AI Agent vs Hiring vs Outsourcing Support: Cost per Ticket in 2026

A support hire, an outsourcer and an AI support agent are priced in different units: a salary, an hourly rate, a fee per chat or per resolution. A fair comparison needs one unit, the cost of answering one ticket, built from the same volume and assumptions for every option. This guide works it out at 500, 2,000 and 10,000 tickets a month, including the setup most small teams end up with: AI in front, a person behind.

The guide is for founders, store owners and support leads deciding whether to post a job ad, sign an outsourcing contract or let an AI answer first. Wage and benefit figures come from the US Bureau of Labor Statistics (BLS), outsourcing prices from the five providers we found that publish them, and AI prices from our AI customer support pricing index.

Every price was read on the provider's own page on 19 October 2026, and every assumption sits next to the number it drives, so you can swap in your own. PepoChat is our product and appears as one AI option under the same assumptions; it publishes no resolution rate, so the rates below come from Intercom, Salesforce and Klarna's own reports.

What are the three ways to staff customer support?

You can hire your own reps, pay an outsourcer to staff the queue, or put an AI support agent in front of customers to answer what it can and pass the rest to a person.

In-house support means employees on your payroll who answer tickets in your own tools. You pay for their hours whether or not tickets arrive, so the cost per ticket falls as each person gets busier and jumps each time you add a person.

Outsourced support means another company staffs the work. A business process outsourcing (BPO) firm sells dedicated agents by the hour or the month; an answering service sells individual chats or calls handled from your script, which suits messages and simple questions better than work inside your systems.

An AI support agent is a chatbot that answers from your own content, can look things up in connected tools, and hands the conversation to a person when it cannot help. Vendors bill the AI per resolution, per conversation or credit, per seat, or as a flat plan with an allowance, as our AI chatbot cost guide explains.

Cost per ticket is the month's total cost of answering support divided by the tickets answered that month. A ticket here is any customer question that needs an answer, by chat, email or form, and one conversation counts as one ticket.

What inputs does the cost model use?

The model uses a wage, a benefits multiplier, a tool cost, a productivity assumption, published prices and an AI resolution rate. The table lists each with its source and date; the subsections explain the four that need judgment.

Last verified 19 October 2026.

What does an in-house support rep cost a month?

A full-time US customer service representative at the median wage costs about $5,466 a month once benefits, payroll taxes and a helpdesk seat are added. The BLS Occupational Outlook Handbook lists median pay of $44,770 a year, or $21.53 an hour, for May 2025, on a page last updated on 27 August 2026. That is $3,730.83 a month in wages.

A benefits multiplier turns a wage into what the employer pays: one divided by the wage share of total compensation. In the BLS Employer Costs for Employee Compensation table for June 2026, wages make up 68.7% of compensation for private-industry office and administrative support workers, the group that includes customer service representatives. The other 31.3% covers paid leave, supplemental pay, insurance, retirement and legally required payments such as Social Security. Dividing $3,730.83 by 0.687 gives $5,430.62 a month, a multiplier of about 1.46, or $31.34 an hour.

The 70.0% wage share for all private-industry workers gives the $30.76 an hour our chatbot ROI guide uses; the occupational group's figure adds about 58 cents an hour. A $35 helpdesk seat, Freshdesk's Omni Growth plan on monthly billing, brings one rep to $5,465.62 a month. Office space, equipment, recruiting and a manager's time are not included; the hidden-cost section returns to them.

How many tickets does one rep handle in a month?

We found no neutral, dated source for how many tickets a support rep handles in a month, so the model treats it as an assumption: 600 tickets for one full-time person, tested at 400 and 800. Six hundred is 30 tickets on each of 20 working days, or one every 16 minutes of an eight-hour day once breaks, meetings and training share the day.

Ticket mix moves the figure most: a delivery-date question takes a minute or two, a billing dispute or bug report can take an hour. Your own number is last month's solved tickets divided by the full-time equivalents who answered them.

A fully used rep costs $5,465.62 divided by the tickets handled. The model rounds up to whole people, because a fraction of a hire is still a hire unless you can hire part-time.

What do outsourced support providers charge?

Few outsourcers publish prices, so the table lists only prices read on each provider's own page on 19 October 2026, not an industry average. Hugo, Influx, Smith.ai, Ruby and HelpSquad publish prices for outsourced support or chat answering; SupportNinja, Helpware and SupportYourApp ask for a quote, and Chatdesk shows its pay-per-ticket plan without a price.

