Office & managementNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.

AI answers customers' 'where is my order?' emails from live ERP data (promotional-products maker, medical device maker)

Added 2026-10-19 · Last updated 2026-10-19

Role

Office (customer service and inside sales: order-status inquiries) · Typical users: customer service representative, inside sales representative, order desk, customer service manager

The problem

A large share of the customer service inbox at a manufacturer is routine: what is the status of my order, when will it ship, what is your returns policy. Each one means reading the email, looking up the order in the ERP and typing a reply, while complex requests and quotes wait. When volume rises, response times slip, satisfaction drops, and sales staff get pulled in to help answer the queue instead of selling.

How AI helps

The AI reads each incoming email and first classifies it: needs no reply (acknowledgements, notifications, marketing), routine and in scope (order status, shipping dates, policy questions), or out of scope (custom orders, complaints, escalations), which goes to a person. For routine questions, it looks up the order in the ERP and drafts the reply from that data. A quality check scores each draft; depending on confidence, the AI either replies directly, places a draft in the representative's inbox for review, or hands the email to a person without drafting.

Step by step

Sample one or two weeks of the customer service inbox and sort it: no reply needed, order status, shipping dates, policy questions, and everything else; count each and note the time per reply.
Pick the routine categories to automate first and write the answers you allow: which order fields may be quoted, the policy texts, and what always goes to a person (complaints, price, custom requests, credit holds).
Give the AI read-only access to the ERP order, shipment and tracking data it needs, nothing more.
Run in draft mode first, in parallel with the team: the AI drafts, the representative edits and sends, and every edit is logged.
Let the AI reply directly only on categories where the drafts were accepted almost unchanged, with a confidence threshold and a hand-off to a person below it.
Track reply time, share handled, customer follow-ups and satisfaction against the baseline; review a sample of sent replies every week.

Tools

Custom build on OpenAI, n8n, NetSuite and Microsoft Outlook (HatchWorks AI): used at the promotional-products manufacturer
Quiq email AI agent integrated with Oracle: used at the medical device manufacturer
Other tools to evaluate: the AI features of your help desk or CRM, Microsoft Copilot in Outlook, or a workflow tool (n8n, Power Automate) with read-only ERP access
Any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below, on a few real emails and an ERP export of the matching orders, for a first test

Result

Promotional-products manufacturer and wholesaler (anonymous, Los Angeles; tens of thousands of customer emails a month, seven-person inside sales team), built by HatchWorks AI on OpenAI, n8n, NetSuite and Outlook: 47% of the emails that needed a reply are now drafted automatically and pass the quality gate; about 50% of incoming emails turned out to need no reply at all; up to 70% of responses to routine inquiries are send-ready (HatchWorks case study). Medical device manufacturer (anonymous), Quiq email AI agent on its Oracle environment, run silently alongside human agents before go-live: more than 85% of the emails the agent answers directly are fully resolved with no follow-up from the customer, response-time SLAs were exceeded, and sales staff pulled into support went back to selling (Quiq customer page). Both are anonymous vendor case studies.

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

Start in draft mode: a person reviews and sends until the drafts are reliably right; only then allow direct replies on narrow, routine categories.
The AI must quote dates and status only from the ERP; a wrong promised date damages the customer relationship more than a slow reply.
Keep complaints, price questions, credit holds and anything unclear with a person.
Give the AI read-only access to the ERP and check what customer data leaves your systems; use an approved enterprise tool.
Both cases are anonymous vendor case studies.

Try it yourself

You are an experienced customer service representative at a manufacturer.

I will paste: (1) a batch of customer emails [sender company, subject, body]; (2) an export of the matching orders from our ERP [customer, PO number, our order number, items, quantities, promised date, shipped date, carrier and tracking number, status]; (3) our policies for shipping, returns and lead times [text].

Do this for each email:
1. Classify it: No reply needed / Order status / Shipping date / Policy question / Needs a person (complaint, price, custom request, anything unclear), with one line of reasoning.
2. Match it to the order using the PO or order number in the email; if no match or several matches, say so.
3. For routine emails, draft a short, polite reply that quotes only the fields in the export (status, dates, tracking) and the policy text as written.
4. For 'Needs a person', write a two-line summary for the colleague who will handle it.
5. End with a table: email, category, matched order, draft ready yes/no, and what is missing.

Never promise a date, price, credit or exception that is not in the data I paste.
Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Want more like this? Join the newsletter

Related cases

All 8 Office & management use cases →