Calculator
How much does running AI customer support actually save you?
Real numbers, not marketing math. Plug in your monthly tickets and your real human agent cost; see live savings, payback, and the per-ticket economics on whichever LLM you prefer.
Your numbers
Plug in your real values. Everything updates live.
Total inbound conversations a month, across all channels.
Each site is $3/site/month billed annually ($36/site/year); $5 if billed monthly.
Salary + benefits + tooling. US average is ~$65,000; lower in many other regions.
Mix of fast (1 minute) and slow (30 minute) tickets. 6 is a typical average.
Share of tickets the AI handles end-to-end. 60-75% is typical with a well-grounded knowledge base; higher if your queue is heavily FAQ.
Token cost on a typical 1,800-input / 220-output ticket. You pay the provider directly.
Estimated monthly savings
$20,826
That's $249,918 a year, roughly 70% cheaper than all-human support at the same ticket volume.
All-human (today)
$29,762
per month
5,000 tickets × $5.9524 per ticket.
AI + humans on OpenAgent
$8,935
per month
3,500 AI + 1,500 human tickets, plus $3.00 platform fee.
Show me the math
| Human cost per ticket | $5.9524 |
| All-human monthly cost | $29,762 |
| AI tickets (70%) | 3,500 |
| Token cost (AI tickets) | $3.85 |
| Remaining human tickets | 1,500 |
| Remaining human cost | $8,929 |
| Platform fee (1 site) | $3.00 |
| Monthly savings | $20,826 |
Assumes 1,820 productive hours per agent per year. Token costs are mid-2026 list prices for the provider you selected; your actual bill comes directly from the provider, not from OpenAgent. Containment rates above 80% are achievable but usually mean the escalation seam is too narrow, which hurts CSAT. We'd recommend 65-75% as the realistic ceiling for most teams.
Ready to flip the switch?
Sign up, paste a model key, drop in a knowledge document, paste the embed snippet. Ten minutes, four steps.
How the calculator works
We don't pull any numbers out of thin air. The formula:
- Human cost per ticket = (annual agent cost) ÷ (1,820 productive hours) ÷ (tickets per hour). 1,820 hours is roughly 35 productive hours per week × 52 weeks, which accounts for meetings, breaks, training, and paid time off.
- AI cost per ticket = published list-price token cost for a typical 1,800-input / 220-output support ticket on the provider you pick. You pay the provider directly; OpenAgent never marks up tokens. See BYO LLM, explained for the source math.
- Platform fee = $5 per site per month billed monthly, or $36 per site per year billed annually ($3 per site per month). The calculator uses the annual rate.
- Monthly savings = (all-human cost) − (AI cost + remaining human cost + platform fee).
The calculator is intentionally honest about the limits. AI containment rates above 80% are achievable but usually mean the escalation seam is too narrow, which hurts CSAT. For most teams, 65-75% is the realistic ceiling. The math also doesn't factor in the secondary savings from faster first-response times (CSAT compounds), or the secondary costs of running a worse experience if you over-rotate to AI.
Vendor cost comparison
The follow-up question: the same AI volume priced on Fin, Zendesk, Agentforce, and your own LLM keys.
The deep ROI guide
Per-ticket math, queue mix, the five rules for human-in-the-loop design. The article the calculator is built on.
BYO LLM, explained
Why hosted AI bills you $0.99/resolution and BYO LLM bills you $0.0011. The architecture difference.