Published Aug 13, 2026 · Updated Nov 19, 2026 · 10 min read
AI Support Cost Per Resolution by Industry (2026)
How fully loaded cost per verified resolution changes by industry, built from BLS wage data for customer service representatives, published AI pricing and stated assumptions you can replace with your own.

AI support costs about the same per resolution in every industry: Intercom Fin bills $0.99 per outcome and Zendesk AI $1.50–$2.00 per automated resolution whatever you sell. What changes by industry is the fully loaded cost per verified resolution (CPVR), which adds the human cost of everything the AI does not resolve. On BLS wages and the assumptions below, that figure runs from about $3.30–$3.90 when the AI verifiably closes 75% of tickets to $16–$20 at 30%. Industry wage differences move it far less than the share the AI closes or the length of a human contact.
No public dataset reports AI resolution rates by industry, so this post does not invent one. It takes a sourced input (BLS wages by industry), states every other assumption, and shows the math so you can replace each input with your own.
Why does CPVR vary by industry at all?
Vendors price the same way across industries, as the pricing-model comparison shows. If the price per resolution is flat, the variation in fully loaded CPVR comes from four other drivers:
- Automatable share. The fraction of tickets the AI closes end to end with no identity check, judgement call or system write it is not trusted to make. Measure it on your own queue.
- Verification strictness. How long it takes to know a resolution held. A shipping question settles in hours; a billing dispute can reopen after the 72-hour window RTCI uses.
- Escalation floor. The share of tickets that must reach a human for regulatory, contractual or safety reasons. It caps automatable share however capable the AI is.
- Fully loaded human cost per contact. Wage plus benefits, tooling, QA and management, divided by contacts handled. Handle time matters as much as the wage.
Only the last is about money. The first three describe your ticket mix, and they are why two companies on the same contract can end up several times apart on CPVR.
What is the model?
Fully loaded CPVR is total support cost divided by verified resolutions:
CPVR = [F + (V × a × C_ai) + (V × (1 − a) × C_h)] ÷ (V × a)
- V = monthly conversations
- a = automatable share, the fraction the AI verifiably closes
- C_ai = AI fee per resolution
- C_h = fully loaded human cost per contact
- F = monthly fixed AI platform cost
Assumptions used throughout: V = 20,000, C_ai = $0.99 (Intercom Fin's published per-outcome rate), F = $3,000 a month. Replace them with your contract.
Human cost per contact is built from public data, then one assumption:
- Wage. The BLS Occupational Outlook Handbook lists May 2025 median hourly wages for customer service representatives in the industries that employ the most of them: retail trade $17.96, business support services $17.68, professional, scientific and technical services $21.77, healthcare and social assistance $21.49, and insurance carriers and related activities $22.47.
- Benefits. The BLS Employer Costs for Employee Compensation release for June 2026 puts wages and salaries at 70.0% of private-industry employer compensation costs. Dividing the wage by 0.70 adds benefits. It does not add tooling, QA or management, so the result is a floor.
- Contacts per hour (assumption). Four contacts per paid hour, or 15 minutes each including wrap-up and idle time. This is the input most worth replacing with your own figure.
Retail example: $17.96 ÷ 0.70 = $25.66 an hour, ÷ 4 contacts = $6.41 per contact.
What does CPVR look like by industry?
The industries are the ones BLS lists; it does not break out ecommerce, SaaS or fintech, so read retail trade for ecommerce, insurance for regulated financial services, and professional and technical services for B2B services. The automatable-share columns are scenarios, not industry measurements.
| BLS industry | Median wage (May 2025) | Loaded hourly cost (÷ 0.70) | C_h at 4 contacts/hr | CPVR at a = 30% | 45% | 60% | 75% |
|---|---|---|---|---|---|---|---|
| Retail trade | $17.96 | $25.66 | $6.41 | $16.46 | $9.16 | $5.52 | $3.33 |
| Business support services | $17.68 | $25.26 | $6.31 | $16.22 | $9.04 | $5.45 | $3.29 |
| Healthcare and social assistance | $21.49 | $30.70 | $7.67 | $19.40 | $10.70 | $6.36 | $3.75 |
| Professional, scientific and technical services | $21.77 | $31.10 | $7.78 | $19.63 | $10.83 | $6.42 | $3.78 |
