes the economics completely. Trained only on your documents, it answers the repetitive 80% instantly and accurately - with citations - at any hour. When it isn't confident, it doesn't guess: it collects the customer's name and question, and escalates with full context so a human picks it up in the morning with zero back-and-forth.
The result isn't just faster replies. It's that the 10pm window - the one you were losing by default - becomes the most productive selling time you have, without anyone staying awake for it.
What changes when the answer is instant
Three things shift, and only the first is obvious.
The obvious one: questions that used to go unanswered get answered, so conversations that used to die quietly convert instead. The second: your morning stops beginning with a backlog - which matters more than it sounds, because a backlog compounds rather than accumulates. Instead of forty messages of which four matter, you open the inbox to the four, each with the full conversation attached. The third is slower and more useful - the questions the bot could not answer become a list of the documents you are missing, so the knowledge base improves in the direction your customers are actually pushing it.
None of this requires anyone to be awake. That is the entire point.
The objection: won't customers hate talking to a bot?
Some will. The evidence from our own inboxes is that it depends almost entirely on two things: whether the bot answers the question, and whether it admits when it can't.
Nobody minds an instant, correct answer at 11pm. What people hate is a bot that loops, that answers a different question than the one asked, or that refuses to hand over to a human when it is clearly out of its depth. Those are design failures, not bot failures - and the fix is a low confidence threshold and a visible route to a person. We wrote about where that line should sit in designing escalation customers actually like.
Start with your fifteen questions
You already know your midnight queries - they're the same fifteen questions you answer every day. Write them down, upload the documents that answer them, and let the bot take the night shift. Setup takes about 5 minutes, and tonight's 10pm customer gets an answer in four seconds instead of eleven hours.
Two things worth reading next: how grounding stops a bot inventing answers, because an after-hours bot that makes things up is worse than silence, and what to put in the knowledge base first. If you sell online, the e-commerce walkthrough covers the specific documents that answer the after-hours questions.
Frequently asked questions
What share of support questions arrive outside business hours?
In the inboxes we have studied, around 60%. The bulk of them are the same fifteen questions - shipping, returns, sizing, availability - which is exactly the profile that can be answered from documents rather than by a person.
Is hiring a night-shift agent a reasonable fix?
Rarely. The volume is unpredictable - three questions on a quiet night, three hundred during a sale - so you cannot staff for the variance, and the work itself is repetitive enough that a person is an expensive way to read a policy aloud.
Won't customers be annoyed at getting a bot at 11pm?
Not if it answers the question and admits when it cannot. What frustrates people is a bot that loops, answers something else, or refuses to hand over to a human. An instant correct answer at 11pm beats a reply at 9:30am comfortably.
How quickly can after-hours cover be set up?
About five minutes for the widget, plus the time it takes to upload documents covering your most-asked questions. Most businesses already have those documents written.
