.png)
Last updated: July 31, 2026
Every support leader has a cost-per-ticket number somewhere in a slide deck. Very few can defend it when finance pushes back, because the number in the deck measures contacts, and the budget conversation is about problems solved. Those are not the same unit, and the gap between them is where most support cost quietly lives.
Cost per resolution closes that gap. It tells you what it actually costs your company to make a customer's problem go away, not what it costs to send one reply. This guide covers how to calculate it properly, which benchmarks to compare yourself against, the four drivers that push the number above benchmark, and how AI-handled resolution changes the arithmetic for a B2B support organization.
What's in this guide
- What cost per resolution actually measures
- Benchmarks by industry and channel
- Four drivers that push your number above benchmark
- How AI resolution changes the math
- A 30-day plan to get a number you can defend
- Frequently asked questions
What cost per resolution actually measures
Cost per resolution is the total cost of fully resolving one customer issue, divided across every issue your team closed in the period. "Total cost" is broader than most teams assume: agent salaries and benefits, tooling licences, QA and management overhead, training, and the cost of escalations. In service-desk cost structures, agent labour typically accounts for 70–80% of cost per ticket, with tooling and overhead making up the rest. That single fact explains why cost programmes that only renegotiate software contracts rarely move the headline number.
The formula, and the version that matters
Most teams calculate this:
Cost per contact = total support operating cost ÷ total contacts handled
That is a useful operational number, but it answers the wrong question. The version that survives a budget review is:
Cost per resolution = total support operating cost ÷ unique issues resolved
The difference between those two numbers is your repeat-contact overhead. If customers need multiple touches before a problem is actually closed, your cost per resolution is a multiple of your cost per contact. One benchmark analysis puts the average at 2.3 contacts per issue — meaning a team reporting $12 per contact is really spending closer to $28 per problem solved.
Why B2B teams get burned worst
In B2B support, issues are fewer but heavier: integration failures, permission disputes, billing exceptions, multi-system troubleshooting. Handle times vary wildly, one account can generate tickets from a dozen users, and escalations pull in engineering time that never appears in a support cost model. Averages hide all of it. A blended cost per contact across a queue where 80% of tickets take five minutes and 20% take two hours tells you almost nothing about where your money goes.
Key takeaway: Report both numbers. Cost per contact is your efficiency metric. Cost per resolution is your unit economics. If the two are drifting apart, your resolution quality is deteriorating even when your efficiency dashboard looks green.
Benchmarks by industry and channel
Before you can say your cost is high, you need to know what "normal" looks like for your shape of business. Cross-industry analysis shows cost per ticket spanning roughly $2.70 for simple retail interactions up to $60 for complex B2B support cases, with a global baseline of about $6–$7 per contact across all sectors. The spread reflects handle time, issue complexity, and how specialised your agents have to be.
Channel matters just as much as industry. Voice is consistently the most expensive path to a resolution, self-service the cheapest — when it genuinely resolves rather than simply deflecting.
| Channel or segment | Typical cost per ticket | Why |
|---|---|---|
| Self-service | $1–$4 | No agent time; cost is content maintenance |
| $8–$15 | Asynchronous, but slow to close | |
| Live chat (human) | $10–$16 | Concurrency helps; real-time attention costs |
| Phone | $17–$25 | One agent, one customer, longest handle time |
| SaaS support (blended) | $18–$35 | Technical complexity, specialised agents |
| B2B enterprise support | $30–$60 | Long handle times, contractual SLAs, escalations |
Channel and self-service figures above are drawn from cross-channel cost-per-resolution benchmarks, which also cite Gartner data putting self-service at $1.84 per contact against $13.50 for assisted channels. That roughly sevenfold differential is the single largest structural lever in support economics — and the reason ticket deflection programmes get funded so readily, even when they are measured badly.
Four drivers that push your number above benchmark
When cost per resolution runs above benchmark for your segment, the cause is almost never "agents are slow." It is structural, and it is usually one of four things.
1. Repeat contacts
This is the biggest and least-tracked multiplier. Industry FCR sits at an aggregated 69% across all industries, which means roughly one in three issues needs a second touch before it closes. Reopened tickets add more: the average reopen rate is 5.4%. Every one of those touches carries full agent cost while producing zero additional resolutions. Improving first contact resolution from 55% to 70% removes an estimated 25–30% of repeat contacts without automating a single ticket.
2. Inconsistent answers
Repeat contacts usually trace back to knowledge, not effort. When three agents answer the same billing question three different ways, one of those answers generates a follow-up. This is why teams that review conversations at scale typically find a handful of ticket categories where agent guidance conflicts. Fixing the source content is cheaper than fixing the symptom, which is why a well-maintained AI-powered knowledge base shows up as a cost metric, not just a CX one.
3. Misrouting
A password reset sitting behind three integration escalations in a two-hour SLA queue costs you queue position, context switching, and specialist attention it never needed. Routing that matches issue type to the cheapest path capable of actually resolving it — automation, self-service, tier one, or specialist — removes that overhead without touching headcount. Organisations with clearly defined support tiers reach 72% FCR versus 45% for those without structured tiering.
