What an AI Shopping Assistant Actually Does on Your Store

Most merchants approach this as a purchase: which AI shopping assistant do we buy, and what does it cost. In 2026 that’s the wrong first question. If your store runs on Basic, Grow, Advanced, or Shopify Plus, you can turn on a customer-facing AI shopping assistant today for nothing, and Shopify documents in unusual detail what it reads, what it can do, and exactly where it stops.

That changes the decision. The useful questions are what the assistant can see, what it can’t do, and what a paid vendor would actually be selling you on top. We build and run these stacks for DTC brands, and this is the analysis we’d give a client sitting on a quote.

Every number below links to the vendor page it came from. Vendor marketing claims are labeled as vendor claims in the sentence where they appear, and all pricing was read off live pages on September 14, 2026, which vendors change without notice. We won’t re-caveat that in every paragraph.

This piece is about the conversational layer on your own store. If your question is how to get found inside AI surfaces in the first place, that’s a different job, covered in our look at getting discovered in AI and closing the sale on your own store.

Shopify Inbox is listed on the Shopify App Store at a price of Free, published by Shopify, with a 4.6 rating across 5,791 reviews as of September 14, 2026 (app listing (opens in new tab)). The listing’s own description of it is blunt: “Inbox puts an AI-powered sales associate on your storefront.” The plan gate is stated in the help documentation just as plainly: “Shopify Inbox is available only for stores on the Basic, Grow, Advanced, or Shopify Plus plan” (Inbox agent doc (opens in new tab)).

It isn’t on by default for stores that already had Inbox. Shopify’s documentation gives three assignment modes, all conversations, only when your staff is unavailable, or never, and says of the last one: “This is the default setting for stores that had Inbox installed before the Inbox agent was available.” The same doc adds, “Until you turn on the Inbox agent, your conversations stay with your staff.” If you installed Inbox years ago for live chat, the agent is sitting in your admin switched off.

What it does once it’s on, verbatim from that doc: “The Inbox agent can answer product questions, make personalized recommendations, help with order lookups, and add items to a customer’s cart.” It also replies in the customer’s language. The app listing describes a further layer for signed-in shoppers: “When buyers sign in with Shop, your Inbox agent can personalize replies using their preferences and purchase history from across Shopify.”

The free floor isn’t only Shopify’s. Microsoft’s Clarity app is listed on the Shopify App Store at Free with a 4.6 rating across 2,160 reviews (listing (opens in new tab)), Tidio publishes a free tier covering live chat with 50 users and Flows reaching up to 100 visitors a month (listing (opens in new tab)), and Rep AI is free to install with 100 visitors a month included (listing (opens in new tab)). A chat widget with a language model behind it is no longer something a merchant buys. Paid tiers buy volume, actions, integrations, and measurement, which is a much narrower purchase than most vendor pitches imply.

Shopify names the assistant’s sources exactly: “The Inbox agent answers customer questions by using your product catalog, store policies, Knowledge Base facts and files, and published storefront content such as pages, blog posts, sizing guides, and product care guides” (Inbox agent doc (opens in new tab)). It also draws a hard boundary around the open web: “It uses your store sources, not web search, for product facts, care instructions, usage ideas, and styling guidance. The Inbox agent bases product recommendations on your store catalog only.” Web search is permitted only as a cited secondary source for qualitative material like reviews, ratings, and social proof on a product that’s already in the catalog.

Facts and tone come from different places, and conflating them is the most common way merchants break this. The persona controls how the assistant communicates, its tone, wording, personality, and service style (persona doc (opens in new tab)). Facts typed into the persona box are explicitly second-class: “Store facts added only to the persona aren’t treated as source information and might not be used consistently.” A return window pasted into the persona field is a stylistic suggestion, not a policy the assistant is reading from.

The place those facts belong is the Knowledge Base app, which Shopify describes in one line: “This information acts as a data source that AI agents can use to generate responses” (Knowledge Base doc (opens in new tab)). Per the same doc, the app generates business facts out of your store settings, including languages, customer account options like submitting return requests or paying with store credit, shipping and delivery settings such as which countries you ship to and whether pickup in store is offered, and returns settings. That’s the native, free answer to “how do I stop it inventing my policies,” and it costs nothing but an afternoon.

Two consequences follow for merchandising. Your product copy, policy pages, and guides now have a second reader that answers customers out loud, so a thin product description isn’t just a conversion problem any more. And the liability is yours, stated in a caution box: “You’re responsible for the accuracy of the information that the Inbox agent provides to your customers.”

