If the CEO said “add AI to outbound this quarter”, decide four things before you connect an AI assistant to anything: who approves replies, what the AI may change without asking, which actions need a confirm step, and how the team reads results. The connection itself takes minutes. The four decisions are the project.
This checklist is for a RevOps lead at a B2B company with a small sales team. It uses the GTM Bud connector as the example, because the connector marks each tool as read or write and puts a confirmation in front of the changes that start something. The decisions apply to any tool that lets an AI act on outbound.
Why the decisions come before the connection
An AI assistant with a connector can do more than write text. It can read real replies, change a campaign and ask for a launch. The moment you connect it, your team has a new operator with no job description.
Most teams skip this step for a simple reason: the request from the CEO was about speed. So they connect first and write rules after the first mistake. The order below is the opposite, and it costs one working session.
The one page rule: if the four decisions do not fit on one page, they are not decisions yet. They are a discussion.
The checklist at a glance
| Decision | The question to answer | Where it lives in GTM Bud |
|---|---|---|
| 1. Who approves replies | One named reviewer, and a backup | Reply review before each send |
| 2. What the AI may change alone | Which write actions need no extra rule from you | Tools marked read or write |
| 3. What needs a confirm step | Which changes a person must confirm | Preview and confirmation before a change applies |
| 4. How to read results | Which numbers, how often, and how old they may be | Results per campaign with an as-of time |
Decision 1: Who approves replies?
One named person approves replies, and a second person covers their days off. Write both names down. The AI shows each reply it writes for review and is instructed to send only after approval, so decide who that reviewer is.
Three points to settle:
- The reviewer. The person signed in to the GTM Bud account reviews in their AI session. In a small team this is often a sales manager or RevOps for the first month.
- The standard. The reviewer does not judge from taste. They judge against the reply playbook. If a reply follows the playbook and still feels wrong, the playbook needs a new rule.
- The daily load. GTM Bud limits replies to 40 a day for each LinkedIn account. With three sender accounts, the upper bound is 120 reviews on a busy day. Plan the time.
The reply playbook is the written standard. GTM Bud keeps one for the account and one per campaign, and the AI follows them for each reply it writes. For the content of the rules, start from our guide on how to handle cold outreach replies.
Decision 2: What may the AI change without asking?
Sort every action into three groups: read, prepare and start. The AI may read anything and prepare drafts without asking. It may not start anything alone.
- Read. Campaign list, results, replies, conversations, playbooks, launch readiness, campaign controls. These change nothing. Let the AI use them freely.
- Prepare. A campaign draft, the targeting of a draft, a prospect preview, value offers, a message sequence, a duplicate of a campaign. A draft sends no messages. Let the AI do this work, and review the result.
- Start. Launch, a change that starts sending, a change that starts lead finding, ending a campaign. These wait for a person.
Two actions need a rule of your own, because they are small and easy to overlook:
- Reply labels. The AI can label a reply Interested, Warm or Not interested. Decide if the reviewer spot-checks the labels each week. A wrong label hides a good lead.
- Removing a lead. The AI can remove a lead from a campaign. Decide when that is allowed, for example only when the lead asks to stop.
The reply playbook is a special case. The AI can propose new wording. It shows the exact new text and saves it only after you approve it in the chat, and each save keeps the earlier text in the version history. Decide who gives that approval. It should be the same person each time.
Decision 3: What needs a confirm step?
Everything that starts contact with real people needs a person to confirm a preview. GTM Bud enforces this for four changes: launch, a change that starts sending, a change that starts lead finding, and ending a campaign. Your decision is who may confirm, and what they must check first. A reply is different: the AI shows it for review and is instructed to send only the approved text, with no second call, which is why Decision 1 names the reviewer.
A workable rule for a small team:
- Before anyone asks for launch, the AI runs the launch readiness check. The check lists what is missing.
- The list must be empty.
- The person who confirms reads the launch preview, which shows the first message of each channel, and reads the full sequence with the campaign sequence tool before confirming.
- The person who confirms is not the person who wrote the messages, if your team is large enough for that.
A change that stops sending applies at once, with no preview. Leave it that way. Anyone who sees a problem should be able to stop a campaign in one sentence.
What GTM Bud does not do here. It does not give you an approval chain with roles. The confirmation happens in the AI session of the person signed in. Point 4 above is your rule, not a product setting.
Decision 4: How will you read results?
Agree on which numbers the team quotes, on which day, and how old a number may be. GTM Bud returns results per campaign, and the numbers carry an as-of time, which is the time they were calculated.
