Does InboxPilot reply to customers on its own?

Not until you tell it to. Every draft waits in Gmail or Outlook for one-click approval. Auto-send is switched on per request type, on the Growth plan and above, once you have watched that category get it right. Refunds, complaints and anything legal can stay on review for good.

Where do the answers come from?

From what you connect: your policies and FAQs, your website, past replies, uploaded documents and tools such as Shopify, Notion, Zendesk and Pipedrive. If the answer is not in any of them, the draft does not invent one; the email goes to a person.

Do we have to move out of Gmail or Outlook?

No. InboxPilot works inside the inbox and the shared addresses you already have. Teams that run a helpdesk for chat or phone keep it and automate the email channel, which is where the volume lands.

Which customer service emails should stay with a person?

Complaints, refund requests above whatever threshold you set, legal or safety issues, and anything from an angry customer. Routing rules and plain-English instructions send those to a named teammate with the thread summarised, and they arrive faster because routine mail no longer buries them.

How long does it take to set up?

About a minute to connect an inbox, and the first drafts usually appear within a few minutes. Connecting your policies and store data takes longer and is what turns generic drafts into specific ones, so plan an afternoon for that.

Pricing is by conversations handled, not per seat, with inbound email unlimited. The Free plan covers 25 conversations a month on one inbox, Starter is $39 a month for 300, and Growth is $99 a month for 1,500 with auto-send and unlimited workflows. Current plans are on the pricing page.

AI Customer Service Email Automation

Why customer service emails deserve automation

Customer service email is the mail a business cannot stop: where is my order, can I change the address, what is your return policy, the invoice is wrong, the part arrived cracked. Most of it repeats, and almost all of it has an answer that already exists somewhere in the company, in the order system, in a policy document, or in a reply a colleague sent last week. The cost is not any one message. It is the lookup before each one, the tab switching, and the twenty minutes a day that turn into a hire. AI customer service email automation, done properly, is not a bot that answers everything. It is a first draft for every message, written from the company's own material rather than from what a general model assumes, with a person deciding what leaves the inbox. InboxPilot does exactly that inside the Gmail or Outlook inbox a team already runs: it sorts each incoming email by request type, drafts a specific reply from the connected sources, sends the routine categories on their own only once you have turned that on, and hands anything sensitive to a named person with the context summarised.

What makes this workflow painful

The same twenty questions

Order status, returns, invoices, delivery changes and product questions make up most of a service inbox, and each one gets answered by hand as if it were new.

The answer is in another system

Replying means opening the order, checking the policy, finding the last thread. The typing takes thirty seconds; the lookup takes five minutes.

Slow replies become complaints

A question left overnight turns into a second email, a chargeback threat or a public review. Speed decides how the conversation ends.

How teams handle it today

None of these are wrong. They just spend skilled people on assembly work, and they all get slower exactly when volume spikes.

Answer everything by hand

One or two people work through the queue, copying tracking links and policy lines into fresh replies, reaching for a saved template on the common ones.

Skilled time spent on lookups, and a template with no specifics earns a second email asking about this particular order.

Move the address into a helpdesk

Route the shared inbox into a ticketing tool with macros, tags and assignment, and train the team on it.

A migration and a new interface for a problem that is mostly writing. The macros still cannot read the question.

Put a chatbot on the website

A chat widget answers the common questions on the site, so fewer people write in.

It only reaches people who came to the site. The email still arrives, often as a reply to an order confirmation.

How InboxPilot automates it

Step 1

Connect the inbox the questions land in

support@ or hello@ in Gmail or Outlook, shared addresses included. One click, scoped access, nothing to migrate.

Step 2

Give it what your team answers from

Your policies and FAQs, your website, past replies, and the systems the answers live in: Shopify or your order data, Notion, Zendesk, Pipedrive, a price list.

Step 3

Every email is sorted and drafted

Each message is labelled by request type as it arrives and a specific reply is drafted from your material. Newsletters and notifications are skipped.

Step 4

Approve, or let the routine ones send

Drafts wait in the inbox for one-click approval. When a category is consistently right, let it send on its own and keep review for refunds, complaints and anything legal.

A real example, before and after

The incoming email on the left, the InboxPilot draft waiting for approval on the right.

Customer, four days after deliveryIncoming email

Wrong size, can I exchange?

Hi, I ordered the medium (order #7731) but it runs small. Can I swap it for a large, and do I have to pay for shipping again? I need it by the 14th.

AI draft by InboxPilotWaiting for your approval

Hi, yes, we can swap the medium for a large on order #7731 at no charge. Exchanges are free within 30 days, and the return label is attached to this email.

Once the carrier scans the return, the large ships the same day, so it should reach you before the 14th. If the scan has not shown up by Thursday, reply here and we will send the large ahead anyway.

The draft takes the exchange window and the free-label rule from your returns policy, the order details from your store, and commits only to what the policy allows.

Where automated customer service email goes wrong

The failure is rarely a rude reply. It is a confident one that promises a refund the policy does not allow, or answers the wrong order. Grounding and a review step prevent both.

Confident and wrong

A general model produces a fluent answer to any question, including ones your policy does not cover. A draft written only from your own material cannot invent a refund window you do not have.

The wrong order

Customers write from a different address, quote the wrong number or reply to an old thread. The draft states which order it found, so the person approving can catch a mismatch at a glance.

Angry customers on autopilot

The last thing a furious customer should get is an instant machine reply. Routing rules on topic and urgency, plus a plain instruction such as 'if the customer is upset, hand the thread over', keep complaints, large refunds and anything legal on a person's desk.

Turn a sentence into automated actions

Describe what should happen in plain English and InboxPilot builds the workflow: a trigger, an AI decision, and the actions that run on their own.

Describe your workflowPlain English

“When a customer asks to exchange or return an item, check the order and the returns policy, draft the reply with the label, and label the thread Returns.”

Build workflow

InboxPilot builds

TriggerIncoming email recognised as a return or exchange request

AI DecisionAI matches the message to the order and checks the returns policy

Actions

Draft a reply with the exchange terms and the return label
Apply the Returns label for reporting
Escalate to a person when the order is outside the window or the customer is upset

Related guides and pages

Order status and WISMO email automation

AI support for e-commerce and retail

Shopify customer support automation

Customer service: frequently asked questions

Software that reads each customer email as it arrives, works out what it is asking, and drafts a reply from the company's own information: order data, policies, product details and past replies. The useful kind stops before sending so a person can approve, and hands anything it should not answer to a teammate.

Customer service email, answered from your orders, policies and past replies