When to Add a Helpdesk, and Why the AI Agent Might Not Work on Your Platform

If you run WooCommerce, Magento or BigCommerce, the AI agent in the leading ecommerce helpdesk does not work on your store. That's not a review. That's Gorgias's own documentation: "AI Agent is not supported on stores that use BigCommerce, Magento or WooCommerce"^[1].

Same restriction on their AI Agent for Sales beta^[2]. A connected Shopify store is a listed requirement^[1]. So if you're on one of the other three, you can buy Gorgias as a helpdesk and you cannot buy the thing they market.

Do you actually need a helpdesk yet?

Three signals. You need at least two.

Volume. Under about 100 tickets a month, a shared inbox is genuinely fine and probably better. Above 300 a month, a shared inbox is costing you more in confusion than a helpdesk costs in money.

More than one person answering. This is the real trigger, not volume. Two people in one inbox means duplicate replies, missed messages and no accountability. That's what a helpdesk fixes first and best.

More than one channel. Email plus Instagram plus WhatsApp plus a contact form, all landing in different places, is where things get genuinely lost.

If you have one person doing 80 emails a month, you don't need this. You need canned responses and a better FAQ page.

What we found: the resolution numbers are three different metrics

Here's the analysis worth having before any demo.

Every AI helpdesk quotes a resolution rate. Almost nobody defines it, and there are three different things being measured that get used interchangeably.

Deflection. The ticket never reached a human. Includes customers who gave up and left. Highest number, least meaningful.

Containment. The conversation stayed in the channel. Doesn't mean it was solved.

True resolution. The customer's problem was actually fixed. Lowest number, only one that matters.

The spread between deflection and true resolution can exceed 30 percentage points, which is why a vendor quoting 70% and your team measuring 40% can both be telling the truth.

The three questions that cut through it

Ask these in the demo. The answers are more informative than the numbers.

"Is that deflection, containment or true resolution, and how do you define the difference?"

"What's your AI-handled CSAT compared to your human-handled CSAT, for the same team?" It's normally several points lower and almost nobody publishes it.

"When the AI hands off to a human, does the human get the full conversation context, and can they take over mid-conversation?" Nobody markets handoff quality, which is odd given it's where the customer experience actually breaks.

The pricing, and the double-billing pattern

Gorgias

Priced per ticket rather than per agent, with unlimited users. Their pricing page frames it as AI Agent on every plan, paying only when it resolves a conversation, with the helpdesk scaling from 50 to 5,000 tickets a month and never priced per agent^[3].

The published tiers run roughly $10/month for 50 tickets up to $900/month for 5,000, with annual billing about 16% cheaper, and AI resolutions at $0.90 to $1.00 each depending on billing term^[4].

The mechanic to understand before you buy: an AI-handled conversation can count as both an AI resolution charge and a helpdesk ticket. If a human picks it up within 72 hours it's billed as a ticket only^[4]. That's a reasonable design and it means your unit cost depends on your handoff rate in a way that's hard to model in advance.

Those tier figures come from third-party analysis rather than Gorgias's own page, which doesn't publish the full grid^[3]. Confirm them in writing.

Zendesk

Per agent, per month. UK annual pricing: Support Team £15, Suite Team £45, Suite Professional £89, Suite Enterprise on request. Copilot is a £40/agent/month add-on, the Workforce Engagement bundle another £40, Contact Centre £72^[5].

AI agents are included across Suite and Support plans on a consumption model, and Zendesk is explicit that you pay only for requests successfully resolved by the AI agent without escalation to a human^[5].

That exclusion is worth noting, because it's the opposite of how Intercom does it, and it's a genuine merchant-favourable design.

Intercom Fin

$0.99 per outcome across all plans, plus seat pricing on Essential, Advanced and Expert tiers, with 20 free Lite seats on Advanced and 50 on Expert^[6].

Read the definition of outcome carefully. It counts when a customer confirms resolution, when they don't ask for more help after Fin responds, or when Fin completes a workflow, including handoffs^[6]. You're charged once per conversation even if multiple questions are answered^[6].

