Using AI to answer
your customers’ emails.

Why a reply written from the email alone goes wrong

An email from a customer carries the question, and rarely the facts needed to answer it. To reply to “where is my order?”, someone on your team looks up the order, checks the date that was promised and knows how you handle a late delivery. To reply to a request for a better price, they know what this customer already pays, what you allow and who can agree to an exception. The AI has none of that unless it is given to it, so it fills the gaps with a plausible answer written in general terms.

That is why the drafts need so much checking. The person reviewing one has to find the facts, correct them and adjust the tone, which often takes as long as writing the reply. Most of the tools you will find when you search for this connect the AI to your inbox and, sometimes, to your customer records. That gives it the conversation, and it still leaves out your rules, your exceptions and the way you word things, which usually sit in people’s heads and in old emails.

Write down what your replies depend on

The first piece of work is to write down, once, what your team knows when they answer a customer: the main types of email you receive, what each reply needs to check, your rules on prices, refunds and delivery, who can approve an exception, and the phrases you use or avoid. This fits on a short page in plain words that your team can read and correct. When a rule changes, someone edits the page, and every draft after that follows the new rule.

The way you write comes from your own sent emails. A reply to a long-standing customer does not read like a reply to a supplier or to someone who has just enquired, so the examples are grouped by the kind of person you are writing to, and the AI drafts from the group that matches. When you change a draft before sending it, the difference is compared with what the AI wrote, and a correction you make more than once is added to the written rules, so the same mistake does not keep coming back.

This is one part of a wider structure: your business knowledge written down once and read at the start of each task, which we describe in what an AI adoption consultant does.

How a draft is produced

When an email arrives, the AI reads the whole conversation before it writes anything, including what was said in earlier messages, and checks whether the question has already been answered in another conversation with the same person. It then drafts the reply in your voice from your written rules and your examples, and leaves it in your drafts for a person to read, correct and send.

The draft is never presented as final. Where the AI is unsure what the customer is asking, or which rule applies, it says so and puts the email in a separate list with its best guess and the reason, so the decision stays with the person who knows the customer.

What the AI may do with your email

An assistant that answers your customers works inside your mailbox, so what it is allowed to do there is settled before it is switched on. In the email assistant we install, the AI prepares drafts and nothing else: it does not send an email, delete one or follow a link inside one. Sending stays with a person because a reply is a commitment to a customer, and a draft that nobody has read should never become a message the customer receives.

When an email contains a link, the AI shows it to you and does not open it. A payment request or a sign-in link from a sender you do not know is flagged as possible phishing instead of being followed. We explain how access and approvals are set for the rest of the business in adopting AI securely and on our security page.

Which emails to start with

Start with one type of email that arrives often and is answered in a similar way each time, such as questions about orders and deliveries or requests for documents. The rules for it are short to write down, the drafts are easy to check, and each correction your team makes shows where the written rules are incomplete.

Some replies are better kept with a person: a complaint from a customer who is upset, a negotiation on price, anything with legal or financial consequences. The AI can still gather the history and the figures for the person writing the reply, which saves the looking up and leaves the judgement with them. Deciding which emails stay manual is part of the setup, and the list can change as the drafts prove reliable.

Your team also needs to know how to read a draft critically and correct it well, which we cover in AI adoption training for your team.

How we start

The first step with us is a free AI pre-audit: 45 minutes on a call where we look at how you and your team use AI today and which work is worth handing over first. Bring the emails that take your team the most time. You leave with a clear next step, with or without us.

If you decide to go further, we set this up as part of the Sprint, alongside the rest of your business knowledge and the access for each role. See how we help and what we have built for other businesses, including the email assistant described here. If your first question is where your emails and customer data would be stored, we compare where the main automation tools keep data, for a business in the UK or in Australia, in Zapier, Make or n8n, and cover the Australian case in detail in AI tools and your business data in Australia. In French, the same question is answered for businesses in France, Belgium and Switzerland in Répondre aux emails clients avec l’IA.