AI sales agent: Draft the pitch, let a person send

An AI sales agent can research a prospect, find relevant details about their business, draft a first email and remind you when it is time to follow up. It can do most of the preparation that makes sales outreach slow. The first message, however, should wait for a person to approve it. Contacting someone who has never spoken to your business creates a relationship before the recipient has agreed to one, and a bad opening can be difficult to repair.

The useful division of work is straightforward. Let the agent gather evidence, prepare the message and organise the next steps. Keep the decision to contact someone with the person who owns the relationship. This gives a sales team time back without allowing an automated system to decide, on its own, who should hear from the business and what it should say.

What an AI sales agent should do before the first email

A good first message usually takes longer to prepare than to read. Someone has to work out whether the prospect fits the business, understand what the company does and find a credible reason to make contact. Then comes the writing, followed by the checks that stop a plausible email from becoming an embarrassing one.

An AI sales agent can take on much of that preparation. Give it a clear brief describing the kind of organisation you want to reach, the problem your offer addresses and the evidence it should look for. The agent can research a prospect's public business information, collect relevant details and prepare a short account of why the person might be worth contacting.

That account should separate facts from assumptions. A company announcing a new office is a verifiable detail. The claim that it must therefore be struggling with a particular operational problem is an inference. The agent can suggest that inference as a possible reason for outreach, but the draft should not present it as something the company has confirmed.

The next task is to turn the research into a draft. A useful opening might refer to a specific business development and explain why it relates to the sender's work. The agent can propose a subject line, write the message and suggest a reasonable next step. A person can then judge whether the connection is persuasive or whether the email sounds like an excuse to pitch a service.

For teams setting up this workflow, the Delagents outreach writer template provides a starting point for delegating outreach preparation. The brief still needs to define the audience, the information the agent may use and the kind of message the business is willing to send.

Why the first contact needs a human decision

A first email is a small action with consequences outside the business. It reaches a real person, represents the sender and makes a claim about why a conversation might be useful. Even a well-written message can be wrong for its recipient. The research might be outdated, the apparent business need might not exist, or the proposed offer might be inappropriate for that organisation.

There is also a difference between permission to prepare outreach and permission to send it. A brief asking an agent to find ten suitable prospects does not grant authority to contact all ten. Nor does approval of one message establish that every later message is acceptable. The recipient, the context and the claims in the draft can change.

A human review should therefore be specific to the proposed action. The reviewer needs to see who will receive the message, what it says and which facts support the opening. They should be able to correct the wording or reject the contact, rather than choosing between a vague approval and a complete restart.

Consider an agent that finds a company's recent hiring announcement and drafts an email offering recruitment support. The message may read well, but the owner knows the company is an existing partner, not a prospect. That one detail changes the decision. Without a checkpoint, the agent can turn an otherwise useful piece of research into an unwanted approach.

For this reason, first contact with anyone outside the workspace should always require human approval. It is a firm boundary, not a setting to relax because the draft looks convincing or the sales team is busy.

The approval should show the actual message

An approval process only helps if the reviewer has enough information to make a decision. A notification saying that an agent is ready to send outreach leaves too much work to the person who must sign off. They need to inspect the action itself, not just its category.

For a first-contact email, the approval card should make the recipient and exact proposed message clear. The reviewer should be able to check the greeting, the reason for contact, any factual claims and the requested next step. If the message needs a change, that change should happen before sending. If the contact is unsuitable, rejecting the action should stop it.

The same principle applies to the work behind the message. A person should be able to see how much of the delegation budget has been used and when the approval expires. Those details help the reviewer decide whether to approve now, request a revision or stop the task. An agent should not gain permission to send simply because nobody has responded to its request.

In Delagents, actions that affect anything outside the workspace wait in an approvals inbox. Each action appears as a card showing what will happen, the cost so far and a deadline. A person can approve, change or reject it. The agent can prepare the work, but the approval remains a separate decision.

This distinction matters when a team handles several prospects at once. Reviewers can make decisions based on the proposed messages instead of relying on a general instruction that outreach is allowed. If a draft contains an unsupported claim or a tone that does not fit the relationship, the problem is visible before the recipient sees it.

Follow-up is where the saved time continues

The first email is only one part of the sales process. Someone must also remember when to follow up, check whether a reply has arrived and decide what should happen next. These small administrative tasks compete with calls, proposals and conversations that need direct attention.

An agent can prepare a follow-up schedule based on the team's instructions and create reminders for the owner. It can organise the prospect's research alongside the draft, record the planned next step and flag a message that needs attention. If the prospect replies, the agent can prepare a summary or draft a response for review.

A reminder is different from a licence to send another message. If the prospect has asked not to be contacted, a follow-up should not proceed just because a date has arrived. If the conversation has changed direction, the original schedule may no longer make sense. The agent can bring the task back to the owner, who can decide whether to continue, pause or close it.

The team should also distinguish routine preparation from actions that create new external commitments. A follow-up draft can be prepared in advance. A first contact must wait for approval, and any other action covered by the team's approval rules must stop at its own checkpoint. The point is to remove the need to remember every administrative detail, not to remove judgement from the process.

Give the agent a brief it can actually follow

Sales automation tends to become unreliable when the brief describes a desired outcome but leaves the working rules unstated. Asking an agent to find leads and start conversations does not tell it what counts as a good prospect, which claims it may make or where its authority ends.

A useful starting brief should cover the following points:

• Define the target prospect by business type, relevant need and any conditions that make an organisation unsuitable.
• Specify which sources the agent may use and require it to distinguish verified details from assumptions.
• Describe the offer accurately, including claims the agent must not make without evidence.
• Set the desired tone, a reasonable message length and the next step the draft should propose.
• Require approval before first contact and state what the agent should do when information is missing or a prospect's status is unclear.
• Set a hard spending limit for the delegation and retain a record of the work, tool calls and approval decisions.

The budget limit matters because research can expand. An agent might begin with five named prospects, discover several related companies and keep collecting information long after the original task has become useful. A hard cap places a boundary on that work. An append-only activity log gives the owner a record of tool calls, approvals and policy changes to inspect later.

Before delegating, test the brief on a single prospect. Read the research, inspect the draft and check whether the agent has made assumptions that sound like facts. Ask for a revision where needed. The purpose of this first run is to find gaps in the instructions before the agent handles a larger list.

Start with one prospect and one approval

A practical first use for an AI sales agent is a single prospect your team already has a reason to investigate. Ask the agent to gather the relevant business facts, explain why the prospect may fit, draft an opening email and prepare a follow-up reminder. Do not ask it to send anything.

Review the result as if you were about to put your own name on the message. Check the evidence, the relevance of the offer and the tone. If you would not send the draft yourself, identify the missing instruction and revise the brief. Then run the same task again and compare the result.

Once the preparation is dependable, use an approval checkpoint for the actual first contact. Keep the recipient, message and decision visible, and retain the right to change or reject the proposed action. Start with one prospect today, and measure the work the agent removes before expanding the delegation to a larger list.