So Intercom charges for a handoff and Zendesk doesn't^[5][6]. On a team with a high handoff rate, that difference is larger than the headline price gap. Neither is hiding it. It's just the sort of thing nobody compares.

Tidio

Worth mentioning specifically because Lyro has native WooCommerce support, which puts it in a category of roughly one among the tools that do real commerce actions^[7]. Pricing is per conversation and turn-independent, which means a long conversation costs the same as a short one.

The platform problem, in detail

This is the section that made me want to write this article, because it's the clearest case of marketing and documentation disagreeing that I've seen in this industry.

Shopify

Everything works. Gorgias, Siena, Fin, Zendesk, all of it, with deep order actions: look up an order, edit it, cancel it, refund it, all from inside the conversation. Gorgias explicitly requires a connected Shopify store for AI Agent^[1].

If you're on Shopify, this is a straightforward buying decision and you can pick on price and UX.

WooCommerce

Gorgias's AI Agent does not work^[1]. You can connect Woo for data visibility, and you cannot use the AI automation, which is the product they sell.

The platform-agnostic vendors (Zendesk, Intercom, Ada) will connect to Woo over API, and the depth of what they can actually do varies enormously. The question to ask, and to make them demonstrate rather than assert: can it issue a refund on my Woo store, or can it only read the order and tell me to go and do it?

That distinction is the entire value. An AI that reads your order data and drafts a reply saves a few seconds. An AI that processes the refund saves the ticket.

Tidio Lyro's native Woo support makes it worth a look on this platform specifically^[7]. The other realistic route on Woo is a generic helpdesk plus your own automation against the Woo REST API, which is more work and gives you exactly the actions you need.

Magento and Adobe Commerce

Same exclusion from Gorgias^[1]. Magento merchants are usually large enough for Zendesk or Salesforce, and usually have the engineering to build real integrations, which is genuinely the better answer at that scale.

Adobe's own agentic story runs through Agent Orchestrator on Adobe Experience Platform, working across Real-Time CDP, Experience Manager, Journey Optimizer and Customer Journey Analytics^[8]. Each of those is separately licensed, so "Adobe has AI agents" and "your Adobe Commerce licence includes AI agents" are different statements.

BigCommerce

Same Gorgias exclusion^[1]. BigCommerce's API is genuinely good, which makes a custom integration more viable here than on most platforms, but you're building rather than buying.

Custom and headless

You build the integration, which means you get exactly the actions you want and you own them. The useful design decision is what the AI is allowed to do without a human. Read order status, yes. Issue a refund under £20, probably. Issue a refund over £200, no.

Put the boundary in the tool definitions rather than in the prompt. Prompts are advisory. Tool permissions are enforcement.

The helpdesk decision table

Situation	Do this	Rough cost	Why
Under 100 tickets/month, one person	Shared inbox and canned replies	£0	Nothing to coordinate yet
Shopify, 300+ tickets/month	Gorgias	From ~$60/mo plus AI[4]	Deepest Shopify order actions[1]
WooCommerce, want AI	Tidio Lyro, or build against the Woo API	Per conversation[7]	Gorgias AI Agent doesn't run on Woo[1]
Magento or BigCommerce, want AI	Zendesk or custom, and test the actions	From £45/agent[5]	Same exclusion applies[1]
High handoff rate expected	Zendesk over Intercom on this axis	Varies	Zendesk excludes escalations from AI billing[5]
Large agent team, low ticket volume	Per-ticket pricing beats per-seat	Gorgias model[3]	Unlimited users
Small team, very high volume	Per-seat pricing beats per-ticket	Zendesk model[5]	Cost stops scaling with volume

What your tickets are actually about

Before you automate anything, categorise a month of tickets. The distribution usually says the software isn't the answer.

"Where is my order" is normally the largest single category, and it's a tracking communication problem. Proactive shipping updates cut it more than any AI agent will.

