If you have ever tried to justify an AI customer support agent on a spreadsheet, you have probably hit the same wall: which number actually proves it?
Cost per ticket is the obvious number. It is also the number that sells the story shortest. Because “what is the ROI?” is a trick question. The answer depends entirely on who you ask.
A founder sees money, in two directions. A CX lead sees the customer’s experience. An operations manager sees control over a queue that used to be chaos. The same agent, the same week, the same tickets, and yet three completely different kinds of return. The mistake most teams make is to let one of those people own “the ROI number”, when the real return is the sum of all three.
In this article we do three things. First we explain which numbers each role cares about. Then we work through the seven metrics that matter, including how they are calculated. Finally we bring the impact of each metric together into one concrete picture of what it delivers.
Which numbers does each role in the business look at?
Before you talk about return, you need to know who is watching. Three roles steer customer support, each with its own goal and its own set of numbers.
| Role | The goal | The metrics they weigh most |
|---|---|---|
| Founder | Profitable growth. Support from cost centre to margin driver. | Cost saving, conversion rate, revenue from support |
| CX lead | Happy customers helped quickly and correctly. | CSAT, first response time, resolution time |
| Operations manager | Predictable capacity that scales without extra hires. | Resolution rate, automation rate, stability of resolution time |
The same agent serves all three of these goals at once. That is why a single headline number does not work: at best it measures the slice one role looks at.
Which metrics matter and how do you calculate them?
A number is only useful once you know how it is produced. These are the seven metrics on which you judge your AI customer support, with the calculation behind each.
-
CSAT (customer satisfaction). The average satisfaction score. Take the sum of all completed survey scores (1 to 5) divided by the number of answered surveys. Only completed surveys count.
-
First response time. The median time between the customer’s first message and the first reply. The median, not the mean, so outliers do not distort the picture.
-
Resolution time. The median time between the customer’s first message and the moment the ticket closes. Spam tickets are excluded.
-
Conversion rate. The share of conversations that lead to an order within three days. Take orders attributed to the agent divided by the total number of conversations, times one hundred.
-
Cost saving. The number of automated tickets times the difference between the human cost and the AI cost per ticket. In this article we use €4 per human ticket.
-
Resolution rate. The share of AI-handled tickets fully resolved without escalation to a human. Take the resolved AI interactions divided by the total number of AI-handled tickets, times one hundred.
-
Automation rate. The share of all conversations handled fully without a human, divided by the total workload (all automated plus all human tickets).
What is the impact per metric?
This is what those metrics look like when you put the Engaige AI agent next to the team average of employees. The figures below come from our own internal data across multiple customers, aggregated and anonymised.
Engaige AI agent versus human
The Engaige AI agent against the team average of employees, per metric.
- First response time lower is better ≈ 180× faster
- Resolution time lower is better ≈ 150× faster
- Resolved first time higher is better 3.5× more often
- CSAT (1–5) higher is better Equal or better
- Conversion rate higher is better over 2× higher
The pattern is consistent. On speed the difference is not incremental but of a different order: seconds against hours, minutes against days. On resolving power the agent sits well over three times higher, and on conversion at least double. And on customer satisfaction, where you would expect to pay the price of speed, the agent stays equal to or better than the team. Faster is only a win if it is not worse, and it is not.
What does this add up to?
Numbers on their own convince no one. The story emerges when you bring them together against the goal of an online store. The same agent then delivers three kinds of return at once.
More revenue on the sales side
The agent converts support conversations into orders at 2.72%, against 0.5% to 1.2% for the employees on the same stores. In this data the agent drove 63% of all support-driven revenue, while handling less than a quarter of the new tickets. A team that answers questions becomes a sales channel too.
Happier customers
A first response within about 40 seconds instead of hours, and a resolution within roughly 15 minutes instead of days. All of it with a CSAT that is equal to or higher than the team average. A customer who gets a correct answer in 40 seconds at 2am, with the same satisfaction as from your best employee during the day, is a customer who comes back.
More margin on the support function
This is the part a founder loses sleep over, in the good sense: this is where the support function tips from cost centre to margin driver. It comes down to two numbers that measure the same thing. A human ticket costs about €4 to handle. And a support employee in the Netherlands earns around €40,000 a year, which with social charges comes to about €48,000, or roughly €4,000 a month.
