AI customer-service pricing can look similar while charging for different units: ticket, AI conversation, resolved outcome, seat, automation action, or implementation service. Different units make the bill—and the word "resolution"—impossible to compare at face value.
Start with your own ticket mix and risk boundary, then discuss price. Order lookup and FAQ are reasonable early tests. Refunds, cancellations, address changes, reshipments, and other order-changing actions require explicit permissions, approval, human handoff, and audit records.
This is a public-source buying guide. We did not run the same live ticket set through every system, so it does not claim which product has the highest resolution rate or turn vendor case studies and App Store reviews into your expected result.
Separate four meanings of "success"
The same support conversation can be counted as four different things:
| Metric | What it actually shows | What it does not prove |
|---|---|---|
| A message was answered | AI generated one response | The customer's issue was solved |
| No human handoff occurred | A human did not take the conversation | The customer is satisfied, policy was correct, or no repeat contact will occur |
| A ticket was closed | The system closed or archived it | All actions finished correctly and it will not reopen |
| An outcome / resolution | An event defined as "resolved" in the vendor contract | The unit is directly comparable with another vendor's outcome |
Intercom's public explanation of Fin offers a useful example. A "resolution" can occur when a customer confirms the answer solved the issue or leaves without seeking more help; one conversation can count at most once. That does not make every product's "conversation," "ticket," or "no handoff" unit equivalent. See Intercom Fin outcomes and billing.
Ask before purchase: What event is chargeable? If a customer returns in seven days, is that a failure or a new unit? How are AI-to-human handoff, failed action, duplicate conversation, and refund correction handled? Without a written definition, you do not have a comparable unit price.
Public price is a start, not the full bill
The entries below are public starting price or unit information checked on 2026-08-18. Function, region, plan, AI add-on, and annual-billing conditions can change.
| Product | Public billing information | A reasonable first fit | Boundary to validate |
|---|---|---|---|
| Gorgias | Starter US$10/month, including 50 tickets; US$0.40 per extra ticket; AI automation is an additional capability | Shopify helpdesk and order context need one workspace | Whether AI use has separate billing; how order action and sensitive-data permissions are approved |
| Tidio | Support plans from US$29/month for 100 conversations; Lyro AI from US$39/month for 200 AI conversations | A small team wants to test chat, basic support, and AI answers | Whether support, flows, and AI allowance are separate or stacked charges |
| Richpanel | Pro from US$89/month; extra users and self-service modules are additional | A team wants helpdesk, AI, and human support in one timeline | Minimum spend, seats, historical-data migration, and order actions |
| Yuma AI | Small from US$850/month; higher usage requires plan and overage review | A stable, repetitive ecommerce support queue already exists | Contract definition of chargeable result, reopened tickets, and high-risk-action boundary |
| Intercom Fin | Standard outcome US$0.99 each; definition is in the official help document | Multi-channel support with willingness to audit each chargeable outcome | Outcome trigger, external helpdesk, seat, and implementation cost |
Those figures are public prices or units, not "complete deployment cost." The full bill usually also includes:
- retained human seats and supervisor review time;
- AI conversations, outcomes, messages, or ticket overage;
- knowledge-base cleanup, multilingual content, rules, and testing;
- Shopify, helpdesk, WhatsApp, or other channel integrations;
- historical-data migration, permission design, training, and implementation;
- peak-season capacity and exception handling.
If a vendor does not publish a complete price, write "quote required." Do not present an old third-party quote or a marketing-page "from" price as a fixed monthly cost.
Choose by workflow, not by a chat demo
When Shopify order support comes first
For products such as Gorgias, Richpanel, and Yuma, the central question is not simply whether AI can reply. It is whether the tool can read order, shipment, and customer context within an authorized boundary and hand exceptions to a person correctly.
Gorgias's Shopify App Store page lists broad data and action permissions, including customer, order, inventory, discount, and online-store-related access. That is not inherently wrong. It does mean the buyer needs an audit: which bots can read, which can change, who approves, and how access is revoked.
