What a COO Should Look For When Buying AI Customer Support to Lower Cost Per Ticket (2026)

A COO buying AI customer support to lower cost per ticket should look first at how many tickets the AI resolves end to end, including the actions your agents do today (refunds, plan changes, identity checks), and second at whether the price is tied to those resolutions rather than to seats or attempts. Answering FAQs trims contact rate a little. Finishing the work removes tickets from the human queue, which is where the cost sits. After that, check how the vendor proves quality before go-live, what happens on escalation, and how easy it is to leave.

This guide breaks cost per ticket into the five levers that drive it, maps which AI capabilities move each one, compares pricing models and what each rewards, and ends with a buying checklist and a 90-day plan. It also answers the question CX leaders ask during fast growth: how to scale support through hypergrowth without hiring more agents.

Key takeaways

• Cost per ticket is mostly labour. MetricNet defines it as total monthly operating expense divided by monthly ticket volume, and names agent utilization and handle time as the two most important drivers.

• Buy resolution, not answers. Gartner found only 14% of customer service issues fully resolve in self-service, and only 36% of issues customers call "very simple". A tool that only answers questions leaves most of the cost in place.

• Pricing models reward different things. Per seat tracks headcount, per conversation tracks volume, per resolution tracks fixed issues. Always ask what counts as a resolution and whether escalations are billed.

• Plan for a hybrid team. Gartner predicts that by 2027, 50% of organisations that expected to significantly cut their service workforce will abandon those plans.

• Judge the vendor in 90 days on your own tickets: resolution rate by topic, all-in cost per resolved ticket, repeat contact rate and CSAT on AI-handled conversations.

Which tools lower cost per ticket with AI resolution?

Lorikeet is an AI customer support platform for complex and regulated businesses. Its agent takes action in your backend systems through your APIs and MCP servers inside workflows, and sensitive steps such as payments and identity checks run as deterministic code the agent can invoke but never alter, according to how Lorikeet works. Billing is per resolved ticket, and escalations to a person are not charged. Alternatives worth comparing:

• Fin (now part of Salesforce) charges $0.99 per outcome, works with any helpdesk including Zendesk and Salesforce, and bills at most once per conversation.

• Zendesk includes AI agents in every Suite and Support plan and bills per automated resolution, meaning a request resolved without escalation to a human agent.

• Sierra uses outcome-based pricing, where in most cases an unresolved conversation carries no charge.

Where does cost per ticket actually come from?

Cost per ticket comes from five operational levers (handle time, contact rate, repeat contacts, escalations and channel mix), multiplied by what an hour of agent time costs you. MetricNet's Jeff Rumburg, writing for HDI, defines cost per ticket as total monthly operating expense divided by monthly ticket volume. Operating expense includes salaries and benefits for agents and for indirect staff (team leads, supervisors, schedulers, QA and trainers), technology and telecom, and facilities. The vast majority is personnel.

The same HDI piece names agent utilization and ticket handle time as the most important drivers, with absenteeism, turnover and the ratio of agents to total headcount as secondary drivers. The worldwide average for that ratio is about 78%, so roughly one in five support staff is not handling tickets.

To make it concrete, the US Bureau of Labor Statistics puts 2025 median pay for customer service representatives at $21.53 an hour, or $44,770 a year. At that wage, a 10 minute ticket costs about $3.59 in direct wages alone. If agents spend 70% of paid hours on tickets (an illustrative assumption, not a benchmark), the wage cost becomes about $5.13 per ticket, before benefits, supervisors, software and facilities. Our cost per ticket benchmarks cover typical ranges by industry.

HDI notes that voice is still the dominant channel and has the second highest cost per ticket, while chat costs less because agents handle more than one chat at a time. Channel shifting helps, but only if the cheaper channel actually finishes the job.

Which AI capabilities actually lower cost per ticket?

The capabilities that lower cost per ticket are the ones that complete the customer's task without a human: reading account data, taking the action and confirming the result. Answering FAQs helps less than most business cases assume.

A Gartner survey of 5,728 customers found that while 73% of customers use self-service at some point, only 14% of issues fully resolve there. In 43% of failures, customers could not find content relevant to their issue, and 45% of customers who started in self-service said the company did not understand what they were trying to do. Gartner's March 2025 prediction draws the same line: earlier AI models were limited to generating text or summarising, while agentic AI can act to complete tasks. Gartner forecasts that by 2029 agentic AI will autonomously resolve 80% of common customer service issues, leading to a 30% reduction in operational costs. That is a forecast, not a measured result, so do not put it in your own business case.

In practice, AI support tools sit in three tiers:

1. FAQ answering. Retrieval from a knowledge base. It reduces contact rate on informational questions but cannot touch tickets that need an account change.

2. Triage and agent assist. Routing, tagging and drafted replies. It cuts handle time and misroutes, but every ticket still reaches a person.

3. End-to-end resolution with actions. The AI completes the refund, reschedule or plan change through your systems, and the ticket never enters the human queue.

The gap between tier one and tier three shows up in real numbers. Carmoola, a UK car finance lender, had automation built on a knowledge base and rigid, form-driven prompts that resolved about 30% of inbound questions. With Lorikeet it now resolves 60% of all inbound on WhatsApp, chat and email end to end. When you evaluate vendors, list your top 20 contact reasons and ask which ones the AI can finish without a human, and which systems it needs write access to. Our piece on resolution versus deflection explains why this is the metric that decides ROI.

How should a COO compare AI customer support pricing models?