Training doesn’t change any of that. Shopify gives two training methods, practice scenarios generated from your catalog and persona, and rating real conversations, then says outright: “Training doesn’t change what your Inbox agent knows about your store” (training doc (opens in new tab)). Ratings produce suggested persona updates you review, and “Updates are never applied automatically.” Thumbs-down feedback tunes voice. Fixing a wrong answer means fixing the source it came from.

The binding limit is the first one in the docs: “It only operates within your online store’s chat widget” (Inbox agent doc (opens in new tab)). This is a storefront assistant, not an omnichannel agent. Email, SMS, Instagram DMs, and marketplace messages are all outside its reach, and the conversations doc closes the obvious workaround: “The Inbox agent can’t send email on your behalf. The customer must initiate email contact” (conversations doc (opens in new tab)).

The volume caps are published and generous. In a single conversation the agent responds to up to 100 messages for customers who aren’t signed in and up to 1,000 for signed-in customers, and each customer can start up to 50 conversations per day with your store (persona doc (opens in new tab)). Those are per-customer abuse controls, not commercial ceilings. A signed-out shopper would have to send a hundred messages in one sitting to reach the first one, so unlike a paid vendor’s monthly allowance, nothing here throttles a busy Black Friday.

Customer accounts are a real gate. The documentation states that “The Inbox agent doesn’t work with legacy customer accounts,” and per the persona doc, order lookup always requires the customer to sign in, whether or not you require sign-in for staff chat. If you’re still on legacy accounts, this isn’t a settings change. It’s an accounts migration, and it stands between you and both order lookup and the Shop-based personalization described above.

You can’t white-label it either. Verbatim: “Customers receive a notice that the chat is powered by Shopify and that they’re chatting with AI. This notice is displayed automatically and can’t be removed or customized.” For most brands that’s fine, and disclosure is where the regulatory wind is blowing anyway. For a brand that deliberately keeps its platform invisible, the chat window names Shopify permanently and can’t be changed.

The persona can’t be used to engineer around any of this: “A persona can’t override Inbox settings or built-in safety behavior.” Shopify’s doc specifically warns against personas written to prevent all staff handoffs, to require a product recommendation in every reply, or to route every conversation to a single contact. If your plan for the assistant depends on one of those, the plan is the thing that breaks.

Can it process a return, a refund, or a cancellation? Those actions aren’t among the capabilities described in Shopify’s Inbox documentation. That’s an absence in the docs rather than a statement that it’s impossible. Put the question to any vendor selling you an assistant that takes actions instead of answering questions. Action-taking is one of the few concrete things a paid tier can sell you over the free one, so make the vendor name the actions and show you the audit trail.

Handoff has two documented outcomes, and only one of them is what merchants picture. When staff handoff is available, “the conversation moves to the Unassigned folder in your Inbox, where your staff can pick it up. The conversation includes the AI’s prior context” (conversations doc (opens in new tab)). When it isn’t available, “the customer receives your store’s sender email address and can contact you by email.”

Availability is driven by the hours you set (availability doc (opens in new tab)). During staffed hours, a customer asking for a person reaches a person. Outside them, they get an email address. A brand that sets the agent to all conversations without staffing the inbox has quietly converted “talk to a human” into “here’s our email,” at the exact moment a frustrated shopper asked for help.

The decision that follows is a staffing decision, not a technology one. Run the agent on “only when your staff is unavailable” if you have coverage worth routing to, or accept the email fallback deliberately and measure how often it fires. That second number, the share of handoff requests that happened while no human was reachable, is one your vendor won’t report and your support team will feel first.

Sidekick is the assistant for you, inside the admin. Shopify defines it as “an AI-enabled commerce assistant in your Shopify admin, to get guidance, generate content, build apps, and complete tasks using everyday language” (Sidekick doc (opens in new tab)), running in two modes, chat and proactive recommendations through Sidekick Pulse on the admin home page. The considerations list adds that it “can’t make changes to your store without your approval, and will only present options for you to review and approve” (set-up doc (opens in new tab)). It’s a merchandiser’s tool, and it never meets a shopper.

