AI for E-commerce: Where It Pays and Where It Doesn't

Under roughly 300 support tickets a month, use the AI already included with your platform and fix the support leaks first. A per-resolution support agent is a volume purchase. Automate order-status questions first, or wait.

Proof status. Prices came from vendor pricing pages on 21 August 2026 and each figure is linked. Secondary-source figures say so. Ticket mix and labour cost in the example are assumptions, not measurements. Replace them with your own. Vendor prices move, so check the links before committing.

The operator and the leak

Maya runs a home-goods Shopify store with roughly 400 SKUs, about 1,800 orders a month, and two people who fit support around other work. She is not drowning. Vendors are pitching. This is not a “15 best AI tools for Shopify” list.

Four questions eat the mornings:

• “Where is my order?” It is the most common and the order data can answer it.
• “Can I change/cancel this?” It is time-sensitive; a wrong answer costs money.
• “Does this fit / what’s it made of?” The product page already holds the answer.
• Returns and exchanges. These follow policy, repeat often, and can become emotional.

Three jobs are repetitive and low-judgement. One is time-sensitive and expensive to get wrong. That ticket mix matters more than a tool feature grid.

Separate copy work from support work

Job	What happens	Cost and risk
Catalogue and copy	Product descriptions, alt text, meta descriptions, subject lines	Cheap, low risk, usually bundled
Support deflection	Agent reads, finds an order, replies without a human	Per resolution; most spend is here
Order-status automation (WISMO)	Look up order, read tracking, reply	High volume, low risk, narrow

Most stores buy support deflection when WISMO is the actual job.

Compare the four options

Approach	Good for	Cost shape	Catch
Built-in platform AI (Shopify Magic / Sidekick)	Copy, store questions, admin	$0 extra	Does not resolve support tickets
Helpdesk + AI agent (Gorgias, Intercom Fin)	Ticket deflection at volume	Base plan + resolution fee	Same ticket can bill twice
Chat-first AI (Tidio/Lyro and similar)	Pre-purchase, lower volume	Lower monthly floor, capped conversations	Weak with live order data
Fix the process first	Stores under ~300 tickets/mo	$0	It is unglamorous. It is often right.

Price the support agent

Shopify plans cost $39 / $105 / $399 per month for Basic / Grow / Advanced in US monthly pricing; annual pricing is roughly 25% less. Shopify Magic and Sidekick are included at no additional cost on every tier. Its plan page lists “Sidekick, AI assistant for commerce” as a core feature. (Shopify pricing, page geolocates; USD figures are secondary.)

Product	Published price	Billing detail
Intercom Fin	$0.99 per outcome; $49/month base includes 50 resolutions	Resolution or procedure handoff bills once per conversation; lead qualification costs more. Fin pricing docs
Gorgias	$10 / $60 / $360 / $900 per month for 50 / 300 / 2,000 / 5,000 tickets	AI Agent adds $0.90–$1.00 per conversation; overage $0.32–$0.40. (secondary; check Gorgias pricing)
Tidio	Free $0 (50 conversations); Starter $24.17/mo; Growth from $49.17/mo; Plus from $300/mo	Growth covers up to 2,000 conversations; Lyro add-on starts at $32.50/mo. Tidio pricing

Maya’s 1,800 orders produce roughly 270 tickets a month, assuming a 15% contact rate. Measure yours. On a Gorgias-style plan, that is the $60 tier. Assume the AI resolves 45%. Intercom case studies put Fin’s real-world resolution rate at 42–50% (secondary), so 45% is a planning number, not an 80% sales-deck claim.

Input	Math	Result
AI resolutions	270 tickets × 45%	≈ 122
AI fee	122 × ~$0.95	≈ $116/month
Support stack	$116 AI fees + $60 base	≈ $176/month
Labour replaced	122 × 4 minutes = 8.1 hours; 8.1 × $20/hour	≈ $162

At this volume, the tool costs slightly more than the labour it removes, before Maya configures it. Higher volume helps amortise the base fee. A more repetitive mix raises resolution. Tickets that need judgment leave a human doing cleanup after the paid AI attempt.

Put hard gates on the risky cases

Helpdesk-plus-agent pricing can bill the ticket and the resolution fee for one conversation. Model the invoice as base + (tickets × resolution rate × per-resolution fee), not a flat subscription.

Order changes need a human gate. “Can you cancel it?” can have a real deadline; saying yes after the warehouse picked the order can create a refund, return shipment, and an angry customer.

A vendor may count a resolution when the customer stops replying. Customers also stop replying when they give up. Watch re-open rate and follow-on ticket rate, not only a deflection dashboard. An hourly catalogue index can also answer “is this in stock” incorrectly during a sales spike.

Do not publish raw bulk-generated copy over hundreds of near-identical SKUs. It produces thin, duplicative pages. Generate a draft, then edit it.

6. Spend $0 before buying an agent

Under roughly 300 tickets a month, make these changes first:

1. Put tracking in the shipping-confirmation email, header order lookup, and order-status page. A large share of WISMO traffic starts with a buried email.
2. Answer fit and material questions on the product page. Twenty identical monthly questions point to a page defect.
3. Write the returns policy in plain English and link it in three places.
4. Use helpdesk macros before agents. Canned replies have zero marginal cost and do not invent refund policy.

If tickets still hurt after that, run the resolution math. The ticket mix will be cleaner too.

Bottom line

My take: use platform AI already included in the plan. Fix the four process leaks that create most tickets. Then compare a per-resolution agent with your measured ticket mix. Below ~300 tickets a month, the math usually does not support a dedicated support agent.

---

More on this decision, three ways to look at it: