AI support pricing explained: replies, conversations and resolutions

An AI support unit is not an industry standard

Two plans with an allowance of 500 can cover very different workloads. One may count outbound messages, another a shopper conversation, and another a resolution or automated outcome. None of those labels alone proves that an order action was completed correctly.

Ask what starts the billable event, what ends it, and how follow-up messages, reopened requests and human handoffs are treated. Keep the provider’s definition attached to the pricing record rather than renaming every unit “tickets.”

Replies and conversations measure different things

Zipchat Starter includes 750 outbound AI replies monthly. A shopper can receive several replies within one conversation, so 750 should not be presented as 750 complete cases. Tidio’s paid Lyro plans use recurring conversation allowances; the free 50-conversation allowance is one-off.

There is no universal conversion factor between replies and conversations. Use your own interaction pattern or a controlled pilot. If that evidence is missing, show the two allowances alongside each other without claiming which processes more complete work.

Add the base product and human seats

Gorgias’s researched Starter example combines a helpdesk fee and AI fee, with distinct ticket and automated-interaction allowances. Starter excludes Actions, so an action-capable workflow may require a different plan.

eDesk combines a per-user plan with automated-outcome charges. Re:amaze Basic includes only five AI resolutions per user each month, with additional resolutions separately priced. A seat fee does not establish unlimited AI usage, and a low AI unit price does not establish the entire helpdesk bill.

Traffic tiers and annual bundles need context

Rep’s traffic-tier example at 80,000 monthly sessions shows separate Support, Sales and Concierge prices. It is not a verified starting-price offer. Use the relevant traffic tier and selected product rather than applying a per-conversation estimate to it.

Yuma’s illustrative annual bundle includes 12,000 full resolutions for $14,400 paid upfront. Its plans require a 12-month commitment, usage pauses at the bundle limit, and an existing helpdesk can add another cost. The final package is quoted; this example is not a universal monthly entry price.

Build a like-for-like scenario

Define your store platform, channels, human seats, required actions and expected workload. Then map that workload to each provider’s own meter. Include mandatory fees and add-ons, and distinguish a published fee from a quoted implementation cost.

A good comparison can conclude that the available public information is insufficient for a reliable total. That is more useful than filling missing rates with assumptions the merchant cannot see.

• Record the specific required plan and action coverage.
• Keep conversations, replies, tickets and outcomes separate.
• Add mandatory seats, platform fees and existing systems.
• Expose annual commitments and usage caps.
• Confirm how the vendor treats escalations and repeat contacts.

Validate the billable event in a pilot

During an evaluation, record both the support outcome and the event charged by the product. An answer may be billable even if a person later completes the request. A support-quality test and a billing test therefore answer different questions.

ToolClue’s launch records are research checked. They do not claim measured resolution rates, cost savings or the number of real cases any allowance will finish. Future tests should publish the setup, sample requests and charging observations together.