To reduce customer support costs with AI, move the transactional tickets, which McKinsey finds are still 50 to 60 percent of interactions, onto an AI agent that resolves them end to end, pay only for tickets it actually resolves, and use the hours it frees to avoid hiring rather than to absorb rework. On a 10,000-contact queue at Gartner's $8.01 per live contact, resolving half the volume at $0.90 per resolution on a $5,100 monthly plan cuts monthly cost by about 35%. The eight levers below are ordered by how much cost they usually remove, and the last sections show the break-even math and where AI adds cost instead.

Key takeaways

• Resolution removes cost; answering does not. Only 14% of service issues are fully resolved in self-service, per Gartner, so a tool that answers without acting leaves the expensive part on your queue.

• Break-even comes early at volume. At $8.01 avoided per contact, a $5,100 plan at $0.90 per resolution pays for itself at about 720 resolutions a month.

• Bad escalations raise cost. A customer who repeats themselves to a person after a failed AI conversation is a more expensive ticket than one that went straight to a person.

• Contract the rate. Gartner predicts GenAI cost per resolution will exceed $3 by 2030, so lock per-resolution prices and escalation billing terms.

Which tools help reduce support costs?

AI agent platforms that price per outcome make the savings checkable before you sign. Lorikeet resolves chat, email, SMS and voice tickets at a published $0.99 or $0.90 per digital resolution on a $2,100 or $5,100 monthly plan, with escalations to a person and unresolved tickets not charged. Alternatives:

• Fin (now part of Salesforce) charges $0.99 per outcome on an existing helpdesk, with no setup or platform fees.

• Zendesk includes AI agents in every Suite and Support plan and bills per automated resolution, which suits teams already on Zendesk.

• Salesforce Agentforce offers $2 per conversation, Flex Credits or per-user licences for Salesforce shops.

How we chose and checked these levers

We included a lever only if a public source shows it moving cost or volume, and we ordered the list by how much of a typical support budget each one touches. Labour figures come from the US Bureau of Labor Statistics; vendor prices and capabilities come from each vendor's own pages as published on 30 October 2026; customer results are quoted only as printed on the customer's published story. Lorikeet publishes this guide and is one of the tools listed. We have tried to describe every product, including our own, using only what each vendor publishes, and we link to the source for each claim.

1. Resolve transactional tickets end to end

End-to-end resolution is the biggest lever because it removes the whole ticket, not part of it. McKinsey's analysis of millions of interactions across more than 30 organizations found 50 to 60 percent remain transactional, around 50 percent of call volume at one European bank. Resolution here means the agent verifies the customer, reads the account, applies the policy, takes the action and confirms the outcome. A UK car finance lender shows the gap between answering and acting: Carmoola moved from about 30% resolved with its earlier automation to 40% end to end from day one with Lorikeet, and now 60% of all inbound on WhatsApp, chat and email. Measure this lever with resolution rate, not deflection; the difference is explained in resolution rate vs deflection rate.

2. Triage and route every ticket automatically

Automated triage cuts the minutes humans spend reading, tagging and reassigning tickets, and it sends the right tickets to the AI agent in the first place. Misrouted tickets are paid for twice: once by the wrong agent and again by the right one. Amber reports 99% accurate ticket routing, and Eucalyptus routes every ticket by topic, tone and complexity and lifted CSAT by 10 points. Triage is also the cheapest AI line to pilot: Lorikeet lists routing or analytics tagging at $0.30 or $0.27 a ticket. Start here if your agents spend the first minutes of each ticket working out what it is.

3. Prevent the repeat contact

Every second contact about the same issue is a full-price ticket, so reducing repeats cuts cost without touching handle time. Klarna reported a 25% drop in repeat inquiries after launching its AI assistant, alongside two-thirds of chats handled in the first month. Gartner's 2019 poll found only 9% of customers resolve their issue completely through self-service, and the rest switch channels, which is how one problem becomes three tickets. Track reopen and recontact rates on AI-handled tickets from week one; a rising reopen rate means the AI is closing tickets it did not fix.

4. Automate QA on every conversation

Automated QA replaces manual sampling, which usually reviews a small fraction of tickets, with a score on every conversation, human or AI. The saving is partly the reviewer hours and partly the drift it catches before it turns into repeat contacts. Lorikeet's Coach scores conversations and proposes fixes to workflows and knowledge; automated QA is listed at $0.30 or $0.27 a ticket. Whatever tool you use, price QA as a per-ticket line in the cost model rather than treating it as free.

5. Move voice and IVR to an agent that can act

Phone is usually the most expensive channel because one agent handles one call at a time, so voice automation has a high value per ticket. The test is whether the voice agent can take actions. McKinsey cites an energy company that cut billing call volume by around 20 percent and shaved up to 60 seconds off authentication by wiring an AI voice assistant into its back-end workflow. An IVR that only routes still leaves the call on the human queue. Wonderschool's published story shows the shape on a small line: answer rate moved from about 10% to 100% in the first month, with 41% of genuine parent calls fully resolved by Lorikeet. Lorikeet Voice runs on local numbers in the US, UK and Australia, so teams elsewhere need another option.

