What Ecommerce Support Calls Actually Cost

Phone support is the line item ecommerce teams model worst, because the three options are quoted in three different units and nobody converts them before deciding.

Most brands answer the phone one of three ways: they hand it to an outsourced ecommerce call center, they staff it in-house, or they put a voice agent in front of it. Those get quoted per agent hour, per annual salary, and per minute respectively. You cannot compare them in those units, and the gap between the headline number and the real one is different in each case.

This post converts all three to the same unit, then shows why the conversion is not the part that decides it.

Normalize everything to the talk minute

The talk minute is the only unit all three share. It is also the unit your call volume is already in, because your telephony bill has it.

Published outsourced rates, from CloudTalk’s 2026 pricing breakdown:

Model	Published rate	Region
Per agent hour	$8 to $14	Offshore (India, Philippines)
Per agent hour	$13 to $23	Latin America
Per agent hour	$28 to $65	US and Canada
Per minute	$0.35 to $0.55	Low-cost international agencies
Per minute	$0.75 to $1.35	US, Canada, and European agencies

SpeechifyAI voice agents are $0.075/min on Starter, $0.07/min on Pro, and $0.068/min on Scale, with the language model, speech-to-text, text-to-speech, and orchestration included and no passthrough fees. That is roughly a fifth of the cheapest offshore per-minute quote and a fifteenth of a mid-range US one.

Those ratios are real but they are also the least interesting thing in this post, so hold them loosely until the containment section.

Reconciling the per-hour and per-minute quotes

An hourly rate and a per-minute rate are not two prices for the same thing. The difference is who carries occupancy risk.

A seat billed at $12/hour does not give you 60 talk minutes. It gives you 60 minutes of a person being available, of which some fraction is spent on calls. At 70% occupancy that is 42 talk minutes, so your effective cost is $0.29 per talk minute. At 50% occupancy, which is what unpredictable ecommerce volume actually looks like outside peak, it is 30 talk minutes and $0.40.

That is why the per-minute quotes land where they do. A vendor quoting $0.35 to $0.55 per minute has priced the occupancy risk and added margin for carrying it. A vendor quoting $12/hour has left that risk with you.

The practical read: take a per-minute quote when your volume is spiky, take a per-hour quote when it is flat and you can keep seats busy. Ecommerce volume is almost never flat, which is a large part of why seasonal support economics are so bad.

What the outsourced number leaves out

Three things, consistently:

• Setup and minimums. Onboarding fees, minimum monthly commitments, and minimum seat counts are quoted separately from the rate and are usually the first thing a small brand runs into.
• Billed minutes versus talk minutes. Check which one the contract meters. Hold time, wrap-up, and after-call work are talk-adjacent and are frequently billable.
• Ramp and attrition. Every agent who leaves takes your product knowledge with them, and the replacement is billed at full rate while learning your returns policy.

None of these make outsourcing a bad decision. They make the quoted rate the floor rather than the number.

What the in-house number leaves out

We are not going to quote you an average support salary, because the only figure that matters is your own fully loaded one, and the loading is where the surprise lives.

The conversion is:

cost per talk minute = fully loaded hourly cost / (60 x occupancy)

Where fully loaded means base pay plus payroll tax, benefits, the helpdesk and telephony seat licences, the fraction of a manager’s time that supervises the queue, and recruiting plus ramp amortized over expected tenure. Run that number for your own team before comparing anything. Teams that do it for the first time usually find the loaded figure is 1.25 to 1.4 times base pay, and that occupancy is well below what they assumed.

In-house wins on one axis that does not appear in any of these numbers: a person who has handled your returns policy for two years resolves an angry call in a way that nothing else in this table does. Budget for that on the calls where it matters, rather than spreading it across every order-status query.

The number that actually decides this

Not the rate. The containment rate: the share of calls that finish without a human.

A voice agent that escalates still costs you the minutes it spent qualifying the caller, and then you pay the human for the whole call anyway. At low containment you are paying twice.

The arithmetic, for a 4-minute call where the agent spends 1.5 minutes before transferring, against an offshore human at $0.45 per minute:

Containment	Blended cost per call	Versus all-human ($1.80)
30%	$1.42	21% cheaper
50%	$1.09	39% cheaper
70%	$0.77	57% cheaper
90%	$0.44	75% cheaper

Against a mid-range US agency at $1.05 per minute, the same containment rates give $3.10, $2.29, $1.49, and $0.68 per call, against $4.20 all-human.