The dates are easy to get backwards, because Shopify’s edition naming runs a season ahead of the calendar. The Winter ’26 Edition was announced on December 10, 2025 (Shopify news (opens in new tab)), and the matching cluster of Sidekick entries in Shopify’s changelog (opens in new tab) carries that same December 10 and 11, 2025 dating: proactive recommendations, app creation, natural-language theme edits, preference memory, and Flow automation generation. Sidekick Skills, the saved and shareable prompts, shipped earlier still, with a changelog entry dated October 3, 2025. Winter ’26 re-announced and expanded them rather than introducing them.

Later capability additions do carry 2026 dates in the changelog: payments and web performance data in Sidekick on January 6, 2026, creating customers and companies on February 10, 2026, and the Online Store Editor with Sidekick on mobile on June 17, 2026. Access isn’t uniform. App generation “is available only to stores on the Grow plan, Advanced plan, and Shopify Plus plan” (app generation doc (opens in new tab)), and Shopify states that “Sidekick Pulse is in early access and is available only to certain merchants,” with Pulse surfacing up to five suggested actions at a time and requiring recent sales activity, store owner or admin access, and Shopify Network Intelligence turned on (Pulse doc (opens in new tab)). Base Sidekick plan availability isn’t stated in the docs we read, so treat that one as undocumented rather than universal.

One seam connects the merchant assistant to the shopper assistant. The Inbox persona editor carries a “Refine with Sidekick” button, described in the persona doc as a way to get suggestions for tone, wording, personality, and service style based on your business needs. That crossover is the only documented one.

Shopify turns on the other half of this on your behalf: “Shopify’s agentic storefronts lets customers discover and purchase your products in AI channels, such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta. Agentic storefronts is active by default for eligible stores” (agentic storefronts doc (opens in new tab)). Ecommerce AI agents are already reading your catalog whether or not you’ve made a decision about it. What your own store already exposes at the protocol level is its own subject, covered in what your store already does for agents without you.

The state of in-surface checkout is genuinely different per vendor, and collapsing it into a single industry verdict gets it wrong in both directions.

OpenAI stepped back from running checkout itself. In a post dated March 24, 2026, the company wrote: “We’ve found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we’re allowing merchants to use their own checkout experiences while we focus our efforts on product discovery” (OpenAI (opens in new tab)). Instant Checkout wasn’t withdrawn, though. OpenAI’s help center, updated roughly August 28, 2026, still documents it: “For some eligible products and merchants, ChatGPT may also show an Instant Checkout option that lets you complete checkout in ChatGPT instead of leaving for the merchant’s site.” That same article tells merchants that “For merchants on Shopify, product data is already integrated into ChatGPT through Shopify Catalog,” and that “No additional work is required from individual merchants” (OpenAI help center (opens in new tab)). Shopify’s channel documentation corroborates the default from the platform side, describing ChatGPT as a product discovery channel whose customers are redirected to buy in your online store checkout, with no toggle for direct checkout in the admin.

Google is live and narrow. Its help page states that agentic checkout “is currently only available in the US and Australia, in English and for select merchants that accept Google Pay as a payment method,” and adds that eligibility requirements apply and merchant availability can change (Google Pay help (opens in new tab)). The same live page contradicts itself further down, where the troubleshooting section says it’s “currently available only in the US, in English.” We’re recording that as a contradiction rather than resolving it, because both statements sit on the page Google publishes for this feature as of September 14, 2026. On the merchant side, per Google’s UCP documentation, merchants remain the merchant of record, native checkout is the default integration path with an embedded iframe path reserved for specific approved merchants, and access runs through a waitlist rather than self-serve activation (Google UCP docs (opens in new tab)).

Microsoft shipped in January 2026 and enrolled you automatically. Its announcement states that “Copilot Checkout is beginning to roll out in the U.S. on Copilot.com, with partner activation across PayPal, Shopify, and Stripe, with more of their retailers ramping throughout the month” (Microsoft Advertising (opens in new tab)). Per that post, Shopify merchants are automatically enrolled following an opt-out window and manage Copilot Checkout from the Shopify admin rather than applying, while non-Shopify merchants apply through a form, and the merchant remains merchant of record owning the transaction, the customer data, and the relationship.