4. Deflection that does not resolve
The most expensive automation is automation that answers without solving. It removes the contact from your queue, closes the ticket, and produces a second contact next week — often on a more expensive channel, and often with an angrier customer. If your containment rate is climbing while cost per resolution is flat, you are moving contacts around rather than eliminating work.
How AI resolution changes the math
The reason cost per resolution has become a boardroom metric is that the cost of one resolution now varies by nearly two orders of magnitude depending on who or what handles it. AI-handled tickets land at roughly $0.50–$2.37 per resolution when the AI takes full ownership of the issue — comparable to self-service, but with the resolution quality of an experienced agent.
Applied to the top 20% of ticket types by volume, which are usually the most repeatable, that shifts those categories from $18–$35 down to a couple of dollars — an 85–95% reduction for that segment. Combine it with FCR improvement and answer-consistency work, and total support operating cost typically falls 20–35% within 6–12 months.
Two conditions decide whether you actually see those savings. First, the AI must complete the job — look up the order, apply the credit, reset the entitlement — rather than summarise a help article. Second, escalation rate has to stay low. If the AI hands more than half of its conversations to a human after gathering context, you are paying for two resolutions and getting one. This is the design principle behind Darwin AI's Eva, a customer experience worker that handles inbound conversations end to end across WhatsApp, email, and chat, and escalates with full context only when the issue genuinely needs a person.
The blended number is what finance cares about
Nobody automates everything. What you are actually managing is a blend: some share of issues resolved by AI at near-zero marginal cost, the rest by agents at $18–$35 or more. Benchmarks suggest SaaS teams achieving AI resolution rates above 40% can pull blended cost per ticket into the $8–$15 range. Track the blend monthly, because it is the number that tells you whether your automation investment is compounding or stalling. If you are building the business case from scratch, our guide to measuring the ROI of AI automation covers how to frame it for a CFO.
A 30-day plan to get a number you can defend
You do not need a data warehouse for this. You need four weeks and an honest spreadsheet.
Week 1 — assemble total cost. Pull fully loaded agent cost (salary, benefits, employer taxes), support tooling licences, the management and QA share of payroll, and training. Resist the urge to exclude overhead because it feels unfair; finance will include it.
Week 2 — count both denominators. Total contacts handled, and unique issues resolved. Most helpdesks can approximate the second by collapsing tickets from the same requester on the same topic within a rolling window. Divide total cost by each. The ratio between the two results is your repeat-contact multiplier.
Week 3 — segment by ticket type. Take your top ten categories by volume and compute cost per resolution for each, using category-level handle time rather than a blended average. This is where the model earns its keep: you will usually find two or three categories consuming a disproportionate share of capacity, and one or two high-volume categories that are trivially automatable.
Week 4 — model one intervention. Pick the highest-volume, lowest-complexity category and model what happens if AI resolves 60% of it end to end. Show the category cost before and after, the effect on blended cost per resolution, and what the freed capacity gets redeployed to. That single before-and-after is a more persuasive artefact than any vendor deck.
Review the model quarterly. Ticket mix drifts, product complexity grows, and a model that was accurate in Q1 is usually wrong by Q3.
Cut cost per resolution without cutting service quality
Darwin AI's Eva resolves inbound customer conversations end to end — on WhatsApp, email, and chat — and escalates with full context only when a human is genuinely needed.
See how Eva worksFrequently asked questions
How do you calculate cost per resolution?
Divide total support operating cost — fully loaded agent pay, tooling, QA, management overhead and training — by the number of unique issues resolved in the same period. Calculate cost per contact alongside it. The gap between the two reveals your repeat-contact overhead, which averages 2.3 contacts per issue across benchmarked teams.
What is a good cost per resolution for a B2B company?
It depends heavily on complexity. Benchmarks place SaaS support at $18–$35 per ticket and B2B enterprise support at $30–$60. Teams with AI resolution rates above 40% typically report blended figures of $8–$15. Compare against your own trailing twelve months before comparing against anyone else.
Is cost per resolution better than cost per ticket?
They answer different questions. Cost per ticket measures handling efficiency; cost per resolution measures unit economics. Because industry FCR averages 69%, roughly a third of issues generate more than one ticket, so cost per ticket systematically understates what each problem costs you. Track both.
Does AI automation actually reduce cost per resolution?
Only when it resolves rather than routes. AI that triages and hands off reduces cost marginally. AI that owns the issue end to end reduces cost for those categories by 85–95%. Escalation rate is the number to watch: high escalation usually means the automation scope is too broad, not that automation cannot help.
Which metric should we improve first?
First contact resolution, almost always. It requires no automation investment and lifting it from 55% to 70% removes an estimated 25–30% of repeat contacts. Automation applied on top of a low-FCR operation just automates rework.