The formula, so you can run it on your own numbers:

blended = c(aT) + (1 - c)(aq + hT)

c = containment rate
a = voice agent rate per minute
T = full call length in minutes
q = qualification minutes before transfer
h = human rate per minute

Two things fall out of this that are worth more than the rate comparison:

1. Moving containment from 30% to 70% saves more than switching from a US agency to an offshore one. Containment is the lever. The per-minute rate is a rounding adjustment on top of it.
2. A cheaper voice agent barely moves the blended number at low containment. At 30% containment, halving a from $0.07 to $0.035 changes the blended cost by about four cents. Vendor selection on headline per-minute price is optimizing the wrong term.

Where the voice agent loses

Three cases, and they are worth naming because pricing pages do not:

• Low containment on a hard call mix. If your volume is dominated by disputes, damaged goods, and anything with a refund decision attached, containment will be low and the double-pay effect eats the saving. Automate the order-status band first and leave the rest alone.
• Calls where escalation is the product. A high-AOV brand whose phone line exists so buyers can talk to a human is not solving a cost problem. Do not let a per-minute comparison talk you out of a deliberate service position.
• Anything needing an identity decision. Refunds to a changed payment method, address changes on an in-transit order, and account recovery all end at a human on purpose. Route them there fast rather than qualifying for 90 seconds first.

What the volume mix actually looks like

Every vendor in this category quotes a statistic for what share of ecommerce support contacts are order-status questions. Across published sources those estimates run from 10% to 25%, 25% to 40%, 30% to 50%, and 40% to 60%, all presented as industry benchmarks.

That spread is not a disagreement about ecommerce. It is evidence that almost nobody is measuring it the same way, and it means any of those numbers is useless for your budget.

Pull your own. Tag last quarter’s calls by first intent, split order-status from returns from pre-sales, and get the average handle time for each. That is a half-day of work and it turns every number in this post into a decision rather than a range. It also tells you which band to automate first, which is the containment question from the previous section restated.

Putting a number on it

Take your own call volume, your own average handle time, and your own loaded human cost. Run the blended formula at a containment rate you can defend, not the one on a vendor’s homepage. Then decide.

If it clears the bar, you can start free with 60 voice agent minutes and 50,000 text-to-speech characters a month, no card, which is enough to point an agent at your order lookup and measure a real containment rate instead of estimating one. Bring your own carrier, SIP, or Twilio number at no extra cost. See the docs for the tool-calling setup, and voice agents for retail and ecommerce for what the deployed shape looks like.

Two related reads: the voice agent platform comparison puts our per-minute rate against eight competitors on assembled rather than headline cost, and choosing text-to-speech for support chatbots covers the case where you already have the conversation and only need it spoken.

One disambiguation, because it causes real confusion: this is SpeechifyAI at speechify.ai, the voice AI developer platform. speechify.com is a separate consumer reading app. Same company, different product, and only one of them has an API.

FAQ

How much does an outsourced ecommerce call center cost? Published 2026 rates run $8 to $14 per agent hour offshore, $13 to $23 in Latin America, and $28 to $65 in the US and Canada. Quoted per minute instead, low-cost international agencies run $0.35 to $0.55 and US, Canadian, and European agencies run $0.75 to $1.35. Setup fees, monthly minimums, and billing on connected rather than talk minutes sit on top of all of those.

Is a voice agent cheaper than outsourcing customer service? Per minute, yes, and by a wide margin: SpeechifyAI voice agents are $0.068 to $0.075 per minute all-in against $0.35 as the cheapest offshore human rate. But blended cost depends on containment. At 30% containment against an offshore agency the real saving is about 21%, not 80%, because escalated calls cost you the agent’s qualification minutes plus the full human call.

What is a good containment rate for an ecommerce voice agent? It depends entirely on call mix rather than on the platform. Order-status and delivery-window questions contain well because they are a database lookup read aloud. Refund disputes, damaged goods, and anything requiring an identity decision do not. Measure containment separately per intent, because a single blended figure hides which band is worth automating.

Should I pay per agent hour or per minute? Per minute when your volume is spiky, per hour when it is flat enough to keep seats busy. An hourly seat only converts to talk minutes at your occupancy rate, so a $12 hour is $0.29 per talk minute at 70% occupancy and $0.40 at 50%. A per-minute quote is the vendor pricing that risk and charging you a margin to carry it.

What share of ecommerce support calls are order-status questions? Published estimates range from 10% to 60% depending on which vendor you ask, which is too wide to budget against. Tag your own last quarter by first intent and measure it directly. The exercise takes about half a day and also tells you which intent band to automate first.