Anyone who runs customer service sooner or later gets the question from management: what do we save with AI? The question is reasonable, but it is often asked with an expectation of fewer staff, and that expectation is the first thing a good calculation should remove. This article shows what a calculation that holds looks like, where the money actually is, what usually gets forgotten and what you can promise.
What does a ticket cost today, and what does it cost with AI?
The starting point is the cost per resolved ticket in your staffed channels, and it is almost always higher than management thinks. The most cited reference comes from a Gartner survey of customer service leaders in 2019: staffed channels such as phone, live chat and email cost on average 8.01 dollars per contact, while self-service on website and in app cost around 0.10 dollars. The figures are American and seven years old, but the order of magnitude holds: the difference between a ticket a person handles and one the customer resolves themselves is manifold.
Work out your own figure: customer service's total cost per month, including salaries, systems and premises, divided by the number of resolved tickets. That number is what you compare everything else with.
The second number is what AI costs per ticket, and it is not zero. Gartner predicts that the cost per resolution with generative AI will exceed three dollars by 2030, more than many offshore contact centres charge for a human agent. Gartner's conclusion is that full automation becomes too expensive for most, and that AI should primarily be used to improve the customer experience, not to cut costs. The calculation should therefore include a real cost per AI ticket, and allow for it rising.
Where is the saving really?
In fewer contacts and shorter handling, in that order. McKinsey describes that a completed transition to AI-enabled customer service can double to triple the use of self-service, reduce the number of service interactions by 40 to 50 percent and lower the cost to serve by more than 20 percent. It is contacts that never need a person, such as order status, return terms and opening hours, that produce the big numbers.
The second source is time per ticket for the tickets that still reach a person: AI drafts the agent reviews and sends, lookups of orders and history, summaries of long threads. That gives shorter handling and more even quality, but less than the removed contacts.
What is not in the calculation in the first year is headcount reduction. According to Gartner only 20 percent of customer service leaders had reduced headcount because of AI by the end of 2025, and Gartner predicts that half of the organisations that planned large reductions will abandon the plans by 2027; in the same survey 95 percent said they intend to keep human agents. A calculation built on fewer staff from day one will not hold.
Which costs get forgotten in the calculation?
The ones that are not on the supplier's quote: the knowledge base, the review, the integrations and the licence per ticket. This is what they look like:
| Cost | What it consists of | Common mistake |
|---|---|---|
| The knowledge base | Writing and reviewing the answers the AI will use, often 20 to 50 articles before anything works | Counted as "already done" because an FAQ exists |
| Knowledge manager | Time every week to prioritise, review and prune, as we describe in the article on the knowledge manager role | Put on top of a full queue and never happens |
| Review | The agents' time to read and change AI drafts in the first months | Counted as zero because it is "just a click" |
| Integrations | Connection to order system, e-commerce platform and carrier so the AI can look up instead of guess | Postponed, and the bot can only answer in general terms |
| Licence per ticket | Price per AI answer, per conversation or per token, growing with volume | Calculated on today's volume and today's price |
| Wrong answers | Tickets that come back, compensation, refunds the bot promised | Not counted at all |
A bot that promises free returns to ten customers a day costs more than the licence, and that is why the calculation should assume AI that answers from reviewed knowledge, with handover to a person when the source material is missing.
What does a simple calculation look like for a team with 300 tickets a week?
Like a comparison of cost per resolved ticket before and after, with three variables you set yourselves: the share of tickets that disappear into self-service, the share that get an AI draft, and the time saved per draft. The worked example below uses assumed values; replace them with your own.
- Current state. 300 tickets a week, two full-time agents. Assume a total cost of 90,000 kronor a month and 1,300 resolved tickets. Cost per ticket: about 69 kronor.
- Fewer contacts. Assume 25 percent of tickets are pure lookups that self-service and chat take over after three months: 325 tickets a month that no longer reach a person.
- Shorter handling. Assume half of the remaining tickets get an AI draft that saves three minutes each: about 24 hours a month.
- New costs. Assume a licence of 8,000 kronor a month, one day a week for a knowledge manager (equivalent to about 15,000 kronor) and 2 kronor per AI-handled ticket.
- Result. The 24 hours and the 325 removed contacts correspond to roughly a third of a post. In money, that covers the new costs with a small margin in the first year. What makes the calculation positive is what you do with the time: longer opening hours without hiring, or not having to bring in an extra person in November.
