AI Chatbot for a Shopify Store: Build vs Buy, With a Real Cost Model

“Should I install a chatbot app or have one built?” is really a question about how you want to be billed. Apps meter you — per conversation, per AI reply, or per resolved ticket — so their cost climbs with your order volume. A custom build is mostly a fixed cost: a project fee, a small upkeep line, and usage charges so small they round away. Neither is “the cheap option.” At 500 orders a month the meter is cheaper; somewhere in the mid-thousands of orders a month, under the assumptions modelled below, the fixed cost wins. This guide prices all four routes — Shopify’s free Inbox agent included — at three store sizes, with the decision framework and training-data checklist that matter more than the last $100 either way. For chatbot pricing in general, see our platform-agnostic guide to AI chatbot costs; this article is specifically about Shopify stores choosing between renting and building.

The four routes, and how each one meters you

Route 1 — Shopify Inbox’s AI agent: $0. Shopify’s own chat app is free, and since June 2026 it includes an AI agent that answers from your catalogue, store policies, pages and Knowledge Base, can check an existing order, and can add products to a cart (Shopify Help Center, as documented in October 2026). Two limits define it: it needs Shopify’s new customer accounts, and it covers storefront chat only — not the email, Instagram and WhatsApp inboxes a growing store accumulates. Your Shopify plan itself (Basic is $39 a month on monthly billing, $29 billed yearly, per Shopify’s pricing page) is a cost every route shares, so it is excluded from the comparison below.

Route 2 — A chat app with an AI add-on (Tidio + Lyro). Tidio sells three meters separately. Its Customer Service plans start at $24.17 a month billed yearly (Starter, 100 “billable conversations” — a conversation counts when a human replies; AI and automation replies don’t). The AI agent, Lyro, is a second meter: billed per conversation it joins — defined on Tidio’s pricing page as any interaction with at least one AI reply — with quotas from $32.50 a month billed yearly for 50 conversations up to $583.33 for 1,000. Every account gets 50 Lyro conversations once, not monthly. Note the asymmetry: you pay for Lyro conversations whether or not the AI resolves them.

Route 3 — A helpdesk with a per-resolution AI agent (Gorgias). Gorgias prices the helpdesk by ticket volume (Basic: 300 tickets, $50 a month billed yearly) and the AI Agent per resolution: $0.90 per automated interaction on annual plans, $1.00 on monthly, per Gorgias’s own pricing FAQ. The catch is in the same FAQ: each AI resolution also counts as a helpdesk ticket, so one conversation consumes two allowances. Plans bundle a small block of AI interactions (30 on Basic with AI Agent, at $77 a month billed yearly); beyond the block you pay per interaction — third-party teardowns of the current plan grid put overage at up to $1.50 each, and the model below uses that higher figure. Intercom’s Fin follows the same per-outcome logic at $0.99 per resolution, on top of per-seat helpdesk pricing from $29 a seat billed yearly — fair per unit, linear at scale.

Route 4 — A custom build. A chatbot built for your store, on your data, connected to the systems you choose. Its cost shape is the mirror image of the apps’: a one-time build, a predictable upkeep line, and per-conversation usage measured in fractions of a cent. The next section covers what that involves on Shopify — because the platform, not the AI, sets the real constraints.

What a custom build on Shopify actually involves

A store chatbot build has five parts: a storefront chat widget (a theme app embed), a knowledge layer over your catalogue and policy content, an order-status lookup through Shopify’s Admin API, guardrails and a test set, and a human handoff into your inbox or helpdesk. That is the scope behind the hours in the model below — and the same shape as the process on AI Dev Expert’s AI chatbot service page: discovery, strategy, design, build, testing, launch and improve, with knowledge sources and handoff responsibilities agreed before anything is built. (Rather assemble it yourself? Our step-by-step guide to building an AI chatbot for your website covers the DIY route.)

Two Shopify rules shape every build. First, protected customer data: a customer’s name, email, phone and address are only returned to an app that Shopify has approved for each field — until then the API returns those fields redacted, so an order lookup that “should work” silently can’t see who it’s talking to (Shopify’s developer documentation). Second, the default order scope covers roughly the last 60 days of orders; full history needs a separate approved scope. Neither rule blocks a build, but both belong in the project plan, not in a bug report after launch.

The running-cost anatomy surprises most people. The AI usage itself is nearly free at store scale: at OpenAI’s published GPT-4o mini rates ($0.15 per million input tokens, $0.60 per million output), a typical support conversation of roughly 4,000 input and 600 output tokens costs about $0.001 — a tenth of a cent. Even 2,500 conversations a month is about $2.50 of AI. What you actually pay for monthly is upkeep: reviewing failed answers, updating content when policies change, and keeping the integration healthy as Shopify evolves its APIs. The model budgets three hours a month for that — the build’s real subscription.