| Insurance carriers and related activities | $22.47 | $32.10 | $8.03 | $20.21 | $11.13 | $6.59 | $3.87 |
AI-only CPVR, the vendor fee plus the platform fee spread over AI resolutions, is $0.99 + $3,000 ÷ (20,000 × a): $1.49 at 30%, $1.32 at 45%, $1.24 at 60% and $1.19 at 75%, in every industry.
Read across a row and CPVR falls by a factor of five or six between 30% and 75%. Read down a column and the highest-wage industry is 18–25% above the lowest. The wage gap between industries is real but small next to what the AI closes.
How much does handle time change it?
Contacts per hour is the assumption with the largest effect after automatable share. Regulated or technical contacts that need identity checks or investigation take longer, which divides the same wage over fewer contacts. Holding the retail wage and changing only contacts per hour:
| Contacts per hour | C_h (retail wage) | CPVR at a = 30% | 45% | 60% | 75% |
|---|---|---|---|---|---|
| 6 (10 min) | $4.28 | $11.47 | $6.55 | $4.09 | $2.62 |
| 4 (15 min) | $6.41 | $16.46 | $9.16 | $5.52 | $3.33 |
| 3 (20 min) | $8.55 | $21.45 | $11.78 | $6.94 | $4.04 |
| 2 (30 min) | $12.83 | $31.42 | $17.00 | $9.79 | $5.47 |
Halving contacts per hour doubles C_h. That is a far larger move than the 27% gap between the lowest and highest BLS wages above ($17.68 to $22.47). If your industry's contacts are long, that, not the wage, is what makes its CPVR high.
Worked comparison: same contract, two operations
Both companies sign the same contract: $0.99 per outcome, $3,000 a month platform fee, 20,000 conversations a month. The automatable shares and contact rates are assumptions chosen to illustrate the drivers.
The ecommerce brand pays the retail wage, handles 4 contacts an hour (C_h = $6.41) and verifiably closes 70%. Numerator: $3,000 + (20,000 × 0.70 × $0.99) + (20,000 × 0.30 × $6.41) = $3,000 + $13,860 + $38,460 = $55,320. Denominator: 14,000. CPVR: $3.95.
The insurer pays the insurance wage, handles 2 contacts an hour because each needs identity checks (C_h = $32.10 ÷ 2 = $16.05), and verifiably closes 40% because claims decisions must reach a human. Numerator: $3,000 + (20,000 × 0.40 × $0.99) + (20,000 × 0.60 × $16.05) = $3,000 + $7,920 + $192,600 = $203,520. Denominator: 8,000. CPVR: $25.44.
Same vendor, same price, same volume, and a 6.4x gap. Benchmarked against the ecommerce number, the insurer would conclude its AI failed. The useful comparison is with itself: raising its verified share from 40% to 50% cuts monthly cost from $203,520 to $173,400 and CPVR to $17.34. Negotiating the fee down 30 cents to $0.69 at 40% saves $2,400 a month and takes CPVR only to $25.14. For the timing of payback, see Break-Even: When AI Support Pays for Itself.
How do I place my own operation?
Skip the industry proxy and measure the four inputs directly:
- Automatable share. Take a rolling 14-day sample using the measurement method. Count only conversations the AI closed with no human message and no same-issue re-contact within 72 hours, and divide by all conversations.
- Fully loaded human cost. Total support payroll plus tooling, QA and a share of management, divided by human-handled contacts. Not by all contacts, and not by headcount.
- Escalation floor. List the categories that must legally or contractually reach a human and total their share of volume. Automatable share cannot exceed 100% minus that floor.
- Reopen timing. If many reopens land after day 3, a 72-hour window flatters you and real CPVR is higher than measured.
If your automatable share is lower than you expected, check knowledge coverage and system permissions before the model; benchmarking against peers helps separate the two. If it is higher, confirm you are not counting silence as success.
What moves CPVR within an industry?
- Knowledge coverage on the top ticket types. The AI cannot resolve what your knowledge base does not document. Run a knowledge base readiness audit before blaming the model, and use the stalled-at-45% diagnostic to find which cause applies.
- Write permissions. An AI that can read an order but not issue a refund hands off every refund. Scoped write access on low-risk actions raises automatable share directly.
- Routing before the AI. Sending regulated categories straight to humans stops the AI using a billable attempt on tickets it may not close.
- Reopen analysis. Each reopen is a resolution you paid for and then paid a human to redo. Cutting reopens raises verified resolutions at the same volume and price.
