Before you pay for any tool, one question should decide it: does this make or save more than it costs? Chatbots get sold on vibes a lot of the time, so let's do the unglamorous thing and actually put numbers to it. The math isn't hard, and once you've done it you'll know whether your bot is an asset or a line item to cut.
The framework is straightforward: add up what the bot creates, subtract what it costs, and see what's left. The care goes into being honest about both sides. Here's how to run it for your own business.
Start with the three ways a bot creates value
A chatbot pays off through some mix of three things. Most businesses lean on one or two more than the others, so figure out which apply to you.
- Time saved: hours your team no longer spends answering repetitive questions
- Leads captured: contacts collected from visitors who'd otherwise have left anonymously
- Sales recovered: purchases that would have been lost to an unanswered question or a stalled checkout
A support-heavy business, like a software company fielding endless how-to questions, gets most of its value from time saved. An ecommerce store gets it from sales recovered and leads captured. Know your mix before you measure, because chasing the wrong number will make a genuinely useful bot look like a waste.
There's a fourth benefit that's real but hard to price, so keep it out of the formula and treat it as a bonus: the insight from transcripts. A month of chats tells you which pages confuse people and which objections keep costing you sales. Acting on that lowers your question volume and lifts conversion over time. You can't easily put a dollar figure on it, so don't try, but don't forget it exists when you're weighing whether to keep the tool.
Put a number on time saved
This is usually the easiest to estimate. Pull a week or two of your incoming questions and figure out what share the bot can handle: the repetitive, fact-based ones. Then estimate the time those used to take.
Say a support rep spends, illustratively, two hours a day on questions that have fixed answers. If the bot handles most of that, you're recovering roughly the better part of a workday each week. Multiply those hours by the rep's hourly cost and you have a monthly figure. Keep it conservative. Assume the bot handles most, not all, of the repetitive load, and don't count time the rep would've spent idle anyway.
The formula in plain terms:
(hours saved per week) x (loaded hourly cost) x (weeks per month) = monthly time-savings value
Put a number on leads and sales
For leads, track how many contacts the bot captures that you can attribute to it, then apply your usual close rate and average deal size. If the bot captures, say, twenty leads a month and you close one in ten at an average sale of a few hundred dollars, that's real revenue with a number attached.
For recovered sales, watch chats that end in a purchase, especially ones where the bot answered an objection at checkout or on a product page. You won't capture every recovered sale perfectly, so estimate on the low side. It's better to understate the value and be pleasantly surprised than to talk yourself into a tool that isn't earning its keep.
Now count the full cost, honestly
The cost side is where people fool themselves by only counting the subscription. There's more to it, and leaving it out makes the ROI look better than it is.
| Cost | What it includes |
|---|---|
| Subscription | Your monthly or annual plan fee |
| Setup time | Hours spent training the bot and testing answers |
| Upkeep time | Reviewing transcripts, updating content, fixing weak answers |
| Integration | Any one-time work to connect it to your CRM or inbox |
The subscription is usually the smallest honest surprise here; the time costs are the ones that hide. That said, they're mostly upfront. Setup is a one-time push, and upkeep settles into a light monthly habit once the bot is tuned. Add them up over the same period you measured value, and use loaded hourly rates for the time, not just wages.
Run the actual calculation
Put both sides together over a consistent window, say one month:
- Total the value: time saved + leads value + sales recovered.
- Total the cost: subscription + setup (prorated) + upkeep + integration (prorated).
- Subtract cost from value to get net gain.
- Divide net gain by total cost to get ROI as a ratio or percentage.
A worked example. Meadowlark Print, a small custom-printing shop, ran the numbers after two months on SpideyChat. Their bot saved an owner roughly six hours a week on repeat questions about turnaround times and file specs, captured a steady handful of quote requests, and recovered a few orders that used to stall on shipping questions. Against a modest monthly plan and a few hours of upkeep, the saved time alone more than covered the cost, and the captured quotes were profit on top. They didn't need a spreadsheet full of assumptions to see it was worth keeping.
Give it a fair timeframe
One week won't tell you much. The first stretch is skewed by setup effort and a bot that's still being tuned, so the ROI looks worse than the steady state. Let it run a couple of months, keep a simple log of the value and cost each month, and judge it on the trend.
Watch the shape of the curve, not just the final number. A healthy bot shows costs front-loaded and value building as it handles more and gets tuned. If two or three months in the value is still flat, that's a signal to investigate rather than to quietly keep paying. Usually the fix is content: the bot can't save time on questions it answers poorly, so weak answers show up as disappointing time-saved numbers. Tightening the content is almost always cheaper than concluding the whole idea doesn't work.
A few honest caveats to keep your numbers grounded:
- Don't credit the bot with sales that would've happened anyway
- Use realistic close rates, not your best month ever
- Count the upkeep time even after setup ends
- Re-measure after any big content or pricing change
If the honest math shows the bot paying for itself and then some, you have your answer, and it's a decision you can defend to anyone who asks. If it doesn't, the same numbers tell you where it's falling short, whether that's weak answers costing you the time savings or a missing handoff losing the leads. Either way, you're deciding on evidence instead of a hunch, which is the whole point of running the numbers in the first place.