How many support questions can bots really close?

Every founder considering support automation asks the same question, and most vendors dodge it: how much of my inbox will a bot actually close? The honest answer is that it depends on what is in your inbox — but "it depends" is not useless. You can predict your number surprisingly well by looking at the shape of last month's questions, and this article shows you how.

Start with the shape of your inbox, not the bot

Support questions come in two kinds. The first kind has exactly one correct answer: "Where is my order?", "Do you ship to Ireland?", "How do I reset my password?". Nobody on your team enjoys these, nobody answers them better on the tenth try than on the first, and the customer does not care who answers as long as it is fast and right.

The second kind needs a decision. "The package arrived damaged and it was a birthday gift." "Can I return this even though it's been six weeks?" "I was charged twice and I'm furious." These conversations involve judgment, empathy, and the authority to spend goodwill — and they are where your team earns loyalty.

Bots belong entirely to the first kind. The mistake behind almost every bad chatbot story is a bot that was allowed to wander into the second.

Here is what that split looks like in practice. Imagine an online shop with 1,420 questions in a month. The team routes the clearly routine lanes — order status, FAQ, standard returns — to bots, and keeps everything else for people:

Questions in a month1,420

Bots allowed to try1,050

Closed without a human660

Handed over with summary390

The exact numbers will differ for your shop — that is the point of doing the exercise with your own inbox. What tends to hold everywhere is the structure: the routine pile is bigger than teams expect, and it is precisely the pile that automation handles well.

What bots close first: orders, FAQ, returns

Three lanes automate earliest, and it is worth understanding why.

• Order status. The answer lives in your systems, not in anyone's head. A bot connected to your order information answers "where is my package?" in seconds, at any hour — and this single question is often the biggest slice of a shop's inbox.
• Repeated FAQ. Shipping costs, size guides, ingredient lists, opening hours. If your help center answers it, a bot fed by that same knowledge base can quote it conversationally — and Botdesk highlights the gaps, so you can see which missing article would quiet the loudest stream of questions.
• Standard returns. "Within policy, unworn, has receipt" is a rulebook, and rulebooks are what bots follow perfectly. The bot walks the customer through the steps and only involves a person when something falls outside the rules.

Notice what these have in common: a single correct answer, verifiable from a source you control. That is the test for whether a lane is ready for a bot — not how "smart" the bot is.

Where a human stays in charge

Just as important is the list you should never automate. Refund exceptions, compensation, angry customers, anything legal, anything where the right answer depends on how much this relationship is worth to you — these belong to people, permanently. Not because bots produce wrong sentences, but because these decisions are yours to make, and customers can tell when nobody made one.

People keep

• Refunds outside policy
• Angry or at-risk customers
• Custom quotes and exceptions
• Anything spending goodwill

Bots take first

• Where is my order?
• Repeated FAQ answers
• Standard returns and swaps
• Password and account basics

The practical consequence: a good support bot is defined as much by its boundaries as by its answers. In Botdesk, each bot runs its own lane — one on FAQ, one on order status, one on returns — with its own playbook, its own knowledge sources, and explicit limits on what it may do. A bot that is not confident does not guess. It hands the conversation to your team with an AI-written summary of everything so far, so the person picks up mid-conversation instead of asking the customer to start over. You can see how that division of labor is set up on the bot teams page.

How to estimate your own number in an afternoon

You do not need a pilot project to predict your result. You need last month's inbox and two hours:

1. Pull a sample. Take 100–200 recent conversations — enough to be representative, small enough to actually read.
2. Tag each one as either "one right answer" (a bot lane) or "needed a decision" (a human lane). Be strict: if you hesitated, it is a human lane.
3. Count the piles. The first pile, minus a safety margin for awkward phrasing and edge cases, is your realistic automation ceiling.
4. Rank the bot pile by volume. The top three question types are your first three bot lanes — start there, not everywhere.

Most retail and service inboxes discover that the routine pile dominates. If yours does not — if every conversation is genuinely unique — automation will help less, and your money is better spent on a good help center first.

What this means for the "should we?" decision

Run the math from the exercise above against what routine questions cost you today: the hours your team spends on them, the evenings and weekends nobody covers, the customers who leave because an answer took a day. Then compare it with the price of automating — Botdesk plans are priced by auto-resolved conversations, with unlimited free seats for your team, and there is a 14-day trial that does not ask for a card, so the worked example above can be re-run with your own numbers before you pay anything. Details are on the pricing page.

The honest summary: bots will not replace your support team, and any vendor promising a specific percentage for your inbox before seeing it is guessing. What bots verifiably do is take the repetitive majority off your team's plate — the order-status stream, the FAQ echoes, the standard returns — and do it instantly, around the clock, in the customer's language. Your team keeps the conversations that were worth a human in the first place. That trade is the entire case for support autopilot, and for most repeat-heavy inboxes it is a strong one.