SMB AI agent pricing for Shopify: rent, build, or break even.

Three honest brackets, with a vendor name attached to each. Under 300 resolved tickets a month: $0 to $39 a month on Tidio Lyro, Shopify Inbox, or Re:amaze. The math is straightforward, you do not need a custom build, and a consultant is not the right spend. Between 300 and 2,000 resolved tickets a month: $0.90 to $0.99 per outcome on Fin or Gorgias, which lands somewhere between $300 and $2,000 a month. This is the band where most 5 to 15 person Shopify shops live in 2026. Above 2,000 resolved tickets a month, or with a tight rule set you cannot fit into a packaged vendor: $500 to $10,000 once for a custom build on top of the Anthropic API plus an optional $1,000 to $5,000 a month retainer. The custom path runs roughly $30 to $80 a month in token cost on Opus 4.7 or Sonnet 4.6 with prompt caching and batch enabled, so the variable cost is the engineering retainer, not the model.

A typical Shopify ticket reply needs about 6,000 input tokens of context (system prompt, the order record, the customer message, the product catalog snippet, the policy excerpts) and produces about 500 output tokens. Opus 4.7 lists at $5 per million input tokens and $25 per million output. Naive cost per ticket: 6,000 / 1M * $5 = $0.030 input, plus 500 / 1M * $25 = $0.0125 output, total ~$0.043. Apply prompt caching (90 percent off cached reads) on the system prompt, policies, and catalog (which dominate the input) and the input cost drops to around $0.005, total ~$0.018. Run non-urgent replies through batch (50 percent off) and you can take that down further. The $0.018 to $0.043 range is the real number to compare against $0.99 per Fin outcome or $0.90 to $1.00 per Gorgias AI ticket. Sonnet 4.6 is roughly a third of the price for tickets that do not need frontier reasoning, which is most of them.

Rough cut, with the assumptions in the previous answer: a custom build at $5,000 once plus a $1,000 a month retainer plus $50 a month in token cost is $11,600 in year one if I deliver on day zero. Fin at $0.99 per outcome on 1,000 monthly tickets is $11,880 in year one. So they cross around the 1,000 ticket per month mark in year one, and from year two onward the custom build is dramatically cheaper because the build cost is amortized. Below 1,000 tickets per month, rent. Above 2,000 tickets per month, build is obvious. Between 1,000 and 2,000, it depends on whether your rule set fits the packaged vendor. If your store has unusual SKUs, complex policy logic, or integrations the vendors do not cover, build wins earlier. If you are a generic Shopify DTC store with a clean policy, rent until you outgrow it.

Real and load-bearing for support agents. The shape of a Shopify ticket reply makes prompt caching close to ideal: a long, slow-moving prefix (system prompt, brand voice rules, FAQ corpus, policy excerpts, catalog snippets) followed by a small, fast-moving suffix (the customer's actual message and the order context). Anthropic charges 90 percent less for tokens read from cache, and the cache stays warm across requests within a session window. On a real workload, the cached prefix is often 4,000 to 5,000 of the 6,000 input tokens, which is exactly where the savings concentrate. The catch: you have to design your prompts so the cacheable content sits in a stable position at the front, and you have to actually flag the cache breakpoints in the API call. If you are using a packaged vendor's API wrapper or a no-code builder, you usually do not get to control this, and the savings go to the vendor's margin, not your bill.

It depends on your resolution rate and your edge cases. Fin only charges when the agent resolves the conversation without a human takeover, so a 60 percent resolution rate means you pay for 60 out of 100 tickets. Gorgias charges per AI-resolved ticket too, but it sits on top of a $60 to $750+ helpdesk plan that you pay regardless of volume. Two scenarios. A 1,000 ticket month at 65 percent resolution: Fin is 650 * $0.99 = $643.50. Gorgias on the $300 plan is $300 + 650 * $0.95 = $917.50. Fin wins on pure variable cost. A 200 ticket month at 65 percent: Fin is 130 * $0.99 = $128.70. Gorgias is $300 + 130 * $0.95 = $423.50. Fin still wins, but you are also paying Gorgias for the helpdesk itself, which has to be valued separately. The fairer comparison is: what would you pay another helpdesk + an AI add-on to replicate Gorgias' core inbox? If the answer is $200 a month, then Gorgias' AI premium over baseline is $217.50, which is comparable to Fin. The headline numbers hide the bundling.

Tidio Lyro and Shopify Inbox: no developer. The Shopify app stores them, you click install, you paste your FAQs into the admin, and they run. Fin and Gorgias: a developer or a sufficiently technical operator helps with the integration but is not strictly required for the standard Shopify connection. The work is mostly content (training the agent on your policies, products, common ticket shapes) and routing rules. Zendesk AI: yes, in practice, especially if you are migrating from another helpdesk. A custom build on Opus 4.7 or Sonnet 4.6 directly: yes, this is the developer-led path. The whole point is to write the integration once, exactly to your rules, with full control over the model, the prompt, and the audit log. The c0nsl tier sheet exists for this case: $500 to $2,000 for a single small integration, $2,000 to $10,000+ for a full custom system, $1,000 to $5,000 a month for ongoing ownership.

Sidekick is the merchant-facing assistant inside the Shopify admin. It helps you, not your customers, so it is orthogonal to the customer support agent question. Shopify Magic generates product descriptions, theme content, and email subject lines, again merchant-facing. Shopify Inbox is the only customer-facing piece, and it is free, but the AI features are limited to FAQ-style replies and quick answers. None of the three are a full ticket-deflection agent. The right way to think about it: Shopify's bundled AI is a useful first step that handles maybe 20 to 30 percent of the easiest tickets at zero marginal cost. Once you outgrow it, you pick a packaged vendor (Fin, Gorgias, Tidio) or build. Treat Shopify's free tools as part of the stack, not as a replacement for the deflection layer.

Buying enterprise tier when SMB tier would have worked. Zendesk AI and Ada are real platforms but their pricing assumes a contact center with 50+ seats, a procurement cycle, and an annual contract. A 5 to 15 person Shopify store does not have any of those things. Operators end up paying a $5,000 a month bundle for capabilities they will never use, then blame AI when the ROI is not there. The second most common mistake is paying for a six-month strategy engagement with an AI agency before any code ships. The third is buying a no-code automation tool for $99 a month, never giving anybody on the team time to own it, and abandoning it after two months without measuring whether it deflected anything. The honest order is: free or near-free tier first, measure for a month, upgrade when the volume signal demands it, custom build only when the rule set genuinely cannot fit a packaged vendor.

Every packaged AI agent that connects to Shopify asks for a set of access scopes during install. Most ask for read-write on orders, customers, and discounts at minimum, because the agent might issue a refund, apply a discount code, or update an order note. After the April 2026 Vercel and Context.ai breach, the right question to ask before you click through the consent screen is: does this agent actually need write access to discounts, or could the workflow live with read-only? Most agents only need read access on most of their scopes and a tightly scoped write on order notes. If a vendor will not let you reduce a scope, that is a real signal. You can also rotate the API token quarterly and revoke any unused integration in the Shopify admin under Apps. SVC-007 on the c0nsl service catalog is a fixed-fee version of this audit applied to every AI tool in your stack, not just the ticket agent.