"It'll save you hours." Every chatbot pitch says it, and none of them show the arithmetic. So let's do the arithmetic, because a real estimate is worth more than a confident promise, and it'll tell you whether automating your support is worth the trouble or not.
The good news is the math isn't hard. You need a few numbers you can pull from your own inbox in an afternoon, and a bit of honesty about the costs that never make it into the sales deck.
Start with what actually repeats
Open your support inbox and look at a normal week. Not the weird tickets, the ordinary ones. Sort them into rough buckets: hours and location, pricing, order status, returns, how-to, and everything else. Count them.
What you'll almost certainly find is that a handful of buckets make up the bulk of your volume. The same dozen questions, asked by different people in slightly different words. That repetition is the entire opportunity. A trained bot is very good at answering the same thing for the thousandth time and never getting tired or terse about it.
Write down two figures for each bucket: how many come in per week, and how many minutes a real answer takes, including the context-switch of stopping what you were doing. That second part matters more than people expect.
Put a believable number on it
Now build the estimate. Say a small support person fields 150 repeat questions a week, and each one takes about four minutes once you count reading, answering, and getting back on task. That's 600 minutes, or roughly ten hours a week, spent on questions that never change.
A trained bot won't catch all of them. Be conservative and assume it handles the clear repeats and forwards the rest. Even at a modest share, you're recovering a meaningful slice of that ten hours. The exact figure depends on your mix, so run it with your own numbers rather than borrowing mine. The method is what matters:
- Count repeat questions per week
- Estimate honest minutes per answer, including the interruption
- Multiply to get total weekly minutes on repeats
- Estimate the share the bot can handle cleanly
- Multiply that share by your weekly minutes for hours recovered
The point of writing it out is to avoid two traps: the vendor's rosy "80% deflection" and your own gut feeling that it "probably won't help much." Measured numbers beat both.
The costs nobody puts in the deck
Time saved is only half the equation. A chatbot also asks for time, and pretending otherwise is how people end up disappointed.
There's setup: pointing the bot at your content, testing it, fixing the awkward answers. There's ongoing review, especially early, where you skim transcripts and patch gaps. And there's the occasional cleanup when the bot confidently says something wrong because two of your pages disagreed.
None of these are dealbreakers, but they're real. A fair way to account for them is to subtract a few hours a week in month one, tapering to well under an hour by month three as the bot stabilizes. Your net savings is the gross hours recovered minus that upkeep. It's still a strong number for most teams. It's just an honest one.
The shape of the cost curve matters as much as its size. Setup and early review are front-loaded, so the first few weeks feel like more work, not less, and that's exactly when people give up and declare the whole thing a wash. Push through it. By the second month, the review time shrinks because you've already patched the common gaps, and the savings you projected start showing up for real. Judging a chatbot by week one is like judging a new hire by their first day of training. The point isn't the ramp, it's where you land after it.
The interruption tax is the hidden win
Here's the part the minutes-saved math undercounts. The cost of a support question isn't only the time to answer it. It's the focus it breaks.
When someone on a small team is deep in a project and a "quick question" pops up, answering it takes four minutes, but recovering their concentration takes far longer. Those interruptions stack up in a way no timesheet captures. A bot that fields the routine stuff means fewer taps on the shoulder, and the person doing focused work stays in it. For a founder juggling sales, product, and support at once, that protection is often worth more than the raw hours.
This is also why the "I'll just answer them myself, it only takes a minute" logic breaks down. The minute is real, but it's not the whole cost. It's the switching, the half-lost train of thought, the second time you have to find your place in the thing you were building. Ten one-minute interruptions don't cost ten minutes. They can cost a whole afternoon of the deep work that actually moves your business. Handing those off isn't laziness, it's protecting the hours that only you can spend well.
A worked example
Take Cedar & Pine, a fictional home goods store run by two people. Before automating, one of them, Dana, spent a chunk of every day on the same questions: is this in stock, when will it ship, do you gift wrap, what's your return policy. She estimated it at two hours a day, most of it in annoying five-minute chunks that wrecked her ability to do anything else.
They trained a bot on the product catalog, shipping policy, and FAQ. Here's a typical exchange it now handles without Dana:
Customer: is the walnut side table back in stock and can it ship before the 20th? Bot: The walnut side table is back in stock as of this week. Orders placed today ship within two business days, so it'd arrive well before the 20th for most US addresses. Want me to note gift wrap on the order?
Dana didn't get those two hours back to the minute. She spent about three hours a week the first month training and reviewing. But by month two, her days had far fewer interruptions, and she reclaimed most of an afternoon each week for actual merchandising work. In SpideyChat she keeps a weekly habit of skimming transcripts, which is where she catches the questions worth adding content for.
Run your own numbers before you commit
Don't take anyone's headline savings figure, including this post's. Spend one afternoon counting your repeat questions and timing a few honest answers. Multiply it out. Subtract a fair estimate for setup and review. What's left is your real number, specific to your business, and it's the only one worth deciding on. If it's a few hours a week net, that's a strong case. If it's marginal, you've saved yourself from automating a problem you didn't really have.