Support Cost Per Contact Benchmarks by Vertical and Volume

Created time

Sep 23, 2026 06:39 AM

Title length (<60)

Author

Mike Heap

Last optimised

Ecomm?

What does a contact cost at each handled share?

Here is the short version, priced for a B2C team at 30,000 contacts a month: the four handling rates in our data, blended out under the two ways of running the AI we model below, set against a team running no AI at all.

Handled share	Blended cost per contact, route by intent	Blended cost per contact, AI attempts everything
No AI, a person answers every contact	$7.00	$7.00
56%, field 25th percentile	$3.15	$6.27
66%, retail / ecommerce / DTC median	$2.46	$4.87
70%, field median	$2.18	$4.31
80%, field 75th percentile	$1.49	$2.91

Why does a single cost per contact benchmark break at volume?

⚡

TL;DR: Published cost per contact figures average two prices into one. Once an AI agent finishes part of your volume, that average stops describing your bill.

The definition is the problem. MetricNet defines cost per ticket as the total monthly operating expense of a service desk divided by the monthly ticket volume.

Klipfolio's contact center version does the same division over contacts, so both are averaging everything you spent over everything that arrived. We run into that limit every time we try to price a rollout from a customer's existing reporting.

The numbers you will hit first are the IT ones. IT service desks publish a tier ladder: about $22 for a Tier 1 ticket, $70 at Tier 2 and $104 or more at Tier 3, against a blended North American IT desk range of $6 to $40 or more, which ScreenMeet attributes to HDI benchmarking.

That ladder prices an internal IT help desk fixing laptops for employees, so it does not transfer to a B2C team answering where-is-my-order. We go through the ones that do in our post on the customer service metrics that actually matter.

There is a fair case for the consensus figure, and MetricNet makes it:

"A higher than average Cost per Ticket is not necessarily a bad thing, particularly if accompanied by higher than average customer satisfaction and service levels. Conversely, a low Cost per Ticket is not necessarily good, particularly if the low cost is achieved by sacrificing service levels or customer satisfaction." MetricNet

That caveat is right, and most sources skip it. A single blended figure is fine for a year-on-year trend line and useless for a decision; the trend line is the only way we ever use one.

Stat callout headed "One contact, two prices", with a large red ~60x above two figures: $7 when a person finishes the contact and about $0.11 when the AI finishes it on chat.

Two support teams can publish the same low cost per contact for opposite reasons. One has simple volume and a good help center. The other has customers who give up before they get an answer.

The average cannot tell those two teams apart. Your board will not be able to either. We ask which one a team is before we quote anything, because the answer changes what the number is worth.

Klipfolio goes further than most sources and declines to publish a band at all, on the grounds that benchmarks depend on industry, geography, wage rates, contact mix and complexity, so outside numbers should be treated as directional. That leaves you with nothing to aim at. What we hand customers instead is the two prices underneath the average.

Cost per contact vs cost per resolution: what is the difference?

A contact is any message a customer sends you. A resolution is one that ended with their question answered. We keep the two apart in every number we report.

On a team where a person handles everything, the two are close enough that nobody bothers separating them. An AI agent in the middle pulls them apart: a contact it finishes costs you almost nothing, and a contact it tries and then hands over costs you two touches.

The label moves the number. In our dataset, deployments reporting a resolution metric come in at a 72.5% median across 108 rated rows, while deployments reporting an automation metric come in at 61% across 52. Check which label a vendor uses before you compare their number to yours.

So before you compare your figure to anyone else's, check what they counted. Our post on containment, deflection and resolution takes that split apart properly.

We use the strict version of that count: a conversation is resolved when the AI handled it without escalating to a human. The escalation is the exact moment the cost changes, so counting it strictly keeps the two prices apart.

The Two-Price Benchmark: what should your cost per contact be?

⚡

TL;DR: Your benchmark is drawn by four figures: what a person costs you per contact, what the AI costs you per contact, the share the AI finishes end to end, and what a handover costs on top.

I find it easier to think about a contact as having two possible prices, one for a person finishing it and one for the AI, and on the figures we model the gap runs about sixty to one, $7 against roughly eleven cents. Your cost per contact is whatever blend of the two you actually bought last month.

