What a Support Ticket Actually Costs (Human vs AI Math)

Most Shopify founders can tell you their customer acquisition cost to the penny. Ask them what a single support ticket costs and you get a shrug, or a guess based on the helpdesk invoice.

That's the wrong number. The subscription is the smallest line in the equation. The real cost of a ticket is buried in salary, payroll overhead, management time, tooling, and the training you redo every time someone leaves.

Once you calculate it properly, two things usually happen. You realise support is more expensive than you thought, and you stop debating automation on vibes and start debating it on unit economics.

The cost-per-ticket formula

Here is the full version. Everything on the cost side goes on top, every ticket you actually handled goes on the bottom, and both must cover the same period.

Cost per ticket = (Fully loaded people cost + Tooling + Management overhead + Training & recruitment) ÷ Tickets handled

The word doing the work is fully loaded. Base salary is roughly 70 to 80% of what an employee actually costs you once you add employer taxes, benefits, equipment, software seats and paid leave.

A worked illustration: three support models compared

This is an illustration, not research data. Every figure below is an assumption we've chosen to make the maths transparent. Swap in your own numbers: the structure is what matters, not our inputs.

The shared assumptions

• The store handles 3,000 support conversations a month
• A support agent handles roughly 1,500 conversations a month at full utilisation
• Agent base salary: $45,000, loaded at 1.25× = $56,250 a year
• Support manager loaded cost: $90,000 a year
• Helpdesk seat: $60 per seat per month
• Blended internal hourly rate for oversight work: $35
• AI cost reference point: 5.7 cents per conversation (AskZoye's usage-based rate)
• Share of conversations resolved without a human: an assumption you must measure for your own store, not a claim about any specific vendor

Model 1: Human-only

Two agents, no automation.

Annualised, that's $135,936, and it buys you coverage during working hours only. Nights, weekends and holidays are uncovered unless you pay a premium for them.

Model 2: Hybrid (AI front line, humans on escalations)

Assume 60% of conversations are resolved without a human, so 1,200 reach an agent. That's one agent, not two.

Model 3: AI-first (AI front line, part-time human layer)

Assume 75% resolved without a human, leaving 750 escalations. Comfortably a part-time role.

Side by side

The number AI does not get to zero

Look again at Model 3. The AI reference rate is 5.7 cents per conversation, but the AI-first cost per ticket is $1.21. More than twenty times that.

That gap is the honest part of the story, and it's made of four things.

1. Escalations. A share of conversations always needs a person. Refund disputes, damaged goods, angry customers, anything requiring a policy exception. Those tickets carry the full human cost.
2. Oversight. Someone has to read transcripts, spot wrong answers, and keep the knowledge base current as products, shipping times and policies change. Budget a few hours a week and don't pretend it's free.
3. Setup. Even with fast onboarding, you still need to document your policies properly, decide escalation rules, and test with real questions before you trust it.
4. Fixed overhead. Tooling and management time don't disappear because volume shifted to AI.

Anyone quoting you a per-ticket cost that equals their per-query rate is quoting you the ingredient, not the meal.

"AI doesn't remove the cost of support. It removes the cost of repetitive support, which is usually the majority of it."

Where the cost actually goes when volume shifts

There's a second honest caveat. Deflecting tickets does not automatically produce a smaller bill.

If you keep the same headcount and the same tooling, your cost per ticket falls on paper while your monthly spend stays identical. The saving only becomes real when one of three things happens:

• You don't make the next hire you were about to make
• You redeploy freed hours to something that generates revenue, proactive outreach, VIP customers, product feedback loops
• You absorb growth without support cost scaling in line with orders

That third one is where most Shopify stores get the value. Volume doubles, support cost barely moves. Track it as cost per ticket alongside your other support KPIs so you can see the curve flatten.

Calculate your own number this week

1. Pull 90 days of ticket data. Count conversations resolved, not received. Divide by three for a monthly figure.
2. Add up loaded people cost. Base salaries × 1.25. Include any fraction of a founder's or ops manager's week spent on support.
3. Add tooling. Helpdesk, chat widget, survey tool, translation, phone. Support-only spend.
4. Add training and recruitment, annualised then divided by 12.
5. Divide. That's your baseline cost per ticket.
6. Segment it. Split by ticket type. WISMO, returns, sizing and pre-sale questions almost always cost the same to handle as a complex complaint, which is exactly why they're the ones worth automating first. Our guide to reducing WISMO tickets covers the biggest category.
7. Re-run it monthly. A single snapshot is a number. A trend is a decision.

The bottom line

Cost per ticket is the cleanest number in customer support, and the one most stores never calculate. Work it out properly (loaded salaries, tooling, management time, training) and the automation decision usually stops being a debate.

Just keep the maths honest in both directions. AI doesn't take support to zero. It takes the repetitive majority off your team's plate at a fraction of the human rate, leaves you paying full price for the escalations that genuinely need a person, and adds an oversight line you should budget for from day one.

If you run a Shopify store, [AskZoye](https://askzoye.com) handles the front line at roughly 5.7 cents per conversation, replies in about three seconds, and goes live in under 60 seconds with no code. Run your own numbers against it before you post that job ad.