You need a paid AI add-on only if the time it saves is worth more than it costs, and that depends on three numbers you already know or can measure: how many tickets you handle, how long they take, and what an hour of agent time costs you. Where agent assist is included in your plan, the question is only whether to use it. Where it is sold per agent, or where the AI agent is billed per resolution, a short calculation tells you the break-even point.
The short version
- Check what is already included. Some plans bundle agent assist; paying for a second one rarely makes sense.
- Per-agent add-on: it pays off if each agent saves more hours a month than the add-on price divided by their hourly cost.
- Per-resolution AI agent: it pays off if a resolution costs less than a human-handled ticket and the questions are ones customers are happy to have answered by AI.
- Use your own numbers. Vendor productivity claims are not a substitute for a short pilot.
- Volume matters most. Below a few hundred tickets a month, the included AI is usually enough.
Step 1: know what your plan already includes
Before any calculation, check the plan. Help Scout includes AI Drafts from Plus, and Zoho Desk includes Zia from Professional. Zendesk gives Suite Professional agents 5 Copilot uses a month without the add-on, and Intercom gives each agent 10 free Copilot conversations a month. A free allowance like that is the cheapest pilot you can run. Each product's pricing page lists what every plan includes.
Step 2: the break-even for a per-agent add-on
Let A be the add-on price per agent per month and W your fully loaded cost of one agent hour (salary, taxes, benefits and overheads divided by working hours). The add-on pays for itself when each agent saves more than A ÷ W hours a month.
With Zendesk Copilot, A is $50; with Intercom Copilot, $29; with Freddy AI Copilot on Freshdesk, $29. Divide by your own W to get the hours. Then ask whether a typical agent, handling around 250 tickets a month in our standard scenario, would save that much time from drafted replies and summaries. That is a question a two-week pilot can answer and a vendor case study cannot.
Step 3: the break-even for a per-resolution AI agent
Let P be the price of one AI resolution and M the average minutes an agent spends on a ticket of the kind the AI would answer. A human-handled ticket costs about W × M ÷ 60. The AI agent is cheaper per ticket when P is less than W × M ÷ 60.
P is $0.75 at Help Scout and $0.99 per outcome at Intercom. For Freshdesk, P is $49 divided by 100 sessions, divided again by the share of sessions that actually resolve, because unresolved sessions are charged too. Two corrections make the comparison honest: count only the questions the AI can answer well, which are usually the short ones with low M; and remember that savings arrive only if you then hire fewer agents or move them to other work.
A decision table by team size and volume
| Your situation | Agent assist | Customer-facing AI agent |
|---|---|---|
| 1–3 agents, under ~750 tickets a month | Use what the plan includes | Try it only if your help center already answers most questions |
| around 10 agents, ~2,500 tickets a month | Pilot the paid add-on with your break-even hours | Worth a pilot on repetitive topics, with a spending cap |
| 50+ agents, ~12,500 tickets a month | Likely worth it if hours saved beat A ÷ W | Likely worth it; negotiate the unit definition and a committed rate |
Whose numbers to believe
Vendors publish impressive figures, and we show them with a label saying who stands behind each. A named customer reporting its own result is stronger evidence than a vendor average. Freshworks, for instance, cites 80% on its product page, while named customers report 54% (iPostal1) and 30% (Hobbycraft) on measures that are not defined identically. Zendesk's named customer Fortnum & Mason reports 41%. Use the lower end when you estimate the AI's share of your volume, and plug that share into the true-cost calculator. Our how we review page explains the evidence labels.
If you are still deciding what the layers are, read what is an AI help desk first. For products ranked with AI costs included, see best AI help desk software and best help desk for small business.
Frequently asked questions
Is the AI add-on worth it for a small team?
Usually not as a paid extra. Small teams get most of the benefit from AI included in their plan. The per-agent add-on becomes easier to justify as ticket volume per agent rises.
How do I measure time saved in a pilot?
Compare average handle time and tickets closed per agent hour for the same ticket types, with and without the AI, over the same weeks. Exclude the first days while agents learn the tool.
Should I buy agent assist or the AI agent first?
Agent assist is lower risk, because a person reviews every reply. The AI agent can save more, but it needs a good help center and careful handoff rules.
Why do you not publish your own productivity numbers?
Because we have not measured them. We have not run hands-on tests, so we give you the formula and the list prices and leave the productivity figure to your own pilot.
Keep reading
- Help desk vs shared inboxguide
- What is an AI help desk?guide
- Per-seat vs per-resolution pricingguide
- Help desk migration checklistguide
- Best help desk softwarebest list
- Help desk true-cost calculator
This page was researched and drafted with AI assistance from the sources listed on it. We have not run hands-on tests of these products. Method: How we review