The decision framework: five tests

1. Volume and trajectory. Below a few hundred support contacts a month, meters are trivially cheap and a build cannot pay back. What matters is where your volume will be in 18 months — per-resolution fees scale with your success; a build’s cost does not.

2. Question mix. If most contacts are order status, shipping times, returns windows and sizing — the routine block every route handles — buy on price. If your answers are advice-heavy, regulated, or specific to how your products are used, control of your own knowledge layer and guardrails matters more than the subscription saving.

3. Actions, not just answers. Order lookup is table stakes. Refund initiation, subscription changes, address updates: vendor agents offer a defined action menu; a build can connect to your returns app, subscription platform or internal tools — scoped action by action.

4. Channels and systems. A helpdesk-grade app earns its fee when chat, email and social land in one queue with order data attached. If storefront chat is your only channel, that fee buys a building you don’t live in.

5. Ownership. With an app, your conversation data, training content and roadmap live inside a vendor’s product and pricing decisions. With a build, the prompts, knowledge base and integrations are yours — as is the maintenance duty.

Test	Buy (app) leans	Build leans
Volume	Under ~2,000 orders/month, or unpredictable	Sustained volume where metered fees pass ~$600/month
Question mix	Mostly order status, shipping, returns, sizing	Advice-heavy, regulated, or brand-specific answers
Actions	Vendor’s action menu covers your needs	Custom actions in your own systems
Channels	Chat + email + social in one helpdesk	Storefront chat inside your own stack
Ownership	Renting is fine; switching cost accepted	Data, prompts and roadmap must stay yours

The Build-vs-Buy Cost Model

Stated assumptions (replace them with your own figures):

• Contacts: support contacts run at 0.25 per order (500 orders → 125 contacts a month). Measure yours: contacts ÷ orders, last 90 days.
• Deflection: the AI resolves 40% of contacts end-to-end; humans handle the rest. This is a planning assumption, not a vendor promise — a sensitivity case follows the tables.
• Human cost: $6 per human-handled contact, fully loaded. Published benchmarks put retail contacts around $2.70–$5.60 each, with US phone contacts averaging $7.16 (ContactBabel, 2025) — $6 is a mid-range planning figure.
• App prices: annual billing, as published November 2026 (Tidio pricing page; Gorgias pricing page and plan grid). Gorgias resolutions beyond each plan’s included block are costed at $1.50 each (see Route 3) — deliberately the conservative figure.
• Build: 120 developer hours at a $75/hour blended rate = $9,000, inside a stated range of 80–160 hours ($6,000–$12,000). This is a modelling assumption, not a quote — AI Dev Expert, like most builders, scopes builds in writing rather than publishing a package price. Amortised over 24 months ($375/month), plus 3 upkeep hours/month ($225), plus a $30/month hosting-and-AI allowance (actual AI usage at these volumes: under $3).
• Your Shopify plan and payment fees are excluded — identical on every route.

Volume math (per month)

	500 orders	2,000 orders	10,000 orders
Support contacts (0.25/order)	125	500	2,500
Resolved by AI (40%)	50	200	1,000
Handled by humans	75	300	1,500
Human handling cost ($6 each)	$450	$1,800	$9,000

Total monthly cost by route (tooling + remaining human time)

Route	500 orders	2,000 orders	10,000 orders
No chatbot (baseline: all contacts human)	$750	$3,000	$15,000
Shopify Inbox agent ($0 tooling)	$450	$1,800	$9,000
Tidio + Lyro (tooling $141 / from $341 / custom)	$591	from $2,141	$9,000 + custom pricing
Gorgias + AI Agent (tooling $107 / $486 / $1,932)	$557	$2,286	$10,932
Custom build (tooling $630, flat)	$1,080	$2,430	$9,630

Three readings. At 500 orders, the free agent is unbeatable on cost and the paid apps cluster near it — the build costs roughly double the best metered route, and buys control you may not need yet. At 2,000 orders, the metered routes still lead, but the build is within ~$150–$300 a month of them; the five tests, not the spreadsheet, should decide. At 10,000 orders, per-resolution billing inverts the table: Gorgias’s meter passes the build’s flat $630 tooling line, while Tidio’s self-serve quotas run out entirely (Lyro tops out at 1,000 conversations a month before sales-led plans). Inbox remains cheapest at every size on pure cost — what it doesn’t sell you is the helpdesk depth, channels and action menu of routes 3 and 4.