Returns and exchange questions are next, and they point at your returns flow rather than your inbox. US returns ran at 15.8% of sales in 2025 and roughly 19.3% of online sales^[9], so at any real volume this is a permanent structural share of your inbox, not a spike.

Product questions before purchase are a product data problem. If people are asking whether it's waterproof, put it on the page as a structured attribute. Baymard's research shows 39% of sites fail feature searches and 66% fail non-product searches like "returns policy"^[10], and both of those failures land in your inbox as tickets.

Disputes and chargebacks are their own thing. First-party misuse is rising for 64% of merchants in the MRC's 2026 survey of 1,278 merchants across 37 countries^[11], and those conversations need a human every time.

One more thing worth checking before you buy anything. On Shopify, semantic search in Search & Discovery is available on Grow, Advanced and Plus for stores under 200,000 products^[14]. If you're on Basic at £19/month billed yearly, upgrading to Grow at £49^[15] may remove more pre-purchase tickets than a helpdesk will, because people who can find the answer don't message you.

That's a £30 a month fix against a several-hundred-pound one. Check it first.

What the AI agent should and shouldn't be allowed to do

Once you've picked a tool, the configuration decision that matters most is scope. Not tone of voice, not the avatar. What actions the agent can take without a person.

A workable default, from most to least comfortable.

Always allowed. Order status lookup, tracking links, delivery estimates, returns policy explanation, product specification questions answered from your own product data, opening hours, stock availability.

Allowed with a value cap. Issuing a return label. Refunding under a threshold you set, which for most stores is somewhere around a fifth of average order value. Reshipping a single low-value item. Applying a goodwill discount within a fixed range.

Never without a human. Anything about a chargeback or dispute. Anything where the customer is angry, which your sentiment signal should route immediately. Anything involving a legal claim, a safety issue or a data request. Anything above your value cap.

Two implementation notes that make the difference between this working and not. Put the boundary in the tool definitions, not in the system prompt, because a prompt is a suggestion and a permission is enforcement. And log every action the agent takes against the order, so a human picking up the conversation can see what's already been promised.

The measurement that tells you if it's working

Not resolution rate. Repeat contact rate within seven days.

An agent that closes a conversation and generates a second ticket four days later hasn't resolved anything, it's deferred the cost and annoyed someone. Repeat contact rate is harder to game than resolution rate and it's the number I'd hold a vendor to. Ask whether they can report it, and watch whether the answer is a straight yes.

Cost modelling that survives contact with reality

Before signing anything, model twelve months at three volumes: your current ticket count, your current count plus 50%, and double.

Per-agent pricing is flat against volume and scales with headcount, which suits a small team handling a lot of tickets. Zendesk's Suite Team is £45 per agent per month on annual billing, with Suite Professional at £89^[5]. Three agents on Suite Team is £1,620 a year regardless of whether you handle 500 tickets a month or 5,000.

Per-ticket pricing is the inverse. Gorgias's model is unlimited users with the bill scaling on ticket volume^[3], which suits a larger team on moderate volume and punishes growth.

Per-outcome AI pricing sits on top of either. At $0.99 an outcome^[6] or $0.90 to $1.00^[4], a store handling 2,000 conversations a month with 50% AI resolution is looking at roughly $1,000 a month in AI charges alone, before the helpdesk.

That's the number to put next to the cost of a part-time person, because that's the actual comparison. And it's the reason the platform question at the top matters commercially rather than just technically: if the AI doesn't run on your platform, that whole line disappears from the model and so does the saving^[1].

The compliance bit nobody mentions in a demo

If you sell into the EU, an AI agent talking to a customer needs to be disclosed as an AI. That's not a nice-to-have and it isn't satisfied by a line in your privacy policy.

The practical implementation is a disclosure before the first AI-generated reply, on every channel, in the language the customer is using. Ask your vendor to show you exactly where it fires, and check it fires over the API as well as in their own widget, because a lot of implementations only cover the widget.

