The hidden fees in AI support pricing

The published price for AI customer support software is a fiction. Vendors advertise a per-resolution fee or a monthly seat cost, then slip eight hidden charges into the contract. You don't see them until month two when your bill is triple the quote.

This guide lists the line items that actually wreck budgets, how they stack, and what to ask before you sign anything.

The double-charge trap: Gorgias

Gorgias bills on two axes at once. You pay a base price per ticket (the "platform fee"), then you pay again per automation that runs, even if that automation is an AI agent resolving the ticket end-to-end. Same ticket, two invoices.

Example: You process 1,000 tickets a month. Gorgias charges you for all 1,000 as platform tickets. Then, if your AI agent auto-responds to 600 of them, you pay an additional per-automation fee on top for those 600. The AI didn't save you money—it cost you more.

Gorgias also restricts its AI Agent to Shopify sellers only, so if you run a SaaS or service business, you're locked into the basic bot tier. Always clarify line-item pricing before demo.

Session expiry and the no-rollover trap: Freshdesk

Freshdesk AI charges per session, not per message or resolution. A session is a continuous conversation. But here's the catch: Freshdesk auto-closes every session after 24 hours of inactivity, even if you purchased session credits and haven't used them yet. When the customer replies a day later, that's a brand-new session and a brand-new charge.

More punishing: Freshdesk AI recently hiked per-session pricing from roughly $0.10 to $0.49 per 100 sessions (effective April 2026). That's a 5× price increase on a platform that auto-wastes your session budget.

If you run a support operation where customers don't reply within 24 hours—typical for any async product or SaaS support—you'll pay for the same conversation twice.

Hidden minimums and seat lockups

Most support platforms hide annual minimums in the fine print. Kustomer, for instance, requires an 8-seat annual commitment if you want to use their bot tier. That's $960–$1,920/year before you answer a single ticket if your team is smaller.

Bland AI, the all-in voice agent platform, has a monthly minimum of $299–$499 per seat. That's $3,588–$5,988 per year just to pilot the technology.

HubSpot Breeze, the cheaper per-resolution option at $0.50/resolution, also charges a $1,500 one-time onboarding fee. If you resolve fewer than 3,000 tickets a month, you'll spend months just breaking even on the setup cost.

Per-failure and per-conversation billing: billing your losses

Some platforms charge you every time the bot tries and fails. Ada and several competitors bill per interaction, not per resolution. If a bot attempts to resolve a ticket and bounces it to a human, you've paid Ada for the failure. Better AI should lower your bill; instead, you're paying for the attempt.

Kustomer bots charge around $0.60 per conversation, regardless of outcome. Salesforce Service Cloud bills per 24-hour session; if your conversation spans 48 hours, that's two charges. You can't win.

Contrast that with genuine per-resolution pricing: HubSpot Breeze at $0.50 and Help Scout at $0.75. If the bot fails, you pay nothing. Better AI = lower bill, as it should be.

Voice platform traps: minimums, idle time, and LLM passthrough

Voice agents sound cheaper at first. Vapi quotes $0.05/minute, Retell $0.07, Synthflow $0.09. But the real cost is almost always higher.

Ring-time billing. Air AI bills the time the phone is ringing before the customer speaks. That's not conversation time; that's wasted caller time. A 30-second hold turns into billable minutes.

Idle websocket billing. Deepgram, a common STT provider used in voice stacks, bills for idle websocket connections. If your voice agent stays on a call for 5 minutes to let a customer think, Deepgram is charging you the whole time.

LLM passthrough. Most voice platforms don't include the LLM (language model) cost in their per-minute rate. Vapi at $0.05/minute becomes $0.20–$0.30/minute once you add Claude or GPT-4 overage. Bland is the exception: $0.11–$0.14/minute is truly all-in, model included.

Platform minimums. Bland requires $299–$499/month. Even if you only pilot with 100 calls, you've paid the full monthly floor.

How to spot hidden fees before signing

Ask these questions in every demo:

• Are there multiple charges per ticket? (Gorgias: yes.) Do platform fees, automation fees, and AI fees stack?
• How long is a session? (Freshdesk: 24 hours; Salesforce: 24 hours; some others: 30 days.) Do unused sessions roll over?
• Is there a monthly minimum or annual seat commitment? (Kustomer: 8 seats minimum; Bland: $299–$499; HubSpot: $1,500 onboarding.)
• What am I billed for if the bot fails? (Ada, Kustomer: yes; HubSpot, Help Scout: no.)
• What model am I using? (Crucial for voice. $0.05/min + GPT-4 passthrough ≠ $0.05/min all-in.)
• Are there setup, onboarding, or training fees? (HubSpot: $1,500.)
• Is ring time or idle time billed? (Ask explicitly for voice agents.)

Use a spreadsheet to model a realistic month: 1,000 conversations, X% to bot, Y% bot failures requiring escalation. Then call and ask: "If this happens, what's my bill?" Most vendors will hedge or refuse to answer. That's a red flag.

Try the AI support cost calculator to compare ticket-based platforms side-by-side. For voice agents, use the voice agent cost calculator to catch hidden per-minute costs. And review vendor pricing pages for one more pass before you sign.

Read the invoice, not the headline

AI support pricing is deliberately opaque. Gorgias charges twice per ticket. Freshdesk sessions expire without notice. Kustomer locks you in for eight seats. Bland requires $3,588–$5,988 annually. Voice platforms bury LLM costs, ring-time charges, and idle-time billing in the footnotes. The cheapest quote becomes the most expensive bill after month one.

Always ask about double charges, session expiry, annual minimums, per-failure billing, and model costs in writing. If a vendor won't commit in email, walk.