How to choose an AI support agent for your store

Every AI support tool demos beautifully. You type a question, it answers in a friendly voice, and you picture the repetitive tickets melting away. Then you install it on your real store, point it at real traffic, and find the polished demo and the day-to-day reality are two different things — including the bill, which can be three or four times what you budgeted for the same volume of conversations.

This is a buyer's checklist, written by people who build one of these tools (Loqara) but would rather you pick the right one than the wrong one. It's vendor-neutral on purpose: the questions below apply whether you're looking at Loqara, an incumbent helpdesk, or a flow-builder. Work through them in order, run the math on your own numbers, and the field narrows fast. (If you're still assembling a shortlist to run through it, start with our round-up of the best chatbot platforms.)

Choose by what decides value, not the feature list: the pricing model (per-resolution, per-conversation, and per-seat can be several times apart on the same traffic), grounding so it answers only from your content, real store actions like product search and order lookup, clean human handoff, and a one-line install. Then test it on real traffic before you pay.

The criteria that actually matter

Feature lists are mostly noise — every vendor checks every box. What separates an agent that pays for itself from one that quietly generates more work is a much shorter list. Here's the checklist, then a closer look at each criterion.

This is a vendor-neutral checklist — hold every tool you're considering, Loqara included, up against the same rows.

Pricing model

The one buyers underestimate most, so it leads the list. Two tools can advertise nearly identical per-unit prices and bill you wildly differently because the unit differs. We give it a full section below — but at the checklist stage the rule is simple: ignore the headline number, ask what counts as one billable event, then multiply by your real monthly volume.

Grounding

The biggest risk in AI support is a confident, wrong answer — a returns window you never offered, a shipping date you can't hit. A grounded agent answers only from your own content (help pages, policies, product data) using retrieval, cites where each answer came from, and says "let me get a human" when it doesn't know. Ask any vendor how they prevent made-up answers; if the demo bot invents a policy, that's behaviour you'll be paying to clean up. It's the difference between trust and a refund policy you never had.

Real store actions

FAQ keyword matching is table stakes. The agents that earn their place can search your live catalog and look up a customer's order after an identity check. That's what deflects the most common ticket of all — "where's my order?" — and turns a browsing question into a sale. A tool that can only search a help center is a glorified FAQ widget. This one capability is usually the feature that pays for the tool.

Human handoff

The feature that makes customers trust a bot is knowing they can reach a person. Check that handoff is clean, carries the full conversation so the customer never repeats themselves, lands in a shared inbox your team already watches, and is available on the plan you'd actually buy — not gated to the top enterprise tier.

Install effort

A one-line embed you can preview and ship the same afternoon beats a six-week integration every time — the longer the setup, the longer you pay for value you can't use. Equally important: make sure you can cancel or downgrade without friction, so a wrong guess isn't expensive.

Languages & voice

Your customers don't all type in English, and a growing share would rather talk than type. Check whether the agent answers naturally in the languages your store serves, and whether it offers a real-time voice agent for phone-heavy categories, older demographics, or hands-busy shoppers. Voice is still rare among support tools, so confirm rather than assume.

Analytics & CSAT

Decide your success metrics before you buy: deflection rate, handoff rate, leads captured, and CSAT. If a tool can't show you those numbers, you can't prove it paid for itself — or tune it. Here are the four metrics that actually matter, and why vanity numbers like "messages handled" mislead.

The pricing-model trap

If you take one thing from this guide, take this. The three pricing models in the market look comparable on a feature page and behave nothing alike on an invoice.

• Per resolution — you pay each time the AI "resolves" a conversation. Predictable at scale, but the vendor defines what counts as a resolution, and that definition can be generous. Common among premium suites: Intercom's Fin charges $0.99 per resolution, Zendesk bills AI usage per resolution on top of per-agent seats, and Gorgias layers per-resolution AI onto ticket-volume tiers (never per agent). Our Intercom, Zendesk, and Gorgias alternatives round-ups dig into each.
• Per conversation / interaction — you pay for every conversation, resolved or not, which means you also pay for the ones the bot fails. One customer thread is one unit, which makes it easy to forecast. Tidio's Lyro meters its AI this way from around $32.50/month (see our Tidio alternatives round-up), and Loqara prices per conversation too.
• Per seat (+ AI add-on) — you pay per human agent, then the AI is either a separate line or a capped inclusion. Zendesk runs $19–$115 per agent/month with Copilot an extra $50 per agent/month, while Re:amaze is $29–$69 per user/month with AI included up to usage caps — either way, the "AI chatbot" you're picturing often isn't fully in the base price.