Second thing: your customer's message goes to a third-party model provider. That's a processor relationship and it needs to be in your records. Ask what PII gets redacted before egress, and what the retention period is on the conversation.

What agentic commerce changes here

The interesting shift is that support increasingly happens where the customer is, not where your widget is.

Google's Universal Commerce Protocol, launched at NRF on 11 January 2026 with Shopify, Etsy, Wayfair, Target and Walmart as co-developers, covers post-purchase support as part of the shopping journey it defines^[12]. The Agentic Commerce Protocol maintained by OpenAI and Stripe covers cart, orders and authentication at spec version 2026-04-17^[13].

Which means a customer may ask an assistant where their order is, and never touch your site. That's fine if your order status is queryable and accurate. It's a problem if the only place that information lives is inside your helpdesk's UI.

The unglamorous implication: your order status API matters more than your chat widget. Most merchants have spent five years on the widget.

What to do this week

1. Categorise one month of tickets. Where is my order, returns, product questions, disputes, everything else. The distribution is the whole diagnosis.
2. Check your platform against the vendor's documentation, not their marketing page. Gorgias's docs and marketing say different things about Woo, Magento and BigCommerce^[1].
3. Make any AI vendor demo a real action on your platform. Issue a refund. Not read an order. Issue a refund.
4. Ask whether handoffs are billed. Intercom counts them as outcomes; Zendesk excludes escalations^[6][5].
5. Fix your top ticket category before automating it. Usually shipping comms, and it's free.
6. Check the AI disclosure fires on every channel. Over the API as well as inside the vendor's own widget.
7. Make sure order status is available outside your helpdesk. Agents will ask for it^[12].

The takeaway

The leading ecommerce helpdesk's AI agent runs on Shopify and nowhere else, and they say so in their documentation while implying otherwise elsewhere^[1][2]. The platform-agnostic vendors will connect to anything and mostly can't take real commerce actions on it. That gap is the actual state of the market and it's why "which helpdesk is best" has a different answer depending on a platform choice you made years ago.

The stance I'd defend: most stores buying an AI helpdesk should first spend two weeks on shipping notifications and product page data, because those two changes remove more tickets than any agent resolves, and they cost nothing per conversation forever.

What's your biggest ticket category? And is the fix really in the inbox?

Sources

1. Set up and use AI Agent on Chat. Gorgias documentation. Accessed 22 July 2026. States AI Agent is not supported on BigCommerce, Magento or WooCommerce stores.
2. Intro to AI Agent for Sales (beta). Gorgias documentation. Accessed 22 July 2026.
3. Gorgias Pricing. Gorgias. Accessed 22 July 2026.
4. Gorgias Pricing 2026: Plans, AI Costs, and Alternatives. Lindy. Third-party analysis; Lindy sells a competing AI product. Confirm figures with the vendor.
5. Zendesk Pricing (UK). Zendesk. Accessed 22 July 2026.
6. Intercom Pricing. Intercom. Accessed 22 July 2026.
7. Tidio Pricing. Tidio. Accessed 22 July 2026.
8. Adobe Launches Adobe Experience Platform Agent Orchestrator. Adobe newsroom, March 2025.
9. Consumers Expected to Return Nearly $850 Billion in Merchandise in 2025. NRF with Happy Returns, 15 October 2025.
10. Ecommerce Search UX Best Practices: the 8 search query types. Baymard Institute, updated 29 April 2026.
11. MRC Releases 2026 Global eCommerce Payments and Fraud Report. Merchant Risk Council, 18 March 2026. 1,278 merchants, 37 countries.
12. Under the Hood: Universal Commerce Protocol (UCP). Google Developers Blog, January 2026.
13. Agentic Commerce Protocol specification. Maintained by OpenAI and Stripe. Spec version 2026-04-17.
14. Modifying search with Shopify Search & Discovery. Shopify Help Center. Accessed 22 July 2026.
15. Shopify Pricing (UK). Shopify. Accessed 22 July 2026.