Microsoft’s second announcement in that post matters more to a store owner than its checkout does. Brand Agents are described there as “AI-powered shopping assistants that speak in your brand’s voice and guide customers naturally from curiosity to purchase,” and per the same post getting started means installing Microsoft Clarity on the Shopify store and then clicking Join Waitlist on the Clarity home page for early access, with insights delivered inside Clarity reporting engagement rates, conversion uplift, and average order value, plus a comparison of agent-assisted sessions against organic traffic. That’s a second free on-site assistant path, reached through a free app and gated behind that waitlist. Microsoft’s claim that agent-assisted sessions outperform unassisted ones is a vendor claim published without methodology, sample, period, or baseline, so treat it as positioning and use the Clarity comparison to produce your own number.

Shopify lists what breaks inside an AI channel’s direct checkout, and the same list appears on both its Google and Microsoft channel pages: subscriptions, product bundles, customizable products, and B2B-only products are unsupported, local delivery, pickup in store, and pickup points are unsupported, Google Analytics and custom pixels don’t fire with only server-to-server pixels running, and “chat or messaging experiences” are among the blocks that might not load. Your assistant doesn’t follow the buyer into someone else’s checkout. Whatever recovery, reassurance, or upsell it does for you on-site simply isn’t present in that flow.

The Shop app is the same story from the discovery side. Per Shopify’s documentation, “Conversational searches are available only to customers in the United States and Canada. Customers in other countries have access to the standard search experience using keywords,” and “You can’t opt out of your products being included in Shop search results” (Shop customer experience doc (opens in new tab)). The same applies to Shop Minis, where merchants can’t choose which products display or opt out.

Shopify also publishes a machine-readable skill file for personal AI agents at shop.app/skill.md (opens in new tab), whose front matter describes it as the “Ultimate personal shopping assistant: find, compare, buy, gift, and reorder products across the Shop catalog containing millions of stores.” It documents a six-step flow from sign-in through search, results, visualization, checkout, and orders, a checkout command set where checkout complete requires an explicit --confirm, and search filters covering ships-to, ships-from, condition, color, size, gender, price, and image lookup. An external agent can already do all of that against your catalog with no work from you, which is the honest frame for what an assistant on your own site adds: relationship, merchandising control, and the conversation itself.

Four vendors bill on four incompatible units, and that, not the sticker price, is the buying decision.

Rep AI bills on visitors. Its Shopify listing shows Pay-as-you-Grow free to install with 100 visitors a month, Starter at $104/month or $995/year for up to 10,000 visitors, Basic at $209/month for up to 25,000, and Standard at $368/month for up to 50,000, each with overage at $12 per additional 1,000 visitors (listing (opens in new tab)). You pay for traffic the widget sees, whether or not anyone talks to it.

iAdvize bills on conversations, with an SKU ceiling attached. Its listing shows Essentials at $290/month for up to 500 conversations and 1,000 SKUs, Starter at $520/month for up to 1,000 conversations and 10,000 SKUs, and Growth at $1,330/month for up to 3,000 conversations and 50,000 SKUs, with annual billing at 25% off monthly (listing (opens in new tab)). Tidio prices its Lyro chatbot at $39/month or $390/year to “Reach up to 200 Lyro conversations,” and its listing describes Lyro selling on the storefront by starting conversations on product pages, comparing products, and adding to cart (listing (opens in new tab)).

Run those against a single store and the divergence is the whole point. Take 100,000 monthly visitors, of whom 2% open a chat, so 2,000 conversations. On Rep AI Standard that’s $368 plus 50,000 visitors of overage at $12 per 1,000, which is $600, for $968 a month, or about $0.48 per conversation by our arithmetic off those published list prices. Hold the traffic identical and drop the chat rate to 0.5%, and the bill is still $968 while the unit cost quadruples to roughly $1.94 per conversation. iAdvize inverts it: Growth at $1,330 covers 3,000 conversations for about $0.44 each at full use, independent of how much traffic you sent.

Gorgias bills an outcome, and defines it tightly in its pricing FAQ: “An automated interaction is a customer request that’s fully resolved without a human agent, whether by AI Agent or another automation feature such as Flows, Quick Responses, or Order Management. These are billed and count toward your plan’s automation allotment. If a support agent ends up replying within 72h of an AI Agent resolution, that interaction is only charged as a billable helpdesk ticket instead” (Gorgias pricing (opens in new tab)). The same FAQ states that “AI Agent interactions are priced at $0.90 each on annual plans or $1.00 on monthly plans. Each AI Agent interaction also counts as a helpdesk ticket and is billed accordingly.”