What you need to know is the mix you are buying and what the handovers cost you on top of it. I call the four figures that answer that the Two-Price Benchmark: the human price, the AI price, the handled share, and the handover surcharge. Work out all four and the number is yours, built from your own mix.

Breakdown diagram with a red "Blended cost per contact" label connected to four boxes: human price $7, AI price about $0.11, handled share 66% for retail, ecommerce and DTC, and a handover surcharge of 2x, or $14 instead of $7.

The human price

This is what one contact costs when a person finishes it. Nearly every published benchmark is quoting this price, whether it says so or not.

Each channel has its own published band. The Office Gurus puts ecommerce and retail chat at $4 to $7 and email at $5 to $9, describing its numbers as ranges observed across mid-market and enterprise operations in North America. LiveAgent publishes a single point figure of $7.16, close enough to the figure we work from.

Count's by-segment table is wider at $5 to $20 for ecommerce, with SaaS B2B at $15 to $45 and fintech at $25 to $60. Voice costs more again on every grid that splits by channel, so your channel mix moves this number before your vertical does. LiveAgent's point figure sits inside The Office Gurus' $5 to $9 email band and Count's band starts in the same place, so we assume $7 for a person-finished contact, the midpoint of that email band.

We watch this one closely, because wages set the ceiling on what any amount of automation can save you. Agents Republic builds the number up from the wage: benefits run close to 30% of an employee's total value, so a $22 hourly salary costs about $31 before that agent takes a single contact.

The same source puts the median assisted contact at $13.50 against $1.84 for one the customer answers from an article, citing benchmarking studies. Our customers move that labor number by hiring offshore or nearshore, and it does not move far.

The AI price

This is what one contact costs when the AI finishes it end to end.

The Office Gurus puts the published field band for a self-served contact, one the customer finishes in your help center or with a bot and no person ever sees, at $0.10 to $0.40. That band prices self-service, which is the closest published comparison to an AI-finished contact, and ours is at the bottom of it. My AskAI bills per ticket, at roughly $0.10 a ticket on helpdesk chat and roughly $0.15 on email, whether or not the AI got there.

At 30,000 contacts a month you would run on our Scale plan at $499, which covers 2,000 credits and charges ten cents a credit after that. A chat ticket costs you one credit and a typical email ticket, with its follow-ups, costs about one and a half, so that $499 covers roughly 2,000 chat tickets or about 1,300 email ones. An email-led team should budget nearer fifteen cents a ticket than ten.

The usage add-ons, things like tagging and actions, are charged separately on every plan. And this is the AI charge only: you keep paying for your helpdesk seats whichever AI agent you run, so our line is added on top of the stack you already have.

Across the middle half of the field, from 56% to 80%, our charge per contact the AI handles moves from about $0.118 to about $0.112, roughly half a cent. Over that same range, routing by intent, the human side of the bill swings by more than $50,000 a month.

You can test all of this before you pay for any of it: our trial runs 30 days with every feature unlocked, unlimited tickets and no card.

The handled share

This is the fraction of your contacts the AI finishes end to end. Published cost benchmarks leave that half of the sum out entirely.

Our AI resolution rate benchmarks study covers 195 rated deployments across 38 vendors, ours and everyone else's. The field median is 70%, and I start from it. The 25th percentile is 56% and the 75th is 80%, so the middle half of deployments finish between just over half and four fifths of their volume.

The number a B2C team needs is worse than the headline. Retail, ecommerce and DTC deployments come in at a 66% median across 43 rated rows (the largest industry group we hold), with a 25th percentile of 50% and a 75th of 77%. That is four points below the field median. If you run B2C support, the line to measure yourself against is 66%, because the 70% headline comes from a field with a lot of SaaS in it.

Vendors publish their wins, so take any field median built from published deployments, ours included, with a grain of salt: it is a self-selected ceiling, and the real field average is likely lower than what you see. Read every median next to its row count too: 195 rows will hold up, a three-row industry cut will not.

All of it holds at 30,000 contacts a month, because nothing in our data says the rate falls away as volume climbs.

One third-party deployment in that 195-row benchmarks corpus, a vendor other than us, handles around 105,000 tickets a month and still posts 64%. Across the field the rate barely tracks volume at all.

Company size does little either, staying between about 65% and 72% across every employee and revenue band we hold. Industry moves it more. And among teams in the same vertical the spread is wider still. Your setup, meaning what you have connected and what you let the agent attempt, moves the number more than your industry does.

