Build vs buy AI agents for ecommerce operations: decide per job

This guide is for CTOs, founders and ecommerce managers deciding how to put AI agents into order handling, catalogue maintenance and customer support. OutsourcingVN is operated by Netbase JSC, which builds custom agents and integrations, so it has an interest when the answer is "build"; the criteria below are written to give the "buy" answer whenever it fits. It is a spoke of the AI agent development services guide and the AI workflow automation buyer's guide.

Contents

• Why decide per job rather than per platform?
• Which ecommerce agent jobs favour buying, and which favour building?
• Which criteria tip the decision?
• What does a hybrid route look like?
• What data plumbing does a custom ecommerce agent need?
• Which Netbase work is closest to a custom commerce agent?
• Which questions should you ask before you buy or build?
• How this guide is sourced
• Common questions
• Choose per job

Why decide per job rather than per platform?

Most commerce teams meet AI first as a feature inside a tool they already pay for: a writing assistant in the store admin, a reply suggester in the helpdesk, a forecast in the inventory app. Those features are a reasonable default. The question is not "build or buy AI" across the business; it is whether each operational job can be done well by an assistant that only sees one system.

The general build-versus-subscribe logic is set out in the custom build versus SaaS guide: subscribe when the capability is common and your process can adapt, build when it differentiates you or the integration makes a subscription unworkable. Agents add one more variable. An agent acts, so the decision also turns on what it is allowed to change and who controls that permission. OWASP's LLM06 entry on excessive agency advises limiting the permissions an agent's extensions hold to the minimum necessary; with a packaged assistant, the vendor decides how fine-grained those permissions can be.

Which ecommerce agent jobs favour buying, and which favour building?

The table compares common operations jobs. "Buy" means an assistant packaged in the platform or an add-on; "build" means a custom agent connected to your systems.

The pattern is consistent: a single-system job with standard logic favours buying, and a cross-system job with your own rules favours building. For order flows that already cross the store and an ERP, e-commerce and ERP integration is often the first step, because an agent can only act on data that is in sync.

Which criteria tip the decision?

Score each row honestly. If most rows point to buying, buy, run the packaged assistant against a sample of your own cases and keep the rules in AI workflow evaluation and testing in mind for when it changes. If most point to building, the guide to scoping an AI agent project is the next step.

What does a hybrid route look like?

Many stores end up with both. The packaged assistant handles single-system tasks inside its own tool, and a custom agent handles the cross-system jobs, reading from the same store through its API. Anthropic's "Building effective agents" recommends finding the simplest solution possible and only increasing complexity when needed; for commerce, that usually means using the packaged feature until a specific job proves it cannot be done there.

The hybrid route needs two disciplines. Each agent's permission map must be written down so two tools do not both act on the same order, and the tool permissions guide gives a template. And a person must stay on the actions that cost money or reach customers, as human-in-the-loop AI workflows describes.

What data plumbing does a custom ecommerce agent need?

A custom agent reacts to events and reads current data. Shopify's developer documentation describes webhooks as a way for an app to subscribe to topics and receive near-real-time data about events on a shop, with examples such as informing shipping companies about changes in orders, returns or refunds. The same page warns that an app should not rely on receiving data from webhooks, because delivery is not always guaranteed, and recommends reconciliation jobs that periodically fetch data so the app stays consistent.

That advice applies to agents on any platform: trigger on events, but check the current record before acting, and run reconciliation so a missed event does not leave a customer waiting. Netbase's ERP consultancy and development experience covers Odoo consultancy and customization, cloud ERP platform development, e-commerce integration with ERP by API synchronisation of product catalogues, inventory, orders and customer data, a Vtiger CRM to ERP upgrade covering customer experience management, sales orders, workforce administration, support tickets, product management, resource allocation and budget planning, and Redmine to ERP development; modules built include finance and accounting, inventory, sales orders, HR and recruiting, CRM, project planning, e-commerce integration and reporting.

Store-side builds sit under e-commerce development, and support automation that combines a packaged helpdesk with custom steps is covered by customer support automation.

Which Netbase work is closest to a custom commerce agent?

For a client that is not named, Netbase built a WhatsApp Business AI chatbot with intent and conversation-flow handling, an LLM API (GPT-4 in the stack) and CRM synchronisation for lead capture, customer data and workflow automation. The WhatsApp chatbot record is the published example closest to a custom assistant that writes to a business system.

When the answer is to build, ownership matters. For custom development the client owns the IP created for it; Netbase productized modules and products are licensed, not transferred.

Which questions should you ask before you buy or build?

• Can the packaged assistant run on a sample of our real cases? If not before purchase, ask for a trial period long enough to score it.
• Which actions can it take, and can we restrict them per role? A tool that can refund but cannot be limited is a risk.
• Where is our data processed, and can we export the configuration? This decides the exit cost.
• For a build: which systems will the agent read and change? Each one needs an account, a permission and a test.
• For a build: what evaluation set and thresholds will accept it? Without them, acceptance becomes an opinion.

How this guide is sourced

The webhook behaviour and reconciliation advice come from Shopify's developer documentation; the permission advice from OWASP LLM06 (2025 edition); the simplicity advice from Anthropic's "Building effective agents". Netbase statements are approved claim-register entries, reviewed as the methodology describes. The job table reflects common commerce operations and names no product's feature list.

Choose per job

Write down the job, the systems it touches and the actions it needs, then score it against the criteria table. AI Workflow Automation is the service that builds custom agents with human control when the score points to building.

OutsourcingVN is Netbase's own outsourcing-services platform. Submit a project with the job you want to automate and the tools you already use, and a person will reply with whether it is a build, a buy or both.

Related services and solutions

Sources

• Shopify, About webhooks
• OWASP GenAI Security Project, LLM06:2025 Excessive Agency
• Anthropic, Building effective agents