A calculation that only compares licence against salary looks good and does not hold. A calculation that counts cost per resolved ticket, including knowledge work and review, looks modest and holds. What a setup consists of is on the pricing page, and how it differs from a price list per agent in the comparison of Supportifier and Zendesk.
How do you measure that the calculation holds?
Through four metrics you measure before the start and then every month. They are the same metrics that steer customer service in general, described in the article on metrics, but here they are read against the calculation:
- Tickets per hundred orders. Should fall as self-service takes over. If it does not fall, no contacts disappeared, whatever the bot reports.
- Share of AI drafts sent without changes. Should rise per ticket type; when it is high the source material is right and the time saving real. How the metric is used to decide what can be automated is covered in the article on AI drafts.
- Handling time per ticket. Should fall for ticket types with drafts, not necessarily in total, since the simple tickets have gone and the hard ones remain.
- Customer satisfaction per channel. Must not fall. A CSAT that drops after automation is the calculation's hidden cost showing itself.
What should you promise management?
Fewer contacts, shorter response time and the same customer satisfaction within two quarters, measured with your own numbers, and not a headcount reduction. According to Gartner's forecast on technology spend, more than half of customer service organisations will double their technology spend by 2028 without a corresponding reduction in staff. Promise what you can measure: the share of tickets resolved without a person, the response time, customer satisfaction and cost per resolved ticket. Promise it per quarter, and promise to stop if the metrics do not move.
What to do
- Work out the cost per resolved ticket today. Total cost per month divided by resolved tickets, per channel if you can.
- Tag a week's tickets into three piles: can be resolved in self-service, can get a reviewed draft, needs a person. The shares are the calculation's most important variables.
- List the new costs from the table above, with your own estimate of time for knowledge base and review. Include a real cost per AI ticket and assume it rises.
- Build the calculation as cost per resolved ticket before and after, with a cautious and a likely scenario. Show both to management.
- Set four metrics with baseline values before you start, and decide when you read them.
- Start where the calculation is safest: the twenty most common questions in self-service and reviewed drafts in the inbox. In Supportifier you see per ticket type how many are resolved without a person and how many drafts are sent unchanged.
Common questions
How quickly does AI in customer service pay off?
Allow two quarters before the saving shows in the cost per resolved ticket, and expect the first three months to cost more than they give. The knowledge base has to be written, the agents review drafts that do not yet hit the mark, and self-service needs time to take over.
Should we count on fewer staff?
Not in the first year. Gartner reports that only 20 percent of customer service leaders have reduced headcount because of AI, and predicts that half of those who planned large reductions will abandon the plans. Count instead on what the freed time is used for: longer opening hours, a shorter queue, knowledge work or not having to bring in extra staff in peak season. If a post becomes surplus in the long run, it happens through attrition, and it should be in the calculation as a possibility, not as the foundation.
What does AI cost per ticket?
It depends on the pricing model: per conversation, per answer or per token, and the price moves. Gartner predicts that the cost per resolution with generative AI will exceed three dollars by 2030. Ask each supplier for a price per ticket at your volume today and at double the volume, and put the rising number into the calculation.
Can we start without integrations?
Yes, with the tickets that have the same answer for every customer: terms, delivery times, how a return works. They only require a reviewed knowledge base. Tickets that require lookups, such as order status and invoice questions, cannot be resolved by the AI without a connection to your systems, and there the calculation should count the integration as a cost and those tickets as a later gain.
Sources
- Gartner Says Only 9% of Customers Report Solving Their Issues Completely via Self-Service — Gartner, 2019
- Gartner Predicts GenAI Cost Per Resolution for Customer Service Will Exceed Offshore Human Agent Costs by 2030 — Gartner, 2026
- The next frontier of customer engagement: AI-enabled customer service — McKinsey, 2023
- Gartner Survey Finds Only 20% of Customer Service Leaders Report AI-Driven Headcount Reduction — Gartner, 2025
- Gartner Predicts 50% of Organizations Will Abandon Plans to Reduce Customer Service Workforce Due to AI — Gartner, 2025
- Gartner Predicts Over 50% of Customer Service Organizations Will Double Their Technology Spend By 2028 — Gartner, 2026
Martin Carlsson
Rickard Collander