Plugin or Custom AI Chatbot for WooCommerce? Costs and Trade-offs

For most WooCommerce stores, a plugin or SaaS chatbot is the right first step. It is live in days, and for a few thousand conversations a month it costs less than building your own. A custom chatbot starts to pay off when you need answers grounded in data a plugin cannot reach (B2B prices, ERP stock, complex returns), when per-resolution fees grow with your volume, or when you need control over data and behaviour for EU compliance.

Below we compare the options, show how to estimate running costs from real API prices, and give a checklist for choosing.

What are the options?

There are three broad routes.

1. WordPress plugin with your own AI key. You install a plugin, paste an OpenAI or similar API key, and point it at your products and pages. You pay the plugin licence plus API usage.
2. SaaS chatbot or helpdesk AI agent. A hosted service with a WooCommerce plugin or integration. Examples include Octocom, whose WooCommerce extension is free while the service itself is a subscription, and helpdesks with built-in AI agents such as Intercom Fin. You pay a subscription, per-seat fees or per-resolution fees.
3. Custom assistant. Built for your store, usually using a model API, your catalogue and policies as a knowledge source, and WooCommerce data through the REST API or the newer Abilities API. You pay for development once, then API usage, hosting and maintenance.

How do they compare?

What does an AI chatbot really cost to run?

The running cost of a custom or bring-your-own-key chatbot is mostly model usage, charged per million tokens. A token is roughly a short word or part of a word. Every message the bot answers sends the instructions, relevant product or policy data and the conversation so far (input tokens), and gets a reply back (output tokens).

Here are current list prices (standard processing, short context) from the providers' pricing pages in September 2026:

OpenAI also notes that its regional data residency endpoints carry a 10% uplift for newer models, which matters if you choose EU processing.

A worked example

Assume a typical support conversation has 6 customer messages. Each reply sends about 4,000 input tokens (instructions, a few relevant products or policy snippets, and the chat history) and returns about 300 output tokens. That is 24,000 input and 1,800 output tokens per conversation.

These are our estimates from list prices, not quotes. Your numbers depend on how much context you send, how long conversations run, and whether you use prompt caching. In our projects we often see the context size, not the model choice, drive the bill. Sending the whole catalogue with every message is the most common and most expensive mistake.

Compare that with per-resolution pricing

SaaS helpdesk agents often charge per successful outcome. Intercom's pricing page lists Fin AI Agent from $0.99 per outcome, plus helpdesk seats (Essential from $19 per seat per month billed annually). If Fin resolved 3,000 of your 5,000 monthly conversations, that is about $2,970 a month in outcome fees before seats.

That is not a reason to avoid SaaS. You get a helpdesk, reporting, handoff and a team that maintains the AI. But the gap between about $330 in model costs and about $3,000 in outcome fees is where a custom build starts to make financial sense, once you add hosting, monitoring and maintenance to the custom side.

What costs do people forget?

• Maintenance. Plugins and SaaS vendors update for you. A custom bot needs someone to update prompts, test after WooCommerce updates and fix integrations. WooCommerce's MCP and abilities are still a developer preview, and the older endpoint is deprecated, so budget for small changes.
• Content work. Any chatbot is only as good as your product data and policies. Missing attributes and vague return rules produce vague answers, whichever option you choose.
• Testing. Before launch, you need a set of real customer questions with correct answers, and a way to rerun it after changes.
• Human handoff. Someone still answers the hard cases. Make sure the tool passes the conversation and context to your team.
• Compliance. EU visitors must be told they are talking to AI. Article 50 of the AI Act has applied since 2 August 2026. You also need processor agreements under GDPR with every vendor that sees customer data.

When does a custom chatbot pay off?

A custom build is worth considering when two or more of these are true:

• Your answers depend on data a plugin cannot see. Customer-specific B2B prices, stock from an ERP, delivery dates from a warehouse system, or compatibility rules for spare parts.
• Your volume makes per-resolution fees expensive. Use the example above with your own numbers.
• You sell in several languages and need consistent answers across them from one catalogue.
• You need control over data. A specific model provider, EU processing, short log retention, no training on your data.
• The bot should take actions, such as building a cart, starting a return or checking an order, with clear permission limits. WooCommerce's experimental commerce agent reference shows this pattern, with cart building for shoppers and human approval for merchant-side changes. It is reference code, not a supported product.

If none of these apply, start with a plugin or SaaS. You can move later, and your cleaned-up product data and FAQ content carry over.

How to evaluate a WooCommerce AI chatbot

Use this checklist for any option, plugin, SaaS or custom:

1. Grounding. Does it answer from your live catalogue, prices and stock, or from a copy that goes stale? How often does it sync?
2. Refusals. What does it do when it does not know? It should say so and offer a human, not guess a price or a policy.
3. Order data. Can it look up an order safely? How does it verify the customer before showing order details?
4. Handoff. Does the conversation reach your team with full context?
5. Test it. Run 30-50 real questions from your inbox, including awkward ones. Count wrong answers, not just good ones.
6. Cost model. Per seat, per conversation, per resolution or per token? Model it at today's volume and at double.
7. Data. Where is data processed and stored? Is there a data processing agreement? How long are transcripts kept?
8. AI disclosure. Does it clearly say it is AI in the first message, in every language you sell in?
9. Accessibility. Can the widget be used with a keyboard and a screen reader?
10. Performance. How much JavaScript does the widget add? Test mobile PageSpeed before and after.
11. Exit. Can you export conversations and knowledge content if you switch?

What does a custom build cost with us?

For clarity, here is how we price this work. An AI readiness audit, which checks your data, current tools and costs and gives a build-or-buy recommendation, starts from $1,500. A single-purpose custom assistant starts from $5,000. Monitoring and ongoing tuning starts from $300 per month, with model usage billed at the provider's cost. We fix the scope and price before work starts.

When to get help

If you are unsure whether a plugin will be enough, test one first with the checklist above. If it cannot reach the data your customers ask about, or the fees grow faster than your sales, our AI shopping assistant service covers the assessment and the build. For an overview of what else AI can do for your store, see AI for WooCommerce.