- Which numbers. Pick the few that leadership will ask about: connection requests sent and accepted, replies, and positive replies. For the definitions that most teams use, see our guide to outbound KPIs.
- Which day. GTM Bud does not send a scheduled report. One person asks the AI for the results of each campaign on a fixed day and writes them into the document that leadership reads.
- How old. Write the as-of time next to each number. When two people quote different numbers, the as-of time shows why.
A worked example: the four decisions on one page
The company in this example sells fleet tracking software to construction firms. It has five reps, and the RevOps lead is the reviewer for the first month. The names below are invented.
The audience sentence. “Fleet managers and heads of operations at regional construction firms with 100 to 1,000 employees in the United Kingdom.”
What the criteria check said. The roles and the country were clear. The check flagged “regional” as a requirement that cannot be read from a profile or a company page, and suggested leaving it to the size range. The AI also asked if civil engineering contractors count as construction firms. The answers: yes, the size range is enough, and yes, include them.
Two previewed prospects.
- Callum D., Fleet Manager at a groundworks contractor with about 340 employees in Leeds. Marked Good fit: right role, right kind of company, inside the size range.
- Priya S., Operations Coordinator at a 12-person scaffolding firm in Kent. Marked Not a fit: too small, and the role does not buy software.
One reply draft. A lead wrote: “We already have trackers in the vans. Why would we change?” The playbook rule for an existing supplier: do not criticize it, ask one question about the gap. The AI wrote:
“No reason to change what works in the vans. Most of the questions we get are about plant and trailers, which van trackers often do not cover. Is that part of the fleet tracked today?”
The reviewer removed “most of the questions we get”, a claim she could not check, and approved the rest.
The results question. “Give me the results for the UK construction campaign with the as-of time, and list the Interested replies that have no answer yet.”
Five prompts for the first week
Use these with the AI in the first week, before any launch. All five are read or prepare actions.
- “List the tools you have from GTM Bud and tell me which ones are marked read and which ones are marked write.”
- “Show me the account reply playbook. Then list the situations it does not cover, based on the replies from the last two weeks.”
- “Label every unanswered reply Interested, Warm or Not interested, and show me the five you were least sure about.”
- “Create a campaign draft for fleet managers and heads of operations at construction firms with 100 to 1,000 employees in the United Kingdom, and stop after the criteria check.”
- “Run the launch readiness check on every draft and give me the missing items as one list.”
Three objections you will hear, and honest answers
“The CEO wants AI to send without a person. Review is the old way.” The AI can prepare everything. The AI still shows each reply for review and is instructed to send only after approval, and a launch still waits for a confirmation. If the goal is zero human time, this setup does not meet it, and you should say so in the first meeting, not in the third.
“One reviewer is a single point of failure.” Yes. That is why Decision 1 asks for a backup by name. GTM Bud does not route replies to people or offer roles in this setup, so the backup is a line in your process and a second person who knows the playbook.
“We cannot prove this works before we commit.” You cannot, and this article gives you no result numbers. What you can do is run the first week in read and prepare actions only, with nothing sent, and judge the drafts. That costs time and no reputation.
Frequently asked questions about adding AI to outbound
What should RevOps decide before connecting an AI assistant to outbound?
Four things: who approves replies, what the AI may change without asking, which actions need a confirm step, and how the team reads results. Write the answers on one page before anyone adds a connector. The page on AI LinkedIn outreach for RevOps shows the same four gates in one table.
How long does it take to add AI to outbound?
The connection takes minutes. The decisions take longer. Plan one working session for the four decisions, one week with the AI in read-only use, and then a first campaign. That is our estimate for a small team, not a measured figure.
Should every rep connect their own AI assistant?
Not at the start. Begin with one person who is signed in to the GTM Bud account and who does the review. Whoever is signed in can confirm a launch, so keep that to the people you named in the checklist.
What do I tell the CEO who expected full automation?
Tell them what is automated and what is not. The AI reads, sorts, drafts and prepares. A person reviews each reply and confirms each launch. That is the version that does not put an unread message in front of a buyer.
Can the AI report outbound results to leadership automatically?
No. GTM Bud does not send a scheduled report. Results come per campaign when someone asks, and the numbers carry an as-of time. Make the weekly review a question that one person asks on a fixed day.
Add AI to outbound in the right order
To add AI to outbound this quarter without a cleanup next quarter, make the four decisions first, run one week of read and prepare actions, and launch only through the readiness check and a confirmed preview. Setup takes a few minutes with the steps on GTM Bud for Claude. If you are new to GTM Bud, start from the page on AI LinkedIn outreach for RevOps, and if you are still choosing a tool, read how to vet an MCP connector for sales.