A small per-resolution discount moves CPVR far less, as the insurer example shows: $0.30 off the fee is worth $2,400 a month, ten points of verified share is worth $30,120.
Which mistakes corrupt industry comparisons?
- Comparing AI-only CPVR with fully loaded CPVR. In the table above the two differ by a factor of 2.8 to 13.6. Compare like with like.
- Using deflection rate as automatable share. Deflection counts conversations that ended; automatable share counts issues solved.
- Mixing verification windows. A vendor resolution rate measured over 24 hours is not comparable to your 72-hour figure. Normalise the quotes first, and check the contract for fees that never reach the pricing page.
- Blending segments. Consumer and enterprise support are two ticket mixes and need two CPVR numbers.
- Treating another company's number as a target. Your escalation floor and handle time may put its CPVR out of reach. Track movement from your own baseline.
Frequently asked questions
Where do these numbers come from? Wages are BLS May 2025 medians for customer service representatives by industry, from the BLS Occupational Outlook Handbook. Benefits use the 70.0% wage share from the BLS June 2026 compensation release. Contacts per hour, volume, platform fee and automatable shares are stated assumptions. Change any of them and rerun the formula.
Why is fully loaded CPVR so much higher than the vendor price? The vendor charges only for what the AI resolves, while you still pay humans for everything else. At a 50% verified share, each AI resolution carries one human-handled contact alongside it.
Which industry has the lowest cost per resolution? Whichever combines short contacts, a high verified share and a low escalation floor. Among the BLS industries above, retail trade and business support services have the lowest median wages; ecommerce order-status tickets are also short and well documented, which helps both other drivers.
Does a higher automatable share always lower CPVR? Yes, if the resolutions are genuinely verified. Raising it by loosening the verification standard lowers CPVR on paper while re-contacts rise.
How often should I recalculate? Monthly. A Google Sheets tracker keeps the inputs in one place. Ticket mix shifts with releases, pricing changes and seasons, and automatable share moves with it.
What if my CPVR is far from the table? If it is much lower, check that silences are not counted as resolutions and that your window matches reopen timing. If it is much higher, check handle time and the escalation floor before the vendor.
Bottom line
Industry sets your wage level and, through regulation and ticket complexity, your escalation floor and handle time. The wage is the smallest of those effects. Measure your automatable share and human cost per contact, run the formula, and work on knowledge coverage and scoped write permissions, the two levers that raise verified share.
Change log
- Replaced the unsourced per-industry automatable shares and human costs with BLS May 2025 wages for customer service representatives and stated assumptions, recomputed every table, and fixed a direct answer that contradicted the table.
- Published.
Read next
Keep reading
Picked for this article: the pieces it cites and the ones that answer the next question buyers usually ask.

Most related
Track Cost Per Resolution in Google Sheets
A copy-pasteable Google Sheets build for cost per verified resolution: a ticket log, a monthly cost sheet, a live dashboard, a budget-variance template and a worked example that shows why the vendor rate understates what you pay.

How to Benchmark Your AI Resolution Rate Against Industry Peers
How to compare your AI resolution rate with peers: fix the definition and denominator, then benchmark against RTCI's 58% median realistic rate and Comm100's 44.8% cross-customer figure.

AI Customer Support Pricing Models Compared: Per-Seat, Per-Resolution, and Usage
AI support vendors price by seat, by resolution or by usage. A worked example at 10,000 conversations a month shows the same 5,000 resolutions costing $874 to $7,200 depending on the price card.

Why Resolution Stalls Around 45%: Diagnostics and Fixes
When verified AI resolution flattens in the 40s, the cause is usually one of five operational gaps. A five-question test finds yours, and each cause has its own fix.

Resolution Rate vs. Deflection Rate: What's the Difference?
Resolution rate measures issues solved end to end; deflection rate measures contacts avoided. Learn the formulas, see examples, and stop vendors from blending them.