Sample store P&L (the 2,000-order store, average order value $75 — an assumption): revenue $150,000/month. The support-chat function costs 2.0% of revenue with no chatbot ($3,000), 1.2% with the Inbox agent ($1,800), about 1.4% with Tidio + Lyro (from $2,141), about 1.5% with Gorgias ($2,286), and about 1.6% with a custom build ($2,430). Every route beats doing nothing — and the gaps between routes are smaller than the effect of the deflection rate, which is why the next paragraph matters.

Sensitivity — deflection is the lever. If your AI resolves 60% instead of 40%, human cost falls a further $150 / $600 / $3,000 a month across the three sizes. On metered routes you hand part of that gain back (more resolutions billed); on a build you keep nearly all of it (more conversations at $0.001 each). Re-run the tables with your own contact rate and a 30–60% deflection range: if the ranking doesn’t change, your decision is robust; if it flips, decide on the five tests instead.

The Shopify training-data checklist

Whichever route you choose, answer quality is decided by the same preparation. This is the checklist the app setup wizards don’t give you:

1. Catalogue truth. Titles, variants, stock status, and the details customers actually ask about — materials, sizing, compatibility — kept in sync, not exported once.
2. Policy text as written. Shipping times by region, returns window and its exceptions, warranty terms — the version you would defend in a dispute, because the bot will quote it.
3. An order-data path. How tracking information flows from fulfilment to the bot (carrier, tracking app, fulfilment service), so “where is my order” gets live data, not a generated guess.
4. Real tickets. Three to six months of past conversations and your team’s macros: real phrasing, real edge cases, the answers your team actually gives.
5. Brand language. How your store names its products and policies, and the words it never uses.
6. Hard exclusions. Questions the bot must never answer alone — safety claims, discount promises, legal threats, anything regulated in your category — routed straight to a human.
7. Escalation triggers. High order value, repeat contact, anger signals, or an explicit request for a person.
8. A test set. 50–100 real questions with approved answers, re-run after every policy or catalogue change. A bot that passed in January can be wrong by March.

Where each route quietly gets expensive

Inbox: the cost is capability, not money — when a second channel or a real ticket queue arrives, you migrate under pressure. Tidio: three separate meters (billable conversations, Lyro conversations, Flows visitors) renew on different clocks, and Lyro bills for conversations it joins, resolved or not. Gorgias: the double count — every AI resolution is also a billable ticket — plus a small included AI block (30 interactions) a mid-size store exhausts in its first week. A build: upkeep hours, rework when Shopify changes an API or your theme changes, and the discipline of the test set. None of these is a reason to avoid a route; all belong in the budget before you choose it.

FAQs

Is Shopify Inbox’s free AI agent enough for a small store?

Often, yes. If your questions are mostly product and order-status questions on your storefront, and volume is low, the free agent covers the core job and the model shows it as the cheapest route at every size. Its documented limits — storefront chat only, new customer accounts required, no multi-channel helpdesk — tell you when you’ve outgrown it.

At what order volume does a custom build start paying?

In this model, between 2,000 and 10,000 orders a month: the build’s flat tooling cost (~$630/month) is passed by the metered bill somewhere in the mid-thousands of orders, depending on your deflection rate and which app you’d otherwise run. Below that range, building is a control decision, not a cost saving.

Can a chatbot take actions — refunds, address changes — or only answer questions?

Both are possible, within limits. Vendor agents ship a defined action menu (order tracking is standard; Gorgias sells additional “Actions”). A custom build can call Shopify’s APIs for the actions you scope, but personal data needs Shopify’s protected-data approval, and write actions deserve human-approval guardrails. Start read-only; add actions one at a time.

What does a custom chatbot still cost me at Shopify itself?

Nothing extra. Your plan and card rates don’t change, and the build replaces app subscriptions rather than adding a Shopify fee. The ongoing platform cost of a build is your own upkeep line, not a Shopify line.

The short version

Price the meter, not the sticker. At low volume, take Shopify’s free agent or a metered app and spend your effort on the training-data checklist. As volume grows, watch the month your per-resolution bill crosses a build’s flat cost — and decide the rest on question mix, actions, channels and ownership. If your five tests point to a build, or you want this model re-run on your store’s real contact rate and deflection numbers, AI Dev Expert scopes Shopify chatbots in writing before any price exists: see our AI chatbot development service, our Shopify development service, or request a custom quote.

— AI Dev Expert. Market figures come from the third-party sources cited inline (Shopify pricing and Help Center documentation; Tidio’s pricing page; Gorgias’s pricing page and plan grid; Intercom; OpenAI’s published API rates), checked November 2026. The cost model is illustrative, with assumptions stated so you can substitute your own. Researched and drafted with AI assistance; reviewed against those sources.