Before you keep paying for any tool, one test should decide it: the tool has to make or save more than it costs. Chatbots often get sold on impressions, so the useful thing is to put numbers to it. The arithmetic is simple. The care goes into being honest about both sides, and into measuring outcomes rather than activity.
This guide walks through the whole calculation: what to count, how to turn it into money, how to attribute sales without overclaiming, and a worked example you can copy with your own figures.
Measure outcomes, not activity
A dashboard that says "4,000 messages last month" feels like progress and proves nothing. A thousand conversations could be a thousand questions answered or a thousand confused visitors. Before you calculate anything, separate what the bot did from what the business got.
A quick filter for any number: if it doubled next month, would you do anything differently? Message volume could double and change nothing. Chat-influenced revenue doubling would change your plans. Keep the numbers that pass.
| Activity number | The outcome worth measuring instead |
|---|---|
| Conversations | Questions resolved, leads captured |
| Messages sent | Staff time no longer spent on repeat questions |
| Session length | Usually nothing on its own; long can mean stuck |
| "Engagement" | Sales where a chat answered a blocking question |
A chatbot creates value in three ways, and most businesses lean on one or two:
- Time saved: repeat questions your team no longer answers by hand.
- Leads captured: contact details from visitors who would otherwise have left anonymously.
- Sales assisted: purchases where a chat answered the question that was holding someone back.
A support-heavy software company gets most of its value from time saved. A shop or a service business gets more from leads and assisted sales. Know your mix before you measure, or a useful bot can look like a waste because you counted the wrong thing. Time saved is cost avoidance and the other two are revenue, so keep them on separate lines from the start.
Turn resolved questions into money saved
Start with how many conversations the bot handled without a person. In SpideyChat the dashboard splits conversations into resolved, handed off and still active, which gives you the raw count.
Next, work out what one human-handled question costs. Take a person's loaded hourly cost (wages plus overheads) and multiply by the minutes they spend per question. As an illustration only: at $30 an hour and six minutes per question, each one costs about $3. Use your own figures.
Then be strict about which resolved conversations count. Not every visitor whose question the bot answered would have emailed or phoned you instead. Some would simply have left. If you can't tell, count half and say so. A conservative estimate you can defend is worth more than a big one that falls apart on the first follow-up question.
money saved = resolved conversations that would have reached your team x cost per human-handled question
Cross-check against repeat contacts. A "resolved" chat followed an hour later by an email asking the same thing was deferred, not resolved, and shouldn't count. For a fuller look at where the hours come from, see how chatbots save your team hours every single day.
Put a number on leads and assisted sales
For leads, the sum is: leads captured, times the share that became customers, times your average sale. Use your real close rate, not your best month. Then apply your gross margin. Revenue isn't return; the profit on that revenue is.
To know which customers came through chat, mark it at the moment you record the lead. SpideyChat's Leads page shows each captured contact with the conversation that produced it, so you can see what they asked before you add them to your pipeline. SpideyChat doesn't sync with CRMs, so note "came through chat" yourself when you copy the lead into whatever CRM or spreadsheet you use. On Pro, a flow can also send collected details to your own systems by webhook.
For assisted sales, look at conversations on product and pricing pages that were followed by a purchase. You won't catch every one, so estimate on the low side. If you want ideas for raising the lead side of the sum, read how to generate more leads with a website chatbot.
Attribute revenue without overclaiming
A buyer might read a blog post, chat with the bot, get an email and buy a week later. Claiming the bot caused that sale is as wrong as saying it played no part. Two simple views help without pretending to precision:
- First touch: was chat the first meaningful contact? That shows its role in starting relationships.
- Last touch: was chat the final contact before the purchase? That shows its role in closing.
Look at both. A chat that keeps showing up as the last touch before a sale is doing closing work, even if it wasn't the only cause.
When you report the result, phrase it so it survives scrutiny: "chat was a touchpoint on deals worth this much, and here's how we tracked them." That is far more convincing than one large number. A finance lead who hears an inflated figure will start pulling threads, and once one gives way they stop trusting everything else you said. Don't credit the bot with sales that would have happened anyway.
Count the full cost
The cost side is where people fool themselves by counting only the subscription.
| Cost | What it includes |
|---|---|
| Subscription | The monthly or annual plan fee |
| Setup time | Hours spent adding content, testing answers and installing the widget |
| Upkeep time | Reading transcripts, filling answer gaps, updating prices and policies |
| Connection work | Any one-time effort to pass leads into your own tools |
The time costs are the ones that hide. Most of them come at the start: setup is a one-time push, and upkeep settles into a light monthly habit once the answers are good. Spread setup over a sensible period, such as six months, and cost all time at loaded hourly rates.
Run the calculation with a worked example
Over one consistent window, usually a month:
- Total the value: money saved plus gross profit on chat-influenced sales.
- Total the cost: subscription plus upkeep plus the month's share of setup.
- Subtract cost from value to get net gain.
- Divide net gain by total cost to get ROI.
- For payback, divide the setup cost by the monthly net gain to see how many months it takes to earn back.
A hypothetical example, with made-up figures to show the method. Meadowlark Print, an imaginary custom-printing shop, has 600 chat conversations in a month and 360 are resolved without a person. The owner reckons about half of those would otherwise have become an email or call, so 180 count. At $3 per question, that is $540 of time saved.
The bot also captured 30 leads. Three became customers at an average order of $400, which is $1,200 in revenue. At a 40% gross margin, that is $480 of profit.
On costs, the subscription is $30 a month and upkeep takes three hours at $30 an hour, so $90. Setup took eight hours ($240), spread over six months at $40 a month. Total monthly cost: $160.
Value is $540 plus $480, so $1,020. Net gain is $860. ROI is $860 divided by $160, about 5.4 times the cost, or 540%. Setup was earned back inside the first month. Even with the sales line removed entirely, the time saved still covers the cost more than three times over. That is the version worth presenting, because every input is written down and can be challenged. To try your own figures quickly, use the chatbot ROI calculator, and compare plan costs on the pricing page.
Judge it over months, not a week
The first weeks are skewed by setup effort and answers that are still being tuned, so ROI looks worse than it will settle at. Let it run for two or three months, log value and cost each month, and judge the trend.
A healthy bot shows costs early and value building as its answers improve. If value is still flat after a few months, investigate before you cancel. Usually the cause is content: the bot can't save time on questions it answers badly. SpideyChat's Answer gaps page lists the questions it handled poorly, which tells you exactly what to add.
Keep the numbers honest as you go:
- Use realistic close rates, not your best month ever.
- Keep counting upkeep time after setup ends.
- Read a sample of transcripts each week, because a high resolved count can hide visitors who gave up.
- Re-measure after any big change to your content, prices or traffic.
If the honest numbers show the bot paying for itself, you have a decision you can defend to anyone. If they don't, the same numbers show where it falls short, whether that's weak answers eating the time savings or leads that never get followed up. Either way you are deciding on evidence, which is the point of doing the sums.