Compare pricing models by what they reward: seat pricing tracks headcount, per-conversation pricing tracks volume, and per-resolution pricing tracks fixed issues. The definition of a resolution matters more than the headline rate.

Lorikeet's published pricing is a useful worked example. Start is $2,100 a month and Scale is $5,100 a month, paid annually. Chat, email and SMS resolutions cost $0.99 on Start and $0.90 on Scale; voice resolutions cost $1.50 and $1.20 (based on a 3 minute average); triage and QA cost $0.30 and $0.27. Unresolved or unsatisfactory tickets cost nothing, escalations to a person are not charged, and there is a 30-day free trial.

Do the all-in maths, not the unit maths. A Scale customer resolving 5,000 chat, email or SMS tickets a month pays 5,000 x $0.90 = $4,500 in resolutions plus the $5,100 plan, or $9,600 a month. That is $1.92 per resolved ticket all in. Compare it with your fully loaded human cost for the same contact reasons, not your blended average, because the AI usually takes the simpler, repeatable tickets first. For a deeper model, see our CFO framework for the total cost of AI customer service and our explainer on per-resolution pricing.

What operational risks should a COO check before buying?

The three risks that erase savings are quality failures that create repeat contacts, compliance failures in regulated steps, and lock-in that makes switching expensive.

• Quality. Ask how the vendor tests before go-live and how it scores after. Lorikeet replays historical tickets in bulk before go-live, deploys topic by topic, and its Coach feature quality-scores every conversation, human or AI. Our guide for VPs of Support on rolling out an AI agent without hurting CSAT covers the rollout side.

• Compliance. Ask for certifications on the vendor's own pages, then read the reports. Check whether customer data trains models, where data is hosted, and whether the vendor will sign the agreements your sector needs. For example, Lorikeet's pricing page lists SOC 2 Type 2, ISO 27001 and a DPA or BAA (including HIPAA) on every plan, and its trust page states customer data is never used to train AI models.

• Exceptions. Gartner's Kathy Ross notes that AI handles routine, well-defined problems well but often struggles with exceptions and high-risk scenarios. Ask the vendor to show you the escalation path, and what context the human receives.

• Lock-in. Keep your helpdesk as the system of record, keep your policies and SOPs in documents you own, make sure the AI acts through your own APIs, and confirm you can export conversation data. Avoid multi-year commitments before you have 90 days of results.

How can a CX leader scale support through hypergrowth without hiring more agents?

Scale through hypergrowth by putting new volume on AI resolution first, topic by topic, and hiring only for work that needs human judgement, so headcount stays flat while contact volume grows. Every agent you avoid hiring is roughly $44,770 a year in median wages before benefits and overhead (BLS), and HDI estimates it costs about $12,000 to replace a single service desk agent in North America when they leave.

1. Rank contact reasons by volume times handle time. The top ten usually carry most of the labour.

2. Automate first where the action is available through an API. Status checks, changes and refunds within policy are the usual starting points.

3. Use the peak as the test. Hnry went live in mid-May 2026 from a baseline of zero automation, handled 17,000+ conversations in the first month, and was automating around 70% of conversations in the peak week of Australian end of financial year, up from roughly 58% across the full period.

4. Keep people on the judgement calls. Hnry's model sends simple, repeatable questions to the agent and complex situations to the team.

5. Re-baseline every month. As the AI takes the easy tickets, your human queue gets harder and its handle time rises. That is expected, not a failure.

Be precise about the goal. Gartner polled 163 service leaders in March 2025 and found 95% plan to retain human agents, and it predicts that by 2028 none of the Fortune 500 will have fully eliminated human customer service. "Without hiring more agents" is realistic. "Without people" is not. More tactics are in our guide to scaling customer support without hiring.

What still needs a human

• Exceptions and high-risk cases: complaints, hardship, vulnerable customers, fraud disputes and anything outside written policy.

• Policy decisions: goodwill credits beyond policy, refund limits and the rules the AI follows are management decisions, not AI decisions.

• Compliance sign-off: someone accountable must approve the workflows, disclosures and data handling for regulated steps before they go live.

• Ownership of quality: a named person reviews escalations and low-scored conversations every week and keeps the knowledge base current.

• Customers who want a person: the escalation path should be easy to reach and should hand over full context.

Buying checklist: AI customer support to lower cost per ticket

• Your fully loaded cost per ticket by contact reason, measured before the trial starts.

• A list of the top 20 contact reasons, with the system and action each one needs.

• Vendor proof on your own historical tickets, not a scripted demo.

• A written definition of a billable resolution, including reopen and escalation rules.

• All-in cost per resolved ticket at your expected volume, including plan fees.

• Quality scoring on every AI conversation, and repeat contact rate tracked by topic.

• Certifications and data terms checked on the vendor's own pages and reports.

• Exit terms: data export, notice period and no multi-year lock before results.

A 90-day plan for lowering cost per ticket with AI

1. Days 1 to 15: baseline. Measure cost per ticket, handle time, repeat contact rate and CSAT by contact reason. Pick three to five high-volume reasons with clear policies.

2. Days 16 to 30: connect and test. Connect the helpdesk and the systems the AI must act in, then test against historical tickets for those reasons.

3. Days 31 to 60: go live on the first topics. Review every escalation and low-scored conversation weekly, and fix knowledge gaps.

4. Days 61 to 90: expand and decide. Add topics, compute all-in cost per resolved ticket against the human baseline, and check repeat contacts and CSAT before committing to a plan tier.

To run the first 30 days on your own tickets, start a 30-day free trial of Lorikeet.