Every AI chatbot vendor will tell you it pays for itself in customer-service savings. Rather than take anyone's word for it, this is a worked example: we take two pricing mechanics that are actually published — Intercom Fin's $0.99-per-resolution model and Callbell's flat per-agent fee — and run them through a hypothetical support team with every assumption stated up front. Treat the dollar figures below as an illustrative framework built for this piece on AI chatbot ROI, not a benchmark published by either vendor.
The exercise, and the assumptions behind it
Here's the setup. Intercom Fin charges a $49/month base fee that includes 50 resolved conversations (“outcomes”); every resolution beyond that is billed at $0.99 each. Callbell charges a flat $17 per agent per month, with the AI chatbot bundled in at no extra cost no matter how many conversations it handles. For this exercise we're pricing a small support team of 3 agents on Callbell — 3 × $17 = $51/month, flat, whether the bot resolves ten conversations that month or ten thousand. The team size is our own assumption for illustration purposes; resize it to your own headcount and the flat side of the math scales linearly. Bump that team to five agents and the flat side becomes $85/month; shrink it to two and it's $34/month — the shape of everything below stays the same, only that one number moves.
The variable that moves everything else is the resolution rate — the share of conversations the AI genuinely closes out without a human stepping in. Vendors like to lead with numbers in the 65-70% range. Real-world reporting across this category — including user reviews on sites like G2, and reviews of comparable tools on this site — tends to land closer to 40-60% once you look past the marketing page. For the cost table below we use 50% as a single working assumption to keep the arithmetic clean. For the time-saved section further down, we widen it back out to the full 40-60% range, since that's a fairer picture of where most deployments actually land.
One more assumption: we're using an average handling time (AHT) of 6 minutes per conversation for a human agent — a reasonable mid-range figure for a typical WhatsApp support exchange — to translate resolved conversations into hours of agent time freed up. To be fully transparent: nothing below is a number Intercom or Callbell has published. It's our own arithmetic, built openly on the pricing mechanics and resolution-rate ranges stated here, so you can swap in your own numbers and redo the exercise for your business.
Monthly cost at three volume tiers
With those assumptions fixed, here's what each pricing model costs as conversation volume grows. At a 50% resolution rate, half of total conversations get closed out by the AI and billed as Fin “outcomes” beyond the 50 already included in the base fee; Callbell's $51/month doesn't move at all, because it's tied to agent seats, not to how busy the chatbot is.
| Conversations/month | Resolved by AI (50% est.) | Intercom Fin — est. cost | Callbell — flat cost (3 agents) | Callbell saves |
|---|---|---|---|---|
| 500 | 250 | $247/mo | $51/mo | ~$196/mo |
| 2,000 | 1,000 | $989.50/mo | $51/mo | ~$938.50/mo |
| 5,000 | 2,500 | $2,474.50/mo | $51/mo | ~$2,423.50/mo |
The crossover happens fast. Under these assumptions, Intercom Fin and Callbell cost about the same around 100 conversations a month — call it 3 to 4 a day. Below that sliver of volume, Fin's $49 floor can actually undercut Callbell's $51. Above it, the gap opens quickly and never closes: by 500 conversations a month, Fin already costs roughly 4.8× what Callbell does; by 5,000, it's about 48.5× — because Callbell's flat fee never “finds out” the chatbot got busier.
Illustrative math only — assumes a 50% AI resolution rate and a 3-agent Callbell team ($51/month flat). Change either input and every number in this table moves.
How many customer-service hours you actually save
Cost isn't the only lens worth using — the hours an AI resolution takes off an agent's plate matter too, especially if you're trying to justify staffing decisions rather than software spend, a trade-off analysts like Gartner track closely across the conversational AI market. Using our 6-minute AHT assumption and the full 40-60% resolution-rate range (since that's genuinely where outcomes vary), here's roughly how many agent-hours a month get freed up at each volume tier, plus what that looks like in full-time-equivalent (FTE) terms using a 160-hour work month.
| Conversations/month | Resolved by AI (40–60%) | Agent hours saved/month | ≈ FTE freed up |
|---|---|---|---|
| 500 | 200–300 | 20–30 hrs | ~0.1–0.2 |
| 2,000 | 800–1,200 | 80–120 hrs | ~0.5–0.75 |
| 5,000 | 2,000–3,000 | 200–300 hrs | ~1.25–1.9 |
These hours are freed-up capacity, not “saved headcount” — most teams reinvest that time into faster replies, higher-touch conversations, or handling more volume with the same staff, rather than cutting the team.
The caveats this framework doesn't capture
Treat everything above as a starting framework, not a guarantee — the fast-growing AI software category that trackers like Business of Apps follow closely is still maturing. A few things move the real numbers around more than this simple model can show:
- Resolution rates vary a lot by vertical and complexity. A store fielding “where's my order?” questions resolves automatically far more often than a team handling technical troubleshooting or billing disputes — 50% is a reasonable planning midpoint, not a universal constant. Think “track my order” versus “why was I charged twice and can you also change my shipping address” — the first resolves itself in seconds; the second usually needs a person.
- Per-resolution pricing isn't automatically the wrong choice. At genuinely low or unpredictable volume — a seasonal business, or a support line that's just getting started — paying only for what gets resolved can cost less than committing to fixed agent seats, exactly like the sub-100-conversation window in the table above, a pattern also visible when comparing plans on marketplaces like Capterra.
- “Resolved” doesn't mean the same thing to every vendor. Before comparing any autonomous-resolution percentage across tools, it's worth understanding what actually counts as a resolution — we go into that in detail in our piece on what “the bot resolves it itself” really means.
Use this as a framework for the conversation, not the final word: plug in your own team size, your own AHT, and a resolution-rate estimate grounded in your actual use case, and the same math will tell you something far more useful than either vendor's marketing page.
The bottom line
At meaningful, growing volume, flat per-agent pricing tends to look better the more successful the chatbot gets, since no line item grows alongside it — the $51/month stays $51/month whether the bot resolves 200 conversations or 2,000. Per-resolution pricing runs the opposite way: the better the automation performs, the higher the bill climbs, because you're billed for every success. That's not a knock on outcome-based pricing in general — it can be the smarter bet at low or unpredictable volume — but for a team expecting the chatbot to actually carry a meaningful share of conversations, a flat per-agent model means your best-case scenario doesn't come with a matching best-case invoice.
If you're building the business case internally, run these numbers with your own assumptions before you present them — the framework matters more than any single figure in this article. Ask your provider for their own resolution-rate data broken down by conversation type, not just a single blended headline percentage — that's usually where the real planning number hides.