Those two figures measure the same thing: the price of human handling, expressed once per month and once per ticket. We take them as the starting point. Divide your €4,000 monthly wage cost by €4 per ticket and you arrive at roughly 1,000 tickets a month. That is not independent proof, but the arithmetic result of those two assumptions.
An AI agent that resolves around 1,000 tickets a month therefore absorbs roughly one full-time employee. That is your break-even line. Below it you prove the model, above it every resolved ticket is margin that flows straight back into your support budget. In the data above the agents together resolved around 2,900 tickets a month autonomously: at €4 per ticket that is about €140,000 a year in avoided human handling cost (gross), the equivalent of almost three full-time employees. Net, after the AI cost of around €0.85 per ticket, about €110,000 of that remains. Your team does not shrink, it shifts: the same people handle the complex, high-value tickets while the volume of repeat questions falls away. That an AI agent does not replace a team but redistributes the work we work through separately.
In summary:
| Metric | Change | Business effect |
|---|---|---|
| Conversion rate | 0.5–1.2% → 2.72% | More revenue from existing traffic |
| CSAT | Equal or higher | Higher retention and repeat purchases |
| First response and resolution time | Hours → seconds, days → minutes | Better experience without extra effort |
| Cost saving | ± 2,900 tickets/month absorbed | ± €140,000 a year in capacity |
One agent, one goal, three kinds of return. Steer on cost per ticket alone and you see only one of the three and badly underestimate the real value. The honest ROI question is not “how much did we save”, but “what did we save, earn, speed up and stabilise at the same time”.
Why these numbers belong to Engaige
These numbers do not come from a chatbot that generates answers, but from an agent that acts. Engaige is the AI layer on top of your existing customer service software: it reads your operational stack (order, warehouse, payment, carrier), interprets the specific carrier events behind a WISMO question, applies your own policy through the AI Manager, and performs the action itself: process a refund, start a return, change an address. That is why resolution and conversion sit high, and why the margin works out: at a fixed monthly price and around €0.85 AI cost per ticket, most of the saved €4 per ticket stays put.
It scales beyond these figures too. HelloPrint automated 70% of its support and brought its support team from 100 to 28 agents, while service quality went up and revenue grew 30% a year. Read the HelloPrint case study. And at Otrium the agent handles 65% of the 120,000 annual tickets autonomously.
Who do you need to convince, and with what?
You sell the business case for AI customer support differently to each role. Use this matrix to decide in advance which number you put in front of whom.
| To convince… | They value most | Show this |
|---|---|---|
| Founder / CEO | Profitable growth and support that pays for itself | ± €140,000 a year in avoided handling cost (gross, net ~€110,000) plus 63% of support-driven revenue: support adds to margin instead of pulling on it |
| CFO / finance | Predictable cost and a clear payback | €4 per ticket times volume, break-even at ± 1,000 tickets a month (one FTE), and a fixed monthly price instead of a cost per resolved conversation |
| CX lead / head of support | Happy customers without losing quality | First response time from hours to seconds, resolution time from days to minutes, with a CSAT equal to or higher than the team average |
| Operations manager | Predictable capacity that scales | Resolution rate of ± 70%, a flat resolution time regardless of volume, and scaling without extra hires |
Frequently asked questions
How do you calculate the ROI of AI customer support?
Look beyond cost per ticket. Add three things: the cost saving (resolved tickets times your cost per human ticket), the extra revenue from converting support conversations, and the value of a better customer experience (faster response and resolution time at equal or higher CSAT).
What is the break-even point of an AI agent?
Around 1,000 resolved tickets a month. At €4 per human ticket that equals roughly one full-time employee: €40,000 salary, about €48,000 including social charges, or some €4,000 a month.
Does speed come at the cost of customer satisfaction?
No. In the figures the AI agent’s CSAT stays equal to or higher than the team average, while first response time drops from hours to seconds and resolution time from days to minutes.
Which metrics should I track?
Seven: CSAT, first response time, resolution time, conversion rate, cost saving, resolution rate and automation rate. Together they cover the three kinds of return: revenue, customer experience and margin.
Does an AI agent only save costs or does it earn revenue too?
Both. Alongside the saved handling costs, the agent converts support conversations into orders. In the figures the agent drove 63% of all support-driven revenue.