The gains come in stages, and each stage takes more work than the last. Connecting and tidying the knowledge you already have gets most of our rollouts to somewhere between 40% and 70%. If none of it is written down yet, you can build that knowledge out of the tickets you have already answered rather than writing a help center from scratch.

Incremental work on top of that adds another 5% to 10%. The next real jump is giving the agent access to your order status and account data, worth another 15% to 50% for a team starting low, with ecommerce at the top of that range.

I would treat any vendor promising all of it on autopilot with suspicion. Someone on your team still has to connect the knowledge and tidy it, and once that is done it does not need repeating for every new contact.

Those benchmark numbers are ours. We build the data from published deployments across 38 vendors, ours included, and we report our rolling rate at 72%. We use it in aggregate against the field and not vendor by vendor, because the definitions underneath each vendor's number are not the same.

The handover surcharge

A contact the AI attempts and then passes to a person costs you more than one a person took from the start. The AI has already tried the quick fixes, so what reaches your team is the harder end of the problem, and whoever picks it up has to read the whole conversation before they can reply. On its own that surcharge can reverse a saving.

The published multiplier comes from HDI, attributing MetricNet: a ticket passed up a tier can cost two to three times one that was answered first time. It is an IT service desk figure and the only published one there is, so treat it as directional until you have measured your own. I model the conservative end, two times, which puts a handed-over contact at $14 against $7 for one that went to a person directly.

Our rollouts show how far that surcharge varies. An iGaming operator we work with escalates 32% of its chats while resolving 44% with the AI, so roughly a third of the conversations the agent starts end up with a person. A digital-goods marketplace, also running on Intercom, escalates 10%, so its handovers cost it a small fraction of what the iGaming operator's cost.

Pushing your handled share up without watching your handover rate can raise your blended cost. If the handover rate stays high while the resolution rate climbs, the second touch absorbs most of what you saved on the first.

At 30,000 contacts a month, with the human price held at $7 and chat billed at about ten cents a ticket on our Scale plan, four handling rates from our data blend out under two ways of running the AI, set against running none at all.

The two differ in who decides what the AI sees. Route by intent means you choose up front which kinds of question the AI takes, such as order status or your returns policy, and everything else goes straight to your team at the normal $7. AI attempts everything gives the AI first go at every contact, and whatever it cannot finish reaches your team as a handover at $14.

We bill for every ticket the AI replies to, finished or not, so under that second policy all 30,000 are billable and our charge holds at $3,299 on every row. That charge prices chat at one credit a ticket; an email-led month would sit higher. I start the table with a no-AI row, the same 30,000 contacts with every one answered by a person at $7, because that is the bill you are actually replacing.

Handled share (rated rows)	AI charge, route by intent	Blended cost per contact, route by intent	Monthly total, route by intent	AI charge, AI attempts everything	Blended cost per contact, AI attempts everything	Monthly total, AI attempts everything
No AI, a person answers every contact	$0	$7.00	$210,000	$0	$7.00	$210,000
56%, field 25th percentile (195)	$1,979	$3.15	$94,379	$3,299	$6.27	$188,099
66%, retail / ecommerce / DTC median (43)	$2,279	$2.46	$73,679	$3,299	$4.87	$146,099
70%, field median (195)	$2,399	$2.18	$65,399	$3,299	$4.31	$129,299
80%, field 75th percentile (195)	$2,699	$1.49	$44,699	$3,299	$2.91	$87,299

Every row comes in under the no-AI line on both policies, and route by intent comes in lower still, because contacts the AI was never going to finish skip the handover. The retail and ecommerce row is where most B2C teams will actually sit.

A B2C team at the median for its vertical pays about $2.46 a contact, against $7 with no AI and the $5 to $9 the published benchmarks quote. Those benchmarks are pricing a contact a person finished, and at 66% a person finishes about a third of yours.

That team, at the same handling rate, pays $4.87 if every contact the AI does not finish arrives at a person as a handover, nearly double the route-by-intent figure and still under the $7 it would pay with no AI. If you want to run it with your numbers in it, our AI support agent ROI calculator does the blend for you.

What does this look like in real rollouts?

