How Much Does It Cost to Outsource Customer Service?
Outsourced customer service is quoted in one of three units, and the unit decides the number more than the vendor does: per hour of agent time, per dedicated seat per month, or per interaction, meaning per call, per ticket, per minute or per resolved contact. None of the three is comparable to the others until you convert it to a cost per resolved contact at your own volume. And the number any quote should be compared against is not a published range, it is the in-house cost it replaces, whose base is federally published: the most recent BLS estimate puts the median US customer service representative at $21.53 an hour, or $44,770 a year, measured across 2,595,750 jobs.
This page does not give you a dollar range, and the reason is worth stating up front rather than discovering three sections in. The ranges that circulate on outsourcing pages, including on this site’s older posts, come from vendor pricing surveys rather than from an audited dataset. They are directional. Quoting them back as though they were measurements is how a buyer ends up anchored on a number nobody computed, and the anchor is the expensive part.
What follows is the arithmetic instead.
The three units, and why they are not interchangeable
Per hour. You buy agent time. Simple to compare across vendors, which is why it is quoted first. You absorb all unproductive time, because you are paying for the clock rather than the outcome.
Per dedicated seat per month. You buy a named person for a fixed block of coverage, billed whether the phone rings or not. Below a steady volume this is the expensive option; above it, it is usually the cheapest per resolved contact, and it is the only one of the three that reliably buys you an agent who learns your business.
Per interaction. You buy outcomes: per call, per ticket, per minute, or per resolved contact. The vendor absorbs unproductive time, which is why this unit prices it in. Read the definition clause carefully, because “handled” and “resolved” are different words and only one of them is worth paying for.
The unit is not a detail. It decides who eats the contacts that never become conversations, and that share is not small.
Building the baseline the quote is compared against
The honest anchor for an outsourcing decision is your own loaded cost per resolved contact. Build it in that order.
Start from the published base. The most recent BLS Occupational Employment and Wage Statistics estimate puts the median US customer service representative at $21.53 an hour, or $44,770 a year, across 2,595,750 jobs. That is a national median for the occupation, and it is a base wage.
Then add your own figures for the things the wage does not include: payroll tax, benefits, a supervisor’s time, telephony and helpdesk software, recruiting and onboarding, and the hours a seat sits idle in a slow week. The multiplier you apply is an assumption you are choosing. It is not a BLS figure and must not be attributed to one. Write it down, show your working, and keep it visible in the comparison, because a buyer who quietly assumes 1.3x and a vendor who quietly assumes 1.6x are having two different conversations.
Then divide by resolved contacts per month, not by contacts handled. Now you have a number the vendor quote can be measured against. The call center cost calculator runs this arithmetic if you would rather not build the spreadsheet.
The cost the quote does not show: contacts that never become conversations
A quote prices what happens on a productive contact. The gap between models is decided by the unproductive ones, and buyers routinely underestimate that share.
On our own corpus of 10,794 outbound roofing appointment-setting calls placed in February and March 2026, every one transcribed and analysed, 2,827 (26.2%) ended in a hang-up, 1,173 (10.9%) in an explicit “not interested”, 439 (4.1%) in a do-not-call request, 329 (3.0%) did not qualify, 212 (2.0%) booked an inspection and 78 (0.7%) asked for a callback.
Two caveats travel with that, and it is a misquote without them. Those six codes account for 5,058 of the 10,794 calls, or 46.9%, and the remaining 53.1% carry dispositions the source documents never enumerate, so the six do not partition the corpus and must not be read as adding to a whole. More importantly for this page: all of it is outbound cold dialling, and none of it describes how an inbound caller behaves. Inbound customer service has a different shape entirely, because the person on the other end chose to call.
It is here because the structural point survives the difference. A large fraction of contact attempts end without the work you are buying having happened. Under per-hour pricing you pay for every one of them. Under per-interaction pricing the vendor does, and has priced it into the unit. Neither is a trick; they are the same cost allocated differently, and the one that is cheaper for you depends on your own ratio, which is why you have to know it before you can read a quote.
The clause that decides what you actually pay
Most of the variance between a quoted cost and an invoiced one lives in four places.
Overage. A low base rate with overage billed at two or three times the base is the most common first-invoice surprise. Ask for the overage rate in writing and model a month at 130% of expected volume.
After-hours and weekend premiums. Overnight and weekend coverage costs more per hour in every model. If your volume is not all in business hours, a business-hours quote is not your quote.
Minimum commitment. A monthly minimum you would not otherwise reach is a real cost even when the per-unit rate is excellent.
Ramp. The weeks before a new agent is productive are paid for by somebody. If the contract does not say who, it is you.
Service level is a contract term, not a legal standard
Buyers often assume there is a rule about how fast a customer service line has to answer. There is not, and the shape of what federal rule does say is instructive.
The FTC Telemarketing Sales Rule governs the opposite direction. Under it, an outbound call is abandoned if a person answers and is not connected to a live representative within 2 seconds of their completed greeting, and the safe harbor caps abandonment at 3% of calls answered by a person while requiring at least 15 seconds or 4 rings before disconnecting. That is a rule about how long a telemarketer may leave someone hanging.
Nothing federal says how long an inbound customer may sit in a queue. So every answer-speed and service-level commitment in a customer service contract is negotiated. If it is not written into the agreement with a remedy attached, it does not exist, and a vendor’s verbal 80/20 during the sales call is worth exactly nothing when the queue backs up in month four.
If you are a contractor rather than a support organisation
The word “customer service” covers two jobs that price differently, and contractors usually want the second one.
Support means resolving problems for people who are already customers, and cost per resolved contact is the right metric. Intake means answering a stranger who is deciding whether to hire you, where a missed call is a lost job rather than a delayed resolution, and cost per booked job is the right metric. If your calls are intake, price it as revenue capture and not as an overhead line. That case is worked through in how to get roofing leads after a storm, and the managed version is our roofing call center.
For the pricing models themselves across the whole category, including per-appointment and per-lead structures that this page does not cover, see call center outsourcing cost. For the geography question that sits underneath most rate differences, see what a nearshore call center is and nearshore vs offshore.
Limitations of everything above
The wage figure is the most recent BLS OEWS estimate for customer service representatives (SOC 43-4051), is a national median, and is a base wage; the loaded-cost multiplier discussed here is an assumption the buyer supplies and is not sourced to BLS. The abandoned-call rule is quoted from the FTC Telemarketing Sales Rule and governs outbound telemarketing, not inbound customer service, and is cited here to show that no inbound equivalent exists. The disposition figures are first-party, measured on one floor over one two-month window on outbound roofing appointment setting; they are not an inbound support benchmark and the six codes shown cover 46.9% of that corpus rather than all of it. No vendor pricing range is stated anywhere on this page as a measurement, and no client, agent or campaign is named.