Three different per-interaction numbers sit on that one page, describing three different situations, and a quote that cites one of them isn’t quoting the others. The feature comparison row shows $1.00, $0.90, $0.90, and $0.85 across Starter, Basic, Pro, and Advanced. The FAQ gives $0.90 annual and $1.00 monthly. And every plan card states that past your included allotment it’s “a $1.50 per automated interaction fee.”

The arithmetic is worth doing before the call, and this is ours off the list prices displayed on September 14, 2026. Gorgias Pro is $471/month on annual billing, including 2,000 tickets and 190 automated interactions. Put 2,000 fully automated assistant conversations a month through it and 190 fall inside the plan while 1,810 bill at the $1.50 overage rate, which is $2,715, landing the month at $3,186 with the ticket allotment absorbing the matching helpdesk tickets. That’s a blended $1.59 per automated interaction, above every per-interaction figure published on the page, because the marginal rate rather than the in-plan rate is doing the work.

The 72-hour reversal then pushes in a direction most merchants don’t anticipate. If a support agent replies to 15% of those shoppers within three days, 300 interactions stop being automated interactions and bill only as helpdesk tickets, which are already inside your 2,000, cutting $450 off the month. The assistant that performs worse costs less. That’s defensible pricing on Gorgias’s part, and it means a quarter-over-quarter drop in your AI invoice is not on its own good news.

Zendesk redrew its unit on May 18, 2026 by splitting containment from resolution. Its documentation introduced three tiers: an “Assisted escalation,” where “The AI agent contributed to the interaction before a human agent completed the resolution,” a “Contained resolution,” where “The AI agent handled the interaction to completion without the customer requesting further assistance,” and a “Verified resolution,” where “The AI agent successfully resolved the interaction” (Zendesk (opens in new tab)). The first two, per that article, don’t count against your resolution allowance. The dividing line is an LLM verification pass: conversations that pass it are Verified resolutions, and those that don’t are Contained resolutions.

A conversation the bot held to the end, with no escalation and no complaint, is free if it fails verification. Containment stopped being the thing you pay for. Any evaluation framework built around a containment or deflection rate is describing the market as it was before that change, and the metric your invoice now responds to is narrower than the metric your dashboard probably shows. Timing matters to the count as well: per the same article, email conversations end 72 hours after the last email, messaging conversations end two hours after the last message by default and can be extended up to 72, and voice ends immediately on hangup.

Two Zendesk billing platforms are live simultaneously, so the question “what does Zendesk charge for an AI resolution” has two answers in 2026. The prior platform persists, and Zendesk states that “Customers who are on this platform remain so until they move to resolution tiers,” with allocations of 5 automated resolutions per agent per month on Team, 10 on Growth and Professional, and 15 on Enterprise, a maximum of 10,000 allocated automated resolutions per year, and overage “billed monthly, regardless of your subscription terms” (prior platform doc (opens in new tab)). Ask which platform your account sits on before you model anything, because we’ve seen renewal conversations run on the wrong one. Zendesk publishes no dollar figure per resolution on either platform.

Intercom’s Fin bills outcomes, and publishes both the definitions and the prices: “You’re only charged for one outcome per conversation, even if Fin takes multiple actions,” across Resolution, Procedure handoff, Disqualification, and Qualification, where “Resolutions, Procedure handoffs, and disqualifications are $0.99 each. Qualifications are $9.99 each. You’re not charged when a conversation is simply passed to your team without an outcome” (fin.ai pricing (opens in new tab)). Point Fin at pre-purchase conversations and you’re generating Qualification events at ten times the price of a resolution, which is a trap we’ve already worked through in detail in our breakdown of how AI vendors define a resolution and what it costs.

So cost per resolution isn’t a comparable figure across these vendors. A Rep AI visitor, an iAdvize conversation, a Gorgias automated interaction, a Zendesk verified resolution, and a Fin outcome are five different events with five different triggers, and only two of them require that no human was involved. Only one requires passing a verification pass. The only way to compare quotes is to take your own conversation log, classify it against each vendor’s definition, and price it five times. If you’re also choosing the helpdesk underneath the assistant, that’s a separate decision, and we’ve worked it through in choosing between Gorgias and Zendesk for a DTC helpdesk.

One product category gets miscounted in this comparison regularly. Octane AI bills credits where, per its pricing page, “1 credit = 1 quiz engagement (when someone completes your quiz),” with Basic at $50/month for 400 credits, Plus at $200/month for 2,200, and overage credits at $0.10 each (Octane AI pricing (opens in new tab)). Guided selling by quiz and conversational selling by chat answer different questions for a shopper, and a quiz platform priced per completion isn’t an AI shopping assistant no matter which list it appears on.