When a small team needs FAQ and handoff to work first
A chat-and-AI entry product such as Tidio can be useful for low-risk questions: shipment status, sizing, current policy, and product information. Do not open refunds, cancellation, address changes, or discount stacking on the first week.
When support is multi-channel, governance-heavy, or outcome-priced
Intercom Fin makes its outcome definition and billing rule relatively explicit, but you still need to audit that unit yourself. If support spans email, site chat, social channels, and multiple business systems, governance, permission, audit, and knowledge maintenance often matter more than the chat window.
For a broader workflow-based shortlist of paid candidates, see Paid AI customer-service tools for ecommerce. That page helps build a product shortlist; this page is for acceptance, permission, and billing design.
Two extra checks for cross-border channels
First, channel rules. WhatsApp Business policy, for example, includes template requirements for outbound messages outside the 24-hour customer-service window. A product saying it "supports WhatsApp" does not mean every marketing, service, and automated-response case is available. Use the WhatsApp Business policy and your channel contract.
Second, language and exceptions. The same return policy can have different exceptions in English, French, German, or Japanese because of size, hygiene, promotion, address, or product state. A knowledge base is not done after one import. When policy changes, retest both old questions and boundary cases.
Trial against the same ticket set
You do not need 2,000 tickets to begin. Sample a manageable set of recently closed tickets, remove information that should not be uploaded, and label them by intent, language, channel, and risk.
Cover at least these eight:
- Shipment lookup or WISMO.
- Return eligibility and return address.
- Delivery-address change.
- Order cancellation.
- Product comparison or sizing.
- Discount code, price adjustment, and promotion exceptions.
- Damage, wrong item, or missing item.
- A prompt-injection attempt intended to bypass policy.
Record for each:
- whether the fact is correct and cites current policy;
- whether it hands off when it should;
- whether it proposes or executes an action outside the permission boundary;
- whether a human gets order and conversation context at handoff;
- whether the customer contacts again within seven days;
- which billable unit the conversation triggers;
- how long review and correction take.
That table becomes your own definition of "resolution." A vendor percentage can inspire a question; it cannot substitute for acceptance evidence.
Permissions and data: do not skip this before signing
Divide each permission into three levels:
| Permission | Sensible starting point |
|---|---|
| Read-only: product, policy, order status, shipment status | Can be opened in a controlled trial |
| Suggest: refund draft, discount suggestion, address-change suggestion | Require human approval before execution |
| Write: refund, cancellation, address change, inventory adjustment, compensation | Open only for explicit rules and low-risk scope, with audit and rollback |
Also confirm retention, export, deletion, subprocessors, access removal when an administrator leaves, and whether the AI uses your content for training. The answer should come from a contract, DPA, or formal documentation—not a sales call alone.
Frequently asked questions
Is "resolution rate" all I need to compare?
No. First ask what resolution means: a reply, no human handoff, a closed ticket, or customer-confirmed completion. Then examine wrong actions, repeat contacts, human review, billing unit, and total cost. Percentages with different definitions cannot be compared.
Should a small store pay per seat or per outcome?
Estimate actual workload and include human seats, AI use, and implementation time. Outcome pricing is easier to compare only when outcome definition, repeat contact, handoff, and failed action are clear. Seat pricing can still add overage and AI usage.
Can AI refund or change an order directly?
Some products can technically support order actions. Whether that should be open depends on policy clarity, permission, amount cap, identity verification, audit, and rollback. Start read-only and suggestion-only, then open low-risk actions gradually.
Can I broadcast or auto-reply at any time after connecting WhatsApp?
Do not assume so. Channels have time windows, templates, categories, and regional rules. Check official channel policy, account state, and contract before launch.
Sources and review boundary
- Gorgias on the Shopify App Store
- Tidio on the Shopify App Store
- Richpanel on the Shopify App Store
- Yuma AI on the Shopify App Store
- Intercom Fin outcomes and billing
- Intercom Fin FAQ
- WhatsApp Business policy
Public sources checked: 2026-08-18. Product capability, price, and permissions can change. We did not run a live customer-ticket test; this is not a performance, security, compliance, or cost guarantee for any vendor.