Last verified 19 October 2026 on each provider's own pricing page.

The published prices fall into two groups. A full-time dedicated agent costs $1,560 a month at Hugo's starting rate and $2,000 on Influx's Managed Operations, 29% to 37% of a US rep's $5,430.62 loaded monthly cost. Chat answering services price each chat, from $3.00 on HelpSquad's largest plan to $14.30 on Ruby's smallest, and their agents work from the script and answers you give them; our 24/7 support guide covers where they fit.

Starting prices are floors: Influx prices by agent level, region and channel, with a 10% premium for languages beyond standard English, and Hugo builds a custom plan for each client. Use them to check a quote, not to replace one.

What do AI support agents cost, and how much do they resolve?

AI support agents cost from under 1 cent to about $2 a conversation on October 2026 list prices, and the resolution rate decides how much of the queue that price covers. Our pricing index puts 2,000 AI conversations a month at $25 on one Zoho SalesIQ seat, $69 on PepoChat's Starter plan at two replies each, $2,019 on Intercom Fin with one seat and $4,059 on Zendesk, counting every conversation as a billable resolution.

A resolution rate is the share of conversations the AI finishes without a person. In an Intercom community answer dated 3 August 2023, an Intercom support engineer wrote that a good Fin resolution rate "starts at around 30-50%", and customers in the thread later reported 25%, 45% and 55% to 60%. Mature figures run higher: Intercom put Fin's average across customers at 76% on 12 March 2026, Salesforce reported 76% resolved and 5% escalated on its own help site in December 2024, and Klarna said in February 2024 that its assistant handled two-thirds of its chats in its first month.

The model uses 30% and 50%, the band Intercom's guidance gives for a good resolution rate, and treats 76% as what a mature deployment reaches after months of content work, as our ticket deflection benchmarks explain. All of these figures are vendor-reported, and most count a customer who goes quiet as resolved.

What does a support ticket cost at 500, 2,000 and 10,000 tickets a month?

On the inputs above, a ticket costs $9.29 to $10.93 with in-house reps, $2.71 to $5.00 with the outsourcers that publish prices, 2 cents to $1.07 a conversation for the two AI rows, and $4.94 to $11.03 with AI in front of in-house reps. Every cell shows its arithmetic under three rules.

• People: an in-house rep costs $5,465.62 a month and an outsourced agent its provider's starting price plus a $35 seat, assuming they work in your helpdesk; each handles 600 tickets, rounded up to whole people.
• AI: two AI replies a conversation on PepoChat, as in our pricing index; on Intercom Fin, every conversation billed as an outcome, which is the ceiling.
• Blended: the AI answers every ticket first, and in-house reps take the share it does not resolve.

Last verified 19 October 2026. Totals are estimates from published list prices and BLS wage data under the stated assumptions, not quotes.

Whole people make small volumes expensive. At 500 tickets a month one full-time rep uses 83% of a 600-ticket month, so each ticket costs $10.93 rather than $9.11. The same rounding makes both blended rows cost $11.03 at 500 tickets, more than people alone: the AI removes a third or half of the tickets but not the one person behind it. With a half-time person behind an AI that resolves 50%, the month costs about $2,799, half the loaded wage plus a full seat and the $49 plan, or $5.60 a ticket.

Savings appear when the AI removes whole people. At 2,000 tickets a month, a 50% resolution rate takes the team from four reps to two and the monthly cost from $21,862 to $11,000, a saving of $10,862; at 30% the team needs three reps and saves $5,396. At 10,000 tickets the blended rows save $27,100 to $43,496 a month against an in-house team of 17.

Outsourcing buys people's time most cheaply on published prices, at $2.71 to $4.07 a ticket for dedicated agents, because Hugo's and Influx's starting rates sit far below a loaded US wage. The answering-service row stays at $5.00 a chat because Smith.ai's Pro plan charges $5 for each chat past its 120; it fits overflow and after-hours messages better than a whole queue.

The AI rows are the cheapest per conversation, but no AI resolves every ticket, so they price a unit rather than a support operation. Intercom Fin bills only outcomes, so at a 50% resolution rate its AI line is about half the ceiling shown.

What can't an AI support agent do, and what does that cost?

An AI support agent should not finish tickets that need judgment, a refund or proof of identity on its own, and those are the expensive part of a queue. The blended rows price that remainder at in-house rates, and the remainder is harder than the average ticket.

