What Does Customer Support Actually Cost?
Jonas Budelmann7 min read
On this page
Customer support bills you in four separate places: the people who answer, the software they answer in, the AI that answers first, and the documentation work that stops the question arriving at all. Almost every published cost figure prices only the first of those, usually as an hourly rate for an outsourced agent. That's a useful number and it isn't your bill. This is what the other three cost, and how to assemble the whole thing from figures you can look up.
The four lines on a support bill
Cost per contact is the metric underneath all of this, and the formula is the boring one: total support spend divided by the number of contacts you handled. The arithmetic isn't the hard part. Those inputs live in four different places, and nobody owns all four of them. Payroll knows the salaries, the finance team knows the software subscriptions, and the AI charge often arrives on whichever card was attached to the help desk account.
The first line is the people, and it's the one every published benchmark prices. If you're hiring rather than outsourcing, that number starts from the wage and grows through benefits, management time and the weeks before a new hire is useful. We went through that arithmetic properly in what a support strategy costs to run, so it isn't repeated here.
The second line is the software: the shared inbox, the knowledge base, the chat widget. The third is the AI, which is now billed separately by most vendors and behaves unlike anything else on the list. The fourth barely looks like a cost at all, because it's the hours somebody spends writing help articles and fixing the confusing checkout step. That work has no invoice, which is exactly why it gets skipped, and it's the only line that reduces the other three.
Working it out once tells you where you stand today. If you want it to be a number you watch instead, cost per contact as a support metric makes that case.
What does an hour of support labor cost?
Between about $7 and $60, and where you land depends almost entirely on geography rather than on quality. EverHelp, an outsourcing provider, published a regional breakdown of outsourcing rates in June 2026 that puts United States agents at $38 to $60 an hour, Eastern Europe at $15 to $35, the Philippines at $9 to $19, and India and Egypt at $7 to $18. Those are the provider's own published rates for its own category, so read them as a market map rather than as an independent survey.
That spread matters more than any single figure in it. A support hour is one of the few business inputs where the same work costs eight times more in one country than another, which is why the outsourcing industry exists at all. An hourly rate also hides what you're buying. You're paying for someone's time, not for the questions they answer. You pay for the hour whether it handled two conversations or twenty, and getting better at support makes each hour cover more ground rather than making the hour cheaper.
What the software line costs, and when it moves
Help desk software is priced per seat per month, so it tracks the size of your team rather than the size of your queue. Help Scout publishes three tiers at $25, $45 and $75 per user a month on monthly billing, plus a discount for paying annually and a free tier for up to five users, all retrieved on 26 August 2026. The differences between tiers are usually about reporting, automation and how many mailboxes you can run, rather than about the inbox itself.
This line is predictable, which is what makes it easy to plan around. Nothing your customers do next month changes it. It moves when you hire somebody, which is a decision you make deliberately with a budget in front of you, and it moves back down when you don't.
Why the AI line behaves differently from the rest
The AI line is priced per successful resolution, which makes it the only part of the bill that rises when your product gets more popular. Help Scout charges $0.75 for each conversation AI Answers resolves without a human, billing once per conversation however many questions it answered. Zendesk charges $1.50 per automated resolution and defines one by a 72-hour quiet period, meaning the ticket stays closed and the customer doesn't come back. Both figures were retrieved on 26 August 2026, and EverHelp puts the wider market somewhere between $0.50 and $5.00 a resolution.
Read the definitions before the prices, because they decide when the meter runs. A vendor that charges on a quiet period is betting on the answer holding for three days. A vendor that charges when the customer doesn't escalate is measuring something slightly different, and neither one measures whether the customer was happy.
Underneath these charges, the model itself costs a fraction of what's billed for it, and the gap between those two numbers is the whole argument of where the money goes in AI support pricing. This line, alone among the four, is indexed to something you don't set.
Which line grows when you get busy?
Only the AI one, if you're paying per resolution. That's the imbalance worth building around, because the other three lines move when you decide they should.
| Line item | What makes it go up | Who controls that |
|---|---|---|
| People | Hiring another person | You |
| Software | Adding seats | You |
| AI, billed per resolution | Every conversation the agent resolves | Your customers |
| Content work | Nothing, it's a fixed effort | You |
A good launch, a viral post or a seasonal rush all raise the third row without touching the others. Worse, the better your agent gets, the more of those conversations it resolves, and every resolution is a charge. Improving the thing you bought makes it cost more.
The third row is the only one priced that way, and it doesn't have to be. That's why Optlo bills for capacity rather than for success events. You connect your own OpenAI, Anthropic or Gemini account, so model usage arrives on a bill in your name at the provider's published rate. No token markup, no resolution fees. A conversation costs the same whether the agent resolves it or hands it to a person, which means a good month for your product is just a good month.
How to work out your own number
Take a full month, add the four lines, and divide by the conversations you handled. Salaries and any outsourcing invoices go in at their fully loaded cost rather than the headline wage. Software goes in at what you paid that month, seats and add-ons included. The AI line goes in as whatever appeared on the resolution charge or the provider bill. The content work goes in as the hours somebody spent, priced at their real cost, which is the line most teams discover they've been getting for free from a founder.
Whatever number you land on is worth more than any benchmark, because your product decides your contact mix and no industry average knows anything about your product. Once you have it, the question stops being whether support is expensive and becomes which of the four lines is growing fastest.
If you'd rather see the shape of it before doing the arithmetic yourself, our support cost calculator runs the same four lines across your own volume and resolution rate. And if the third line is the one that worries you, Optlo's 7-day trial includes AI usage, so you can watch what your own traffic generates before deciding anything. When it ends, you connect your own AI account to keep the agent running and pay your provider directly.
Common questions
What is the average cost of customer support per contact?
There isn't a trustworthy single average, and the ones published are usually industry estimates with no named study behind them. The spread is wide: a conversation handled by a United States agent at $38 to $60 an hour costs many multiples of one resolved automatically at $0.75. Your contact mix decides where you land, so calculate your own figure from a real month rather than adopting somebody else's.
How do you calculate the cost of customer support?
Add every support cost for one month, then divide by the number of conversations handled in that month. The costs to include are fully loaded salaries or outsourcing invoices, help desk software including seats and add-ons, any AI resolution or token charges, and the hours spent writing help content. Dividing gives you cost per contact, which is the standard measure and the one worth tracking over time.
Is outsourcing customer support cheaper than hiring?
On the hourly rate, usually yes, and by a wide margin: EverHelp's June 2026 figures put offshore rates at $7 to $23 an hour against $38 to $60 in the United States. The comparison is incomplete though, because outsourcing buys capacity rather than answers, and product knowledge takes longer to build in a team you don't manage. It tends to work best for high-volume, well-documented questions and worst for anything needing judgment about your product.
Why does AI customer support get more expensive as it improves?
Because most vendors only charge when the AI succeeds. Help Scout takes $0.75 every time its AI closes a conversation without a person, and Zendesk takes $1.50. So an agent that gets twice as good hands you twice the AI bill, on exactly the same traffic. That's the pricing talking, not the technology. Buy a plan with a set number of conversations included, and one conversation costs the same whether the agent resolves it or hands it to you. Put the model on your own account, and you just pay for what it uses.
What is the cheapest way to reduce customer support costs?
Fixing the reason people write in. It has no invoice attached, and it's the only lever that reduces all four lines at once. Sort your last few hundred conversations by question type, and anything appearing more than a handful of times is either a documentation gap or a product problem. Automating a bad answer just makes it arrive faster.
