Customer service AI ROI: how to calculate it in e-commerce

The ROI of customer service software is calculated by comparing the value it delivers (agent hours freed up, tickets handled without a human, customers retained) with its full cost over 12 months: ROI = (gains − costs) ÷ costs. In e-commerce, the most reliable calculation starts from the real cost of a written ticket handled by your team, then applies conservative automation and productivity assumptions, contact reason by contact reason.
This guide gives you the step-by-step method, a worked example for an online store, the metrics to record before and after, and how we build this calculation with brands at Klark.

Key takeaways
To calculate the ROI of customer service AI, start from three numbers you already have: the volume of written tickets per month, the number of agents handling them and the share of their time spent on written channels. From these you get a cost per ticket. The direct gain comes from tickets resolved automatically and from time saved on the ones that stay with agents; the indirect gain comes from satisfaction and repeat purchases. Present the direct gain as the basis of your business case, and the indirect gain separately, with deliberately low assumptions.
What is the ROI of customer service software?
ROI (return on investment) measures what a tool brings in compared with what it costs. The formula is the same for a helpdesk, a chatbot or an AI agent: ROI = (gains over the period − full cost over the period) ÷ full cost × 100.
The hard part is the gains. Some are measured in hours and euros (tickets absorbed, reduced handling time). Others are measured in satisfaction, customer reviews and repeat orders. Mixing both into a single number makes your business case fragile in front of a finance team.
How to calculate the ROI of customer service AI, step by step
Step 1: calculate the current cost of a written ticket
Take the annual fully loaded cost of your agents, multiply it by the share of their time spent on written channels (email, forms, chat), then divide by the number of written tickets handled over the year. This cost per ticket is the basis for everything else.
The share of time spent on written channels is often underestimated or overestimated. In a recent meeting, a customer service manager put it this way: “Depending on the day, we can easily be at 60% on the phone and the rest on tickets.” In that case, only 40% of the team's cost goes into the calculation.
Step 2: sort tickets by contact reason
ROI isn't calculated on "tickets" in general. It's calculated contact reason by contact reason: order tracking, returns, refunds, address changes, product questions, complaints. Repetitive reasons, which require order data and a clear rule, get automated first. Sensitive complaints stay with agents, with a reply prepared by the AI.
Step 3: estimate the direct gain
The direct gain has two sources: tickets resolved with no human involvement, and time saved on the tickets agents handle with a reply suggested by a copilot. You add them up without counting anything twice: the productivity gain only applies to the remaining tickets.
Step 4: add up the full cost
The full cost includes the subscription or usage-based billing, internal setup time (helpdesk connection, rules, testing), supervision time in the first weeks and, where applicable, connectors. The pricing model changes the calculation a lot: per agent, per ticket handled, per conversation or per resolution. Check which unit the invoice is based on before comparing two offers.
Step 5: calculate ROI and payback period
ROI = (direct gain − full cost) ÷ full cost. The payback period is calculated as follows: full cost ÷ monthly gain. Run the calculation with two scenarios, conservative and baseline, rather than a single figure.
What does this calculation look like for an online store?
Here is a fictional example, built on assumptions you will replace with your own numbers. A brand receives 3,000 written tickets per month, or 36,000 per year. Four agents handle them full time on written channels. The annual fully loaded cost of an agent is set at €34,000 (assumption).
- Annual cost of written support: 4 × €34,000 = €136,000
- Cost per written ticket: €136,000 ÷ 36,000 = €3.78
| Item | Conservative scenario | Baseline scenario |
|---|---|---|
| Share of tickets resolved automatically | 25% | 43% |
| Tickets automated per year | 9,000 | 15,480 |
| Gain on automated tickets | €34,020 | €58,514 |
| Productivity gain on remaining tickets | +20% | +50% |
| Gain on remaining tickets | €17,010 | €25,855 |
| Annual direct gain | €51,030 | €84,369 |
| Agent equivalent | 1.5 FTE | 2.5 FTE |
The baseline scenario uses the averages observed at brands that use Klark: 43% of tickets automated and +50% productivity. These are averages: your result depends on your contact reasons, the quality of your order data and your reply history.
Let's take a full cost of €20,000 for the first year. This amount is arbitrary: it illustrates the formula and does not correspond to any actual pricing. The ROI of the conservative scenario is then (51,030 − 20,000) ÷ 20,000 = 155%, with payback in just under 5 months. Another useful way to read it: in the conservative scenario, the tool stays profitable as long as its annual full cost stays below €51,000.