Shopify reports four metrics on the Inbox overview, and names them precisely: “Assisted sessions: The number of chat sessions in which a customer interacted with the Inbox agent,” “Assisted orders: The number of orders placed by customers who interacted with the Inbox agent,” “Satisfaction rate: The percentage of rated responses that received positive feedback,” and “Avg. response time: How quickly the Inbox agent responds to customer messages” (conversations doc (opens in new tab)).

Those are engagement metrics, and two of them need reading carefully. Assisted orders comes with no documented attribution window and no documented model, so the documentation doesn’t tell you over what period the order must follow the chat or whether the interaction has to precede the order at all. Treat it as an influence number, not incremental revenue, and don’t put it in a board deck as lift. Satisfaction rate is a percentage of rated responses rather than of all responses, and chat rating rates are typically small, so it describes the subset of customers motivated enough to press a button.

There’s no containment, resolution, deflection, or handoff metric in that set. If you want those, you define and compute them yourself or get them from a support vendor whose billing depends on them. Shopify does give you conversation topic tags to work from, naming Sales, Support, and Store Policies for AI conversations plus a ten-label classification for customer messages covering checkout, edit orders, order status, pricing and discounts, product details, product interest, returns and delivered orders, shipping, store info, and wholesale, with one caveat stated in the doc: “Inbox conversation automatic topic labels are only available in English.”

Four numbers are worth defining before you switch anything on, and none of these are industry standards, because none of the vendors above publishes them in this form. Set your own definitions, write them down, and keep them stable across vendor changes.

If the free assistant is too limited and the paid ones are priced on a unit that doesn’t fit you, the build path is further along than most merchants assume. Shopify exposes Storefront MCP servers with cart and policy tools at https://{shop}.myshopify.com/api/mcp, and states that “Storefront MCP servers don’t require authentication” (Storefront MCP docs (opens in new tab)). The catalog tools sit elsewhere, and Shopify is explicit that “The UCP catalog tools ( search_catalog , lookup_catalog , and get_product ) are available on a separate endpoint from the standard Storefront MCP tools,” reachable at /api/ucp/mcp (storefront catalog docs (opens in new tab)).

The catalog endpoint has a gate the standard one doesn’t: per the same documentation, the endpoint “requires an agent profile” and “every request must include a meta.ucp-agent.profile URL pointing to your agent’s UCP profile,” and the tools returned depend on the capabilities your agent advertises. Shopify also flags that “Some stores may restrict access. Always test with your specific store” (storefront server docs (opens in new tab)). Its overview describes a Chat UI as a theme extension that shows a customer-facing chat window with your app backend acting as the MCP client, so the storefront surface is a documented pattern rather than a custom build.

What that leaves you building is the assistant layer, the interface, and the evaluation, not the commerce plumbing. Shopify’s own guidance on the policy tool points back at the free surface anyway, advising developers to use only the provided answer and not to include external information that might be inaccurate, and to consider the Knowledge Base app for better store policy management. Build if you need actions, channels, or data the packaged assistants don’t reach. Building to save the subscription is usually the wrong trade once you price the evaluation work.

The pre-launch work is mostly content and policy, not configuration, and Shopify names one legal item directly: per the Inbox documentation, merchants running a custom privacy policy should disclose that they use Shopify to power AI-assisted chat and that Shopify processes customer data including chat communications to generate responses and improve the agent over time. Do that one first, because it’s the only item on this list with a compliance deadline attached to launch rather than to performance.

Turn on the free one first, on the assignment mode that matches your staffing, and give it thirty days against metrics you defined rather than metrics a vendor chose. You’ll learn more about whether conversational commerce works for your catalog from your own assisted-order data than from any vendor benchmark, and the documented limits above tell you in advance which gaps you’d be paying to close: channels beyond the storefront widget, actions the Inbox docs don’t describe, integrations with your helpdesk, and measurement that survives a change of vendor.

Buy when you can name the specific limit you’re buying past, in a unit that fits your traffic and chat rate, and with your own conversation log priced against that vendor’s definition of a resolution. That’s a real purchase. Buying a chat widget in 2026 isn’t.

If you’d rather have someone map the conversion path around it first, book a free Growth Audit and we’ll show you where your store is losing the people the assistant is meant to catch.