AI vs human cost: what it really costs to automate customer support
“AI is cheaper” is only true above a certain volume. Here's how to work out where that line sits for your team — with every number shown.
On this page
Most “AI vs human” comparisons put a monthly software price next to a salary and declare a winner. That skips most of what an AI support system actually costs — and it hides the most useful number of all: the volume at which AI starts to pay.
This guide walks through a complete comparison for a support team, with every figure shown so you can swap in your own.
01The short answer
- AI support has fixed costs (setup and maintenance) and variable costs (usage per ticket, plus the tickets that still need a person).
- People have mostly variable costs — more tickets, more hours.
- So there's a breakeven volume. Below it, people are cheaper. Above it, AI is — and the gap widens with every extra ticket.
02Step 1: What the human version costs
Start with cost per ticket: minutes per ticket ÷ 60 × a loaded hourly cost — salary plus employer taxes, benefits, equipment, software and a share of management time. Multiply by monthly volume for the monthly cost.
Measure the minutes rather than guessing. Include reading the ticket, looking things up, replying and updating the system — not just typing the answer.
03Step 2: The four costs of the AI version
- 01Setup — design, build, integration with your help desk and order systems, and testing. A one-off cost.
- 02Usage — model and infrastructure cost per ticket handled.
- 03The remainder — tickets the AI can't or shouldn't handle still go to a person. This is the cost most estimates forget.
- 04Maintenance — keeping answers current, handling changes in connected systems, reviewing quality. A fixed monthly cost.
04A worked example
A hypothetical support team. Every number is an assumption to replace with your own.
| Input | Value |
|---|---|
| Tickets per month | 2,000 |
| Minutes per ticket | 8 |
| Loaded hourly cost | $28 |
| Resolved end-to-end by AI | 65% |
| AI usage cost per ticket | $0.35 |
| Setup (one-off) | $12,000 |
| Maintenance | $600 / month |
People: 2,000 × 8 ÷ 60 × $28 ≈ $7,467 a month — about $3.73 per ticket.
AI system: usage 2,000 × $0.35 = $700, plus the remaining 35% of the human cost ≈ $2,613, plus $600 maintenance ≈ $3,913 a month.
| Result | Value |
|---|---|
| Monthly saving | ≈ $3,553 |
| Breakeven | Month 4 ($12,000 ÷ $3,553 ≈ 3.4 months) |
| Net saving, year one | ≈ $30,640 |
| Net saving, two years | ≈ $73,280 |
05When people are cheaper
Run the same model at 200 tickets a month. People cost about $747. The AI system costs about $931 — because the $600 maintenance doesn't shrink with volume. At low volume, AI loses.
You can find the line directly. In this example each ticket costs $3.73 by hand and about $1.66 with AI (usage plus the share that still needs a person), so each ticket saves about $2.08. Covering $600 of maintenance takes roughly 290 tickets a month. Also paying back the $12,000 setup within two years ($500 a month) takes roughly 530 tickets a month.
06Where estimates go wrong
- Assuming too high an automation share. Measure your ticket mix first — what share are routine questions with a clear answer?
- Forgetting the remainder. The tickets that still need a person are a real cost.
- Ignoring maintenance. Policies, products and systems change; the assistant has to keep up.
- Measuring cost but not quality. A cheaper answer that's wrong creates a second ticket — or a lost customer.
07Where to start
- 01Export a month of tickets and tag them: routine, needs lookup, needs judgement.
- 02Time a sample of each type to get real minutes per ticket.
- 03Put the numbers into the AI vs human cost calculator and check you're above the breakeven volume.
- 04If the routine share is high, start with those tickets only — and keep a clear route to a person.
Not sure whether your tickets need an AI agent or simpler automation? Our guide to AI agents vs workflow automation covers the difference, and our AI chatbot development page shows how we build support assistants.