Judgment comes first: refunds outside policy, complaints, bug reports and anything your content does not cover. In PepoChat, human handoff, the step where the AI passes a conversation to a person, happens when the visitor asks for one, when the agent finds no answer and the visitor accepts its offer of help, or when a tool or the model fails. The team is emailed the reason and the last three messages, and the conversation waits in the team inbox with its transcript, as the human handoff docs describe.

Money and identity come next. PepoChat's actions run at least once, so a rare retry can repeat a write, which is why refunds and payments are never wired as actions and a person issues them. Email verification gates meeting memory, bookings with the visitor's verified details and the emailed transcript, but not order or ticket lookups; an action's instructions can make the agent ask the visitor to verify first, but the agent is not told whether they did. Account changes that need proof of identity belong with a person.

The tickets left for people are harder than average, because the AI takes the quick ones. If your reps handle 450 of the remaining tickets a month instead of 600, the 2,000-ticket case at a 50% resolution rate needs three reps rather than two, and the blended cost rises from $5.50 to $8.23 a ticket, the same as at 30%. Time a month of escalated tickets before you plan headcount on the blended rows.

The AI's pricing model matters less than its resolution rate. With Intercom Fin in front instead of a flat plan, the same 2,000-ticket case at 50% pays $990 in outcomes plus a $39 seat instead of $69, and the blended cost moves from $5.50 to $5.98 a ticket, because the two reps' $10,931 dominates either AI bill.

Which hidden costs change the comparison?

Six costs sit outside the per-ticket table: hiring and ramp-up, turnover, supervision, coverage hours, outsourcing oversight, and AI setup and upkeep. The table shows where each lands; the paragraphs below price those that can be sourced or computed.

Hiring and ramp-up are paid in weeks: each week a seat stays empty, or a new rep is still learning the product, costs about $1,261 of loaded cost in other people's time or slower answers.

Turnover is built into the occupation. The BLS projects about 289,500 openings a year for customer service representatives against 2,666,000 jobs in 2025, all of them from replacing workers who move to other occupations or leave the labor force, since employment is projected to fall 5% from 2025 to 2035. That is one opening for every nine jobs each year, before counting people who simply change employer.

Coverage hours multiply headcount. A week has 168 hours, so one seat staffed around the clock needs 4.2 full-time people before holidays and sick days, about $22,956 a month at the loaded cost, against about $6,552 at Hugo's starting rate. An AI support agent costs the same at 3 a.m. as at noon. PepoChat's Team availability setting decides what it promises when nobody is online: My team answers during business hours uses your away message outside those hours and asks for the visitor's email before handing over, and No one answers live chat, we follow up by email never promises a person, as the widget customization docs explain.

Outsourcing oversight is the cost of managing quality from a distance. An outsourced team needs a written playbook, access to your helpdesk and order data, and someone on your side who reads a sample of its tickets every week. Hugo's starting price includes onboarding, QA, training and team leads; Influx's Managed Operations includes QA and reporting but requires a team leader at an unpublished price.

AI setup and upkeep are mostly your own time. Our cost guide budgets a day or two to import the site, upload documents and fix wrong answers, then about an hour a week of conversation review: at $31.34 an hour, $251 to $501 to start and about $136 a month, more than PepoChat's $69 bill at 2,000 tickets. Building the agent yourself moves the cost to engineering time, as our build vs buy guide shows.

Should a small business hire, outsource or use AI?

Let volume decide how many people you need and ticket type decide who answers what. The rule from the model is to put the AI in front of every ticket and hire a full-time person once the tickets it does not resolve reach a full workload.

Written as a formula, hire when monthly tickets × (1 − AI resolution rate) reaches the tickets one rep handles. At 600 tickets a rep, the first full-time hire makes sense at about 850 tickets a month with a 30% resolution rate and 1,200 with 50%, and each further 600 unresolved tickets adds a person. Below that line, a part-timer or the founder covers the remainder.

Within each setup, ticket type decides who answers what.

What to do next

Count last month's tickets by type and estimate how many would reach a website chat with our guide to how many chatbot conversations a month to expect. Remove what you can at the source with the tactics to reduce support tickets, then run the rest through the tables above. To test AI in front on your own content, create a free PepoChat workspace with 200 AI replies a month and every feature on the free plan; the pricing page lists Starter at $49 for 2,000 replies and 3 team members.