⚡

TL;DR: Four rollouts at handled shares from 21% to 73%, including one held deliberately at the bottom of the band, and what each one does to the bill.

Four of our rollouts sit at very different points on that range.

Spectrum chart of four AI support rollouts plotted by handled share: 21% with about 105 hours saved, 44% with 170, 58% with 1,768, and 73% with about 5,650 marked in red as the highest, on a rail labeled lower to higher handled share.

One of our own accounts, a high-volume prop-trading platform, runs at about 105,000 Intercom tickets a month, a 73% AI resolution rate and 68% AI CSAT, saving roughly 5,650 hours a month. Price those hours at your loaded agent cost and you have a cost avoided without needing a cost per contact column at all.

A digital-goods marketplace resolves 58% of 36,413 conversations in a 30-day window at 92% AI CSAT, saving 1,768 hours, and escalates only 10%. It sits above 30,000 conversations a month and eight points under the retail and ecommerce median. Its low escalation rate keeps its blended cost near the route-by-intent figure.

An iGaming operator we work with resolves 44% across 4,687 AI conversations at 72% AI CSAT and 170 hours saved, with a 32% escalation rate on a progressive rollout. A third of the conversations the agent starts finish with a person, so it pays the surcharge on a third of its volume.

An ecommerce marketplace on Zendesk sits at 21% across roughly 6,100 Zendesk tickets a month at 66% AI CSAT and about 105 hours saved, and that 21% is a deliberate choice. Their escalation guidance pushes around 77% of tickets to a person because they want marketplace disputes handled by one, and their cost per contact is meant to be high as a result.

We run all four on the same product. The spread from 21% to 73% comes from what each business lets the agent attempt and how much of its data the agent can reach.

What should you do this week?

⚡

TL;DR: Work out your two prices before you compare yourself to anyone. Budget an afternoon for it.

We run this exercise with customers in their first week, and all four steps work on the helpdesk you already have, whether that is Zendesk, Intercom, Freshdesk, Freshchat, Gorgias or HubSpot. None of them needs you to buy anything. The timings assume you paste the exports into whichever AI tool you already have open; by hand they take considerably longer.

Split last month's contacts by who finished them - export the month and group it by whether a person touched the conversation. What comes out is your real handled share, and it is almost never the number on your dashboard. About 30 minutes.

Price the two halves separately - cost the person-finished contacts against your loaded agent cost and the AI-finished ones against your AI bill. The gap between those two methods is what finance has been asking about all along. We cannot work it out for you from the outside. About 45 minutes.

Count the handovers separately - you will come out with a third cost figure, the price of a contact the AI started and a person finished, which is higher than either of the other two on its own. About 20 minutes.

Compare against the row for your vertical - if you run B2C, the median in our data is 66%. About 15 minutes.

Call it two hours, and you come out with all four figures for your business. Once you hold them, our guide to reducing customer support costs goes through what to do about each one.

Video preview — I Let AI Agents Resolve 10,000 Support Tickets, Here's How Much It Cost

If you already run My AskAI, step one is done for you: the Insights view shows what the AI finished and what went to a person without you exporting anything at all.

When does the Two-Price Benchmark not apply?

⚡

TL;DR: Below a few thousand contacts a month, or where the contact is the sale, the blended number answers the wrong question.

Low volume is the common case where this is the wrong tool.

Under a few thousand contacts a month your support cost is mostly fixed, and the blend tells you nothing you did not already know. Count's worked example is five agents and 1,500 contacts a month, producing $17.33 a contact from a $26,000 base.

At that size, dividing the budget by the contact count is the sensible method, and we tell teams that small not to bother with the blend. The Two-Price Benchmark is worth running from the point where a handling rate changes your staffing.

Then there are high-value contacts (a renewal call, a win-back offer, a big order). Where a contact is a retention moment or a sale, driving cost per contact down destroys value.

Keep that volume out of your own blend. A lower cost per contact in that segment means fewer customers came back, which is not a saving. We like how Entrepreneur lays out the arithmetic on it:

"Acquiring a new customer costs five to seven times more than keeping an existing one. A customer who has a poor service interaction and does not complain does not become a satisfied customer; they become a silent churn risk." Entrepreneur

Contact Center Pipeline goes at it from another angle, arguing that an accurate average can still mislead. If your contacts carry revenue, benchmark the revenue instead.

