Skip to content
unmatrixai

Discovery, support, and conversion

AI agents for e-commerce: discovery, support and conversion

AI agents for e-commerce help shoppers find the right product, resolve orders and returns without a ticket, and keep catalogue and merchandising data clean. Unmatrix AI builds them on your commerce platform, order management and support systems, with guardrails that stop an agent promising what your policies do not allow.

Last updated

Workflows

What do we build for e-commerce?

The workflows we usually start with, and the architecture each one is built on.

  • Product discovery and guided selling

    Answer shopper questions from catalogue, specifications and reviews with citations, and recommend within stock and margin rules.

    Built as

    RAG agents

  • Order and returns support

    Look up the order, apply the policy, and issue or draft the resolution: refund, replacement or return label.

    Built as

    LangGraph workflow agents

  • Catalogue operations

    Normalise supplier feeds, generate and check attributes and copy, and flag inconsistencies for merchandisers.

    Built as

    Autonomous agents (ReAct)

  • Marketplace and seller support

    Triage seller cases, verify them against policy and resolve the routine ones.

    Built as

    LangGraph workflow agents

Constraints

What shapes the build?

Policy adherence
The agent can only offer what the policy engine allows. Refund and discount actions are gated by thresholds and, above them, by a person.
Brand and tone
Approved copy and tone guides, with evals run on real conversations.
Peak load
Cost and latency budgets per conversation, with model routing for peak, load-tested before the season.
Privacy
Customer data minimised in prompts, and regional residency respected.

Worked example

A typical build: returns resolution

A retailer's support team spends a third of its time on order status and returns, most of which the policy already answers.

  1. 01

    Ask

    A customer messages about returning an item.

  2. 02

    Identify

    The agent finds the order, the items and the delivery status in the order management system.

  3. 03

    Apply

    It applies the returns policy: window, condition, category exceptions.

  4. 04

    Resolve

    Within the threshold it issues the label or refund. Otherwise it drafts the response for a support agent with the policy reasoning.

  5. 05

    Record

    The ticket is closed with the reasons and the policy version used.

Routine returns resolve in the first message. Support agents handle the exceptions with the context already gathered.

Integration

Which systems do we work with?

  • Shopify
  • Salesforce Commerce Cloud
  • commercetools
  • Adobe Commerce
  • Manhattan Active
  • Zendesk
  • Gorgias
  • Klaviyo

Anything with an API or a file interface can be integrated. These are the systems we meet most often in e-commerce.

Questions

What e-commerce teams ask

Will the agent give unauthorised discounts or refunds?

No. Money-moving actions go through your policy engine and thresholds. Above them, a person approves.

Can it recommend products accurately?

It answers from your catalogue and inventory with citations, and evals check that it does not invent specifications or stock.

Does it integrate with Shopify?

Yes. Shopify, Salesforce Commerce Cloud, commercetools and Adobe Commerce, plus your order management system and help desk.

How does it handle peak season?

Budgets and routing per conversation, load-tested before peak, with a graceful hand-off to people when volume or confidence demands it.

Related

Next step

Pick one workflow. We will be at your office in two weeks.

A thirty-minute call to find the right first workflow, followed by a written scoping note. No deck, no pilot.

Book a call