One point to make to your leadership: 1.5 FTE freed up doesn't mean 1.5 positions cut. In most e-commerce teams, that time goes to absorbing growth and the peaks of sales seasons and Black Friday, or to reopening the phone line in the afternoon.
How do you put a number on the indirect gain without inflating it?
The indirect gain comes from customers who get a faster, more accurate answer and who recommend you instead of leaving. You estimate it with four data points: the number of customers who contact customer service each year, the share who would go from dissatisfied to satisfied, the annual value of a customer and the share of those customers who recommend you.
This calculation relies on strong assumptions. We recommend presenting it separately, using the lowest customer value your team will accept, and never adding it to the direct gain in the headline figure. A low review rating on a public platform is a good starting point: it makes the cost of slow customer service visible.
To tie this gain to actual measurements, track your CSAT separately for tickets handled by AI and tickets handled by agents. Our guide to measuring customer satisfaction covers the methods in detail.
Which metrics should you track before and after rollout?
Record these metrics over the 3 months before launch, then every month afterward. Without a baseline, no ROI holds up.
| Metric | Formula | What it says about ROI |
|---|---|---|
| Cost per written ticket | Cost of written support ÷ written tickets | The basis of the direct gain |
| Automated resolution rate | Tickets closed without an agent ÷ total | The volume actually absorbed |
| Reopen rate | Reopened tickets ÷ closed tickets | A gain that comes back as a ticket isn't one |
| Average handling time | Agent time ÷ tickets handled | The copilot's gain on remaining tickets |
| First response time | Average time before the first reply | The link with satisfaction |
| CSAT by handling channel | Satisfied customers ÷ respondents | The perceived quality of automated replies |
| Backlog | Tickets open at end of day | The capacity regained during peak periods |
Measure resolution, not just "deflection": a customer redirected to an FAQ who comes back by email the next day costs you twice.
How does Klark build this calculation with brands?
In meetings, James Rebours, Klark's co-founder, runs the calculation live with the team, using their own numbers rather than a standard template. He always separates two blocks: direct ROI, expressed in FTEs freed up on written support, and indirect ROI, tied to satisfaction and retention. The first is the basis of the business case; the second shows the order of magnitude of what's at stake, with assumptions the brand can lower itself.
The questions are always the same: how many tickets per month, how many people, what fully loaded cost, and what share of time goes to written channels versus the phone. One customer service manager summed up what she expected from the time freed up: “On quite a lot of topics, the answers are pretty basic. It's a shame to spend time on that when a tool can do it.”
Klark uses your ticket history to identify recurring contact reasons and how your team handles them. More than 150 brands use it. If you want to understand how it all works before running the numbers, read our guide to AI for customer service.
Which mistakes skew the ROI calculation?
- Counting the whole team's cost when part of their time goes to the phone. The gain is calculated on the written share.
- Adding automation and productivity on the same tickets. The copilot's gain only applies to the remaining tickets.
- Forgetting internal time for setup and supervision in the first weeks.
- Using a brochure automation rate. Test on your own past tickets before signing.
- Ignoring changes in unit cost. Gartner predicts that the cost per resolution of generative AI will exceed $3 by 2030 (press release of January 26, 2026). Check how your contract changes if your volumes double.
- Stopping measurement after launch. Contact reasons change with your catalog, carriers and seasons.
To compare solutions on concrete criteria, see our comparison of AI customer service software for e-commerce.
FAQ
What is the formula for customer service software ROI?
ROI = (gains over the period − full cost over the period) ÷ full cost × 100. Gains include tickets resolved without an agent and time saved on the others. The full cost includes the subscription, usage, setup and supervision.
How do you calculate the cost of a customer service ticket?
Multiply the annual fully loaded cost of your agents by the share of their time spent on written channels, then divide by the number of written tickets handled over the year. With 4 agents at €34,000 and 36,000 tickets per year, you get about €3.78 per ticket.
How long does it take for customer service AI to pay off?
It depends on volume, contact reasons and the cost of the tool. In this guide's fictional example, the conservative scenario pays back in just under 5 months. Calculate the payback period with your own numbers and two scenarios.
Should customer satisfaction be included in ROI?
Yes, but separately. The gain from satisfaction and repeat purchases relies on strong assumptions. Present it alongside the direct gain, with low values, rather than adding it to the headline figure.
Which e-commerce tickets pay off fastest?
Repetitive contact reasons that require order data: package tracking, returns, refunds, address changes before shipping. They are frequent, follow clear rules and are easy to verify.
Want to run this calculation on your own tickets? Request a demo: we'll go through your volumes and contact reasons with you.







