FlowsBook a callSign inCreate free accountSign up
AI employees and automation

Letting an AI employee email a client directly: a safe way to start

Before a client gets reports straight from an AI employee: the sender they see, your brand, who can be emailed, a trial with your OK first and the checks.

Part of the guide AI employees

By the xMap team · · 5 min read

You can let an AI employee email a client directly once three things are in place: the email looks like it comes from you (your company name, ideally your logo and your own sending domain), the content has been checked by a person for a few runs, and the employee is set to stop if its data looks wrong. Start with the employee asking for your approval before every send, read the first three or four emails it drafts, then switch it to acting on its own and telling you what it sent. Most teams reach that point within a month.

This article is for agencies, consultants and account managers who send clients regular updates, such as a monthly market note, a rating watch or a list of new competitors, and want an AI employee to do the sending. It covers what the client actually sees, the settings that matter, a made-up rollout, mistakes that embarrass you in front of a client, and the exact blocks to use.

What the client sees

An email from an AI employee is sent from xMap's address under your company's name, shown as "Your Company via xMap Flows", and replies go to the person who owns the flow. That is fine for an internal team, but for a client you will usually want more:

  • Your logo and colour at the top of the email, set once under Settings for the whole company.
  • Your own domain as the sender, such as reports@youragency.com, after you add three DNS records that prove the domain is yours.
  • An unsubscribe link, which every email a flow sends carries. A client who clicks it stops getting that company's flow emails, so tell them what the emails are before the first one arrives.
  • A clear note when figures are examples: if any step ran on sample data, the email says so. Make sure a client email never goes out with that note; fix the setup first.

Three levels of trust

Each AI employee has one setting for how much it does alone, in its panel and in the window you see before it goes live:

  • Ask me first. An Approval step sits before every send. You get a notification, read the email, change it if needed, and approve or decline.
  • Act and tell me. It sends on its own, then leaves a note in the Team inbox saying what went out.
  • Fully on its own. It sends without notes. Keep this for routine internal emails, not for clients.

For client emails, start on Ask me first and move to Act and tell me once the drafts stop needing changes.

A worked example

Here is a made-up rollout. A small marketing agency in Sao Paulo looks after six restaurant clients and wants each to receive, every month, a short note on how the restaurants around them are rated and which new ones opened nearby.

  1. Week one: the account manager builds the flow on the agency's own address first, runs it once and reads the email as if she were the client. She rewrites the AI's instructions twice to drop jargon and to keep it under 200 words.
  2. Week two: she adds the agency's logo and its own sending domain, and switches the employee to Ask me first. The first real client drafts arrive as approvals; she edits one sentence in two of the six.
  3. Month two: none of the six drafts needs a change. She switches to Act and tell me. Each month the Team inbox shows what was sent to whom.
  4. One client replies with a question. The reply lands in her own inbox, because she owns the flow, and she answers it herself.

Mistakes to avoid

  • Going straight to sending on its own. The first drafts always need a person's eye: tone, length, a name spelled the way the client spells it.
  • Letting the AI invent figures. Tell the xMap LLM block to use only the figures from the earlier blocks, and check the first few emails against the run's data.
  • One flow per client with no plan. Six copies drift apart. Use one flow with variables, or linked copies, so a change reaches every client at once.
  • No check for an empty run. If a search finds nothing, the client should not get an empty email. Add a rule that stops the run or emails you instead.
  • Sending from xMap's address for long. A client trusts mail from your domain more; set it up before the first client email if you can.

The blocks, in order

  1. Schedule, for example the first working day of the month at 08:00 in the client's time zone.
  2. Area of interest around the client's address.
  3. Nearby places for the kinds of place the client competes with.
  4. xMap LLM on the whole list, with instructions written for the client and told to use only the figures given.
  5. Check the work, with rules such as "no figure that is not in the data" and "under 200 words".
  6. Approval, added for you when the employee is set to Ask me first.
  7. Email to the client's address, or Make a report to attach a branded PDF.

What lands in your inbox changes with the level: approval requests to read and send at first, then short notes of what was sent. The templates Rating watch for my clients and Reputation programme for my clients are built for agencies and are a quick place to begin.

Questions

Will the client see that an AI employee sent the email?

The sender shows your company's name with "via xMap Flows" unless you send from your own verified domain. You decide how to describe the service to the client.

Can I check every email before it goes to the client?

Yes. Set the AI employee to Ask me first and each email waits for your approval, which you can edit before it goes.

Where do client replies go?

To the person who owns the flow, so a real person answers questions.

Can I email a client on the Free plan?

On the Free plan, flows email only your own team and addresses at your team's email domains. Emailing a client at another company needs a paid plan.

Templates that do this

Read next

More on ai employees and automation · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.