6. Assist the humans who remain

Assist tools lower the cost of the tickets AI cannot take. In the NBER study of 5,179 support agents, an AI assistant raised issues resolved per hour by 14% on average and 34% for novices. McKinsey reports 20 to 30 percent shorter time to proficiency from simulation-led onboarding. Drafted replies, summaries and suggested next steps matter more after automation, because the remaining tickets are harder.

7. Re-forecast staffing and absorb peaks without temporary hiring

Staffing is where AI savings become cash, and it is the lever most often skipped. Once AI resolves a share of volume, re-run the forecast by interval and channel, then apply shrinkage last; Call Centre Helper notes most contact centres run 30 to 35% shrinkage and divides by it rather than adding it. Peaks are the clearest win. Flex uses AI to handle 4x chat spikes during rent week, and seasonal patterns are covered in handling seasonal support spikes. Present the saving as hires and temp contracts you avoid; scaling support without hiring covers the mechanics.

8. Pay per outcome and contract the rate

Pricing structure decides whether the AI's failures cost you money. Per-seat pricing scales with the humans you keep; per-conversation pricing charges for tickets the AI touched but did not finish; per-resolution pricing falls when the AI fails. Convert every quote: price per conversation divided by resolution share is the effective price per resolution, so $2 per conversation at 50% is $4. Then lock it: Gartner predicts GenAI cost per resolution will exceed $3 by 2030 as vendors move toward profitability.

How much does a human support ticket cost?

Roughly $7 to $11 per ticket for a US team on public data, depending on channel. The BLS median wage for customer service representatives was $21.53 an hour in May 2025. In the June 2026 employer cost release, wages were 70.0% of private-industry compensation costs, so the loaded cost is about $30.76 per paid hour; divide by 0.70 for shrinkage and multiply by an overhead factor such as 1.25 and a productive hour costs about $54.92. At seven digital tickets an hour that is $7.85 a ticket; at five calls an hour, $10.98. Gartner's 2019 poll of service leaders put live channels at $8.01 per contact. Use your own loaded number; customer service cost per ticket walks through it.

Where is the break-even point for AI support costs?

Break-even resolutions per month equal the plan fee divided by the human cost avoided per ticket minus the price per resolution. On Lorikeet's published rates:

Worked example: a consumer lender handles 10,000 contacts a month at $8.01, a $80,100 baseline, on Scale. Resolving 50% costs $4,500 in resolutions, $5,100 for the plan and $2,700 for QA on all contacts, $12,300 in total, against $40,050 of avoided cost: a net $27,750, about 35%. If disputes, hardship and complaints push the resolved share to 35%, the net is $17,085, about 21%. Both sit near Gartner's prediction of a 30% cut in operational costs when agentic AI resolves most common issues by 2029. A fuller model with three scenarios is in AI customer support ROI for high-volume teams, or plug your own numbers into Lorikeet's free CX ROI calculator.

Tools that help

If most of your volume is simple questions and you run Zendesk or Intercom, the native agent is often the cheaper route because there is no second contract. A plan fee is the trade-off with Lorikeet: below a few hundred resolutions a month a pure per-outcome tool usually costs less.

Where does AI raise support costs?

AI adds cost when it touches tickets it cannot finish. Bad escalations are the main one: if the handoff loses context, the customer repeats everything and the human ticket takes longer. Hidden exits cost trust; in a Gartner survey, 64% of customers said they would prefer companies did not use AI in customer service. Per-conversation billing charges for failures. Gartner also predicts that by 2028 right-to-a-human regulation will raise assisted service volume by 30%, so budget a human floor that grows with your customer base.

What still needs a human

Decisions with legal, financial or clinical consequences. Under Regulation E, a financial institution must determine whether a transfer error occurred within 10 business days, or take up to 45 days if it provisionally credits the account; the AI can take the intake, but a person decides. In healthtech the AI never gives clinical advice and escalates to clinicians. Hardship, vulnerability and complaints go to people with context already collected. Model all of these as human cost from day one.

Questions to ask before you pick a lever

• Which ten intents make up most of our volume, and which of them can be closed with a system action?

• What is our loaded cost per ticket by channel, agreed with finance?

• How many minutes per ticket do agents spend on reading, tagging and reassigning?

• What is our reopen and recontact rate today, so we can see whether AI moves it?

• Does the vendor bill escalations, handoffs or conversations, and who defines a resolution?

• Which freed hours become avoided hires, and in which month?

Start with the lever that touches the most volume at the lowest risk, usually triage or the top transactional intents, and add voice and regulated workflows once the first results are measured.

To see which of your tickets an AI agent can resolve, start a 30-day free trial.