Whatever model a vendor uses, ask one question: at my real monthly volume, what's the all-in cost — including the AI, including seats, including overage? Run that math on your numbers, not the headline price. And check whether the free tier renews every month or is a one-time taste that expires after a few hundred messages.

A simple way to decide

Match the tool to your stage, not the longest feature list. The right answer for a five-person store is the wrong answer for a 200-agent support org, and vice versa.

1. Just starting out (0–100 conversations/month). You want grounded answers, real order lookup, and a free tier you can prove value on before paying a cent. Predictable cost matters more than enterprise workflows you won't touch. A flat allowance or per-conversation tool fits cleanly here. See the best chatbots for online stores.
2. Growing store (hundreds/month). Now install effort, handoff on every plan, and per-conversation predictability separate the field. This is where a per-resolution suite can quietly outpace your budget — model it carefully. Platform guides help: WooCommerce and Shopify.
3. Scaling, multichannel (thousands/month, a team in the inbox). Deep automation across email, social, and SMS starts to justify a heavyweight helpdesk, and per-resolution pricing can pencil out at volume. Weigh it against the AI-agent-plus-live-chat approach. Shortlists: Gorgias alternatives and Zendesk alternatives.
4. Want to turn chat into a growth channel. If the agent should also capture leads conversationally, confirm that's built in rather than a bolt-on, and that you can measure the leads it produces.

A quick gut-check before you commit: install it on your real store, point it at a week of real traffic, and watch the metrics. A tool that can't earn its keep in a free or trial week probably won't earn it once you're paying. If you're on WooCommerce specifically, our walkthrough on adding an AI agent to WooCommerce shows exactly what that test looks like end to end.

Frequently asked questions

What's the most important thing to check before buying an AI support agent?

The pricing model, because it determines your actual bill more than the headline price does. Per-resolution, per-conversation, and per-seat pricing can differ by several times over on identical traffic, so model your real monthly volume against each tool's billable unit before you commit. Everything else — grounding, store actions, handoff — matters, but the pricing unit is the one that surprises people.

What does "grounded" mean and why does it matter?

A grounded agent answers only from your own content — your policies, FAQs, and product data — and says "I'm not sure" rather than guessing. It matters because a confident wrong answer (an invented returns window, say) creates more tickets and lost trust than no answer at all. Ask any vendor exactly how they prevent fabricated answers, and look for source citations as proof.

Should I pick a tool that charges per resolution or per conversation?

It depends on your volume and how the vendor defines a "resolution," which can be counted generously. Per-conversation or flat-allowance pricing is usually easier to forecast for small and mid-size stores because one customer thread equals one unit. Per-resolution suites can make sense at high volume with a dedicated team, but always run the math on your own numbers first.

Do I really need an AI agent that can look up orders and search products?

For most stores, yes — those are the tickets that actually repeat. "Where's my order?" and product questions make up the bulk of support volume, and only an agent with live order lookup and catalog search (not just help-center search) can deflect them. A tool that can only answer FAQs is a widget, not a support agent.

How long should it take to set up an AI support agent?

A modern, grounded agent with a one-line embed can be live the same day: add your content, connect your store platform, customise the widget, and paste one snippet. Heavier helpdesk suites take longer because there's more machinery to configure. Also confirm you can cancel or downgrade easily, so a wrong choice isn't costly.

Can an AI support agent handle voice and other languages?

Some can. Loqara, for example, includes an optional real-time voice agent and answers in multiple languages, while many tools are text-and-English-only today. If your customers prefer to call or shop in another language, confirm voice and multilingual support explicitly rather than assuming it's included.

---

The bottom line: the best AI support agent isn't the one with the longest feature list — it's the one whose pricing you can forecast, whose answers you can trust, that can take real store actions, and that you can install today. Work the checklist, run the math on your own volume, and test it on real traffic before you pay. Loqara was built for the lean-store end of that list: grounded chat and voice, real product search and order lookup, handoff on every plan, predictable per-conversation pricing, and one line to install.

Try Loqara on your store free — one line, no credit card, live in an afternoon.

Competitor pricing models referenced here are approximate and as of mid-2026 — always confirm on the vendor's site.