The last case is the data behind the table. Our cut of it runs by company size, with no volume cut in it at all. Size barely moves the handling rate, holding between about 65% and 72% across every band, and the largest deployments in it still post rates in the field range.

It cannot show you what 30,000 contacts a month specifically does to your rate.

None of this makes the table a target to chase. That marketplace's 21% is the number we built with them on purpose, and their cost per contact is meant to sit above the table.

The takeaway

⚡

TL;DR: Cost per contact is a band. Two prices, a handled share and a handover surcharge draw it, and the four figures you work out place you inside it.

The figure depends on a mix only you can see, so there is nothing useful to look up. Draw the band yourself from four numbers: your human price, your AI price, the share the AI finishes end to end, and what a handover costs on top.

If you do one thing, pull last month's contacts and split them by who finished them. Your human price, your AI price, and your handover surcharge all depend on that split to mean anything.

The band itself will keep sliding as handling rates rise across the field. We publish and refresh the underlying data, so check your number against it whenever your budget cycle comes round.

FAQs

What is a good cost per contact for customer support?

The published benchmarks give a spread. Blended B2C support is commonly put at $5 to $9 a contact and self-service at $0.10 to $0.40, per The Office Gurus' ecommerce and retail row, which also splits that by channel at $4 to $7 for chat, $5 to $9 for email and $7 to $12 for voice. Every one of those figures prices a contact a person finished, so your number depends on how many of yours a person finishes. At the retail and ecommerce median of 66% we see the blend land near $2.46 a contact, and near $4.87 if every un-finished contact arrives at a person as a handover. If you run B2C, that retail median is the line to use.

What's the average cost per support ticket in ecommerce?

Treat the working band as a per-channel one: $4 to $7 for a chat contact and $5 to $9 for an email one. The Office Gurus splits ecommerce and retail by channel, and those are the bands we plan against.

Channel	Cost per contact
Self-service	$0.10 to $0.40
Chat	$4 to $7
Email	$5 to $9
Voice	$7 to $12

A second by-segment grid, from Count, is wider at $5 to $20 for the same segment, because its band stretches from a two-person team to an enterprise operation.

Cost per contact vs cost per resolution: what's the difference?

A contact is any interaction a customer starts (a question, a complaint, a simple status check). A resolution is one of those that ends with the customer's problem solved. An AI agent in the middle pulls the two apart, because a contact it finishes costs almost nothing and one it hands over costs two touches. The label changes the number too: in our data, deployments reporting a resolution metric come in at a 72.5% median across 108 rows, while those reporting an automation metric come in at 61% across 52, and we count a conversation resolved when the AI handled it without escalating to a human.

How much does AI reduce cost per ticket?

At 30,000 contacts a month, a team with no AI pays $7.00 a contact. Routing by intent, the field 25th percentile of 56% brings that to $3.15, and moving to the 75th percentile of 80% takes it to $1.49, a little over half again. The saving comes from the share of contacts a person no longer touches; the AI rate itself moves by about half a cent over the same range. Our rollouts follow the same pattern.

What is cost per call?

Cost per call is the phone-channel version of the same metric: total contact center costs for a period divided by the calls answered in it. There is a worked calculation here, and a good summary of cost per call as a contact center KPI. The Office Gurus publishes voice at $9 to $16 a call across industries and $7 to $12 for ecommerce and retail, the highest of any channel either way, which is why voice volume is where a cost program usually starts.

How much can AI reduce customer support costs?

Across our rollouts the handled share runs from 21% to 73%, and the hours saved run from about 105 a month to about 5,650. Price those hours at your loaded agent cost and you have the saving. The range is that wide because the result tracks how much of your knowledge and your systems the agent can reach, and how much of your volume you let it attempt in the first place.

What is cost per contact in a call center?

In a contact center, cost per contact is the total operating cost for a period divided by the contacts handled in it. MetricNet's cost-per-ticket definition applies the same division over tickets. Both assume a person handled the contact, and we price the two halves separately because an AI agent finishing most of the volume breaks that assumption.

Support Cost Per Contact: How to Benchmark Yours at 30,000+ Contacts

What does a contact cost at each handled share?

Why does a single cost per contact benchmark break at volume?

Cost per contact vs cost per resolution: what is the difference?