Support is where most businesses feel AI’s value first — and where most of them get it wrong first. The 2024-era chatbot that answered FAQs and then said “let me connect you with an agent” for everything else is gone. In 2026, AI support can actually resolve tickets: check the order, apply the policy, issue the refund, and close the conversation.
Here’s how to decide what to automate, what to keep human, and what it realistically costs.
Chatbot vs Support Agent: The Difference That Matters
A support chatbot reads your help centre and answers questions. Useful, but it deflects rather than solves.
An AI support agent is connected to your systems — store, CRM, billing, shipping — and can take actions inside your rules. “Where’s my order?” gets the live tracking status. “I want to return this” gets a return label, if the order qualifies.
That difference is what moves the numbers. Deflection saves a little time; resolution removes the ticket entirely. The broader picture of what agents can and can’t do is in the AI agents guide.
What to Automate vs Keep Human
| Ticket type | Automate? | Why |
|---|---|---|
| Order status / tracking | Yes, first | High volume, zero judgement, live data answers it |
| Password and account access | Yes | Rule-based, secure flows already exist |
| Returns and refunds within policy | Yes, with limits | Clear rules; set a value cap for automatic approval |
| Product and how-to questions | Yes | Answerable from your docs and catalogue |
| Billing disputes | Triage only | AI gathers details, human decides |
| Complaints and angry customers | No | Needs empathy and discretion |
| Anything legal, medical, or safety-related | No | Risk outweighs the time saved |
The rule of thumb: if you could write the correct answer as a checklist, AI can handle it. If the right answer depends on reading the situation, keep a person in charge.
Built-In AI or Custom?
Use your helpdesk’s built-in AI (Zendesk, Intercom, Gorgias, Freshdesk) if your answers mostly live in help articles and the tool already integrates with your store. It’s quick to switch on, and pricing is usually per resolution.
Go custom when:
- Your key data lives in a custom system the helpdesk can’t reach
- Per-resolution pricing gets expensive at your volume
- You need specific business rules — refund limits by customer tier, approval steps, regional policies
- You want the same AI to work across chat, email, and phone
What It Costs
| Setup | Build cost | Ongoing |
|---|---|---|
| Helpdesk built-in AI | Setup time only | ~$0.50–$1.50 per resolution |
| Custom FAQ/RAG support bot | $1,500–$5,000 | $20–$200/month |
| Custom support agent (reads + acts in 1–2 systems) | $6,000–$20,000 | $100–$500/month |
| Multi-channel agent (chat, email, voice) | $20,000+ | Volume-dependent |
At a few thousand tickets a month, a custom agent often pays for itself within a year compared to per-resolution pricing — but only if it’s resolving tickets, not just answering them.
How to Roll It Out Without Annoying Customers
- Pull a month of tickets and group them. The top two categories are usually 30–50% of volume.
- Automate one category. Build it, test it on past tickets, then turn it on for a small share of traffic.
- Make the human handoff instant. “Talk to a person” should always work, and the person should see the full conversation.
- Be honest that it’s AI. Customers accept AI that solves problems. They resent AI pretending to be human.
- Track resolution rate and CSAT weekly. If satisfaction drops, fix it before expanding.
If phone is a big channel for you, the same approach applies to AI voice agents.
Drowning in repetitive support tickets? Let’s talk. I’ll look at your ticket mix and tell you what can be automated safely, what it would cost, and how much time it would realistically save.