How to run a weekly outbound review with an AI assistant

Run a weekly outbound review with an AI assistant: pull your replies, find the one line costing you meetings, and test a single change each week.

Get started

Short answer

Run a weekly outbound review by putting your sent invites, the actual reply text and your booked calls in front of an AI assistant and asking it to rank what failed. The assistant should name the specific sentence or targeting choice that lost people, not just report how many leads you touched.

Start with a 30 minute block

Most founders look at total numbers during their weekly review. Meetings booked tells you the result, not the cause. The point of this block is to move from counting leads to finding the sentence or the targeting choice that is costing you meetings.

You don't need a dashboard. You need the raw text of what you sent, what came back, and a prompt that pushes the assistant to be critical instead of agreeable. Ten minutes to gather, ten minutes with the AI, ten minutes deciding what changes next week. Book it as a recurring slot, because the review only works if it happens before you write the next batch of messages.

  1. 1

    Export last week's activity

    Copy every invitation you sent, every first message, and every reply into one plain text file. Keep each person's headline and job title next to their reply. The assistant can't judge targeting without knowing who you targeted.

  2. 2

    Give the assistant a role and a bias

    Open with something like: you are a skeptical sales coach reviewing my outbound. Your job is to find the weakest link, not to reassure me. Without that instruction you get a summary that agrees with everything.

  3. 3

    Ask for the funnel, not the totals

    Have it work out acceptance and positive reply share from your own numbers, then say which step loses the most people. A drop between acceptance and reply points at copy. A drop between reply and meeting points at the ask.

  4. 4

    Group the noes

    Ask the assistant to sort every negative reply into wrong person, wrong time, wrong offer. Whichever bucket is largest tells you what to change, and it's usually not the wording.

  5. 5

    Pick one change and write it down

    Record the change, the date and the metric you expect to move in a running log. Next week's review starts by checking that line, so you stop relitigating the same decision.

What to put in front of the AI

An assistant that only sees a meeting count can't tell a targeting problem from a copy problem. Give it the funnel and the text behind each step.

Paste replies exactly as they were written, typos included. Cleaned-up summaries lose the phrasing that tells you whether someone was confused, annoyed, or interested but busy.

Invites sent and accepted

Data
Whether your profile and connection note earn a yes before anyone reads a word
What it tells the assistant
LinkedIn sent invitations and your connections list

Replies sorted by intent

Data
Whether the offer lands with the people you picked, or the picking is wrong
What it tells the assistant
Message history, tagged as interested, not now, or no

Positive replies that never booked

Data
Whether your call to action asks for too much or hides the link
What it tells the assistant
Your calendar and your own notes

Messages with no reply at all

Data
Which opening lines and lengths get ignored
What it tells the assistant
Sent folder, copied as plain text

Prompts that find friction instead of summarizing

Asking for a summary gets you a polite paragraph that changes nothing. Ask for ranked failures with quotes attached.

Comparison prompts work best: people who replied against people who went quiet, or your first message against your follow-up. Forcing a comparison stops the assistant from hedging.

Paste this job

Analyze the 20 replies below. Do not summarize them. List the three objections that appear most often, with the exact quote for each, and tell me which one I can fix with wording and which one is a targeting problem.

My first message got 6 replies and my follow-up got 2. Read both and tell me what the follow-up assumes that the first message didn't. Give me one rewritten follow-up that doesn't repeat the first message.

Look at the 10 people who accepted my invitation and never replied. Group them by job title and company size. Tell me one thing my opening line claims that this group probably doesn't care about.

Last week I changed this sentence: [old line] to [new line]. Here are the replies since. Did it work, and what is the next single change?

Read intent before you read numbers

Counts hide the difference between a no and a not yet. Before you touch a script, read your replies as three piles: interested, wrong time, wrong person.

Wrong time is the pile people misread. Someone who says to check back after their rebrand is a live lead with a date attached. If your review counts that as a loss, you'll rewrite a message that was working.

Example

  1. You

    Hi Priya, I work with agency founders who handle their own outbound. Saw Northwind hired two account managers last month, so I guessed new business is growing. Worth a short call?

  2. Prospect

    Maybe next quarter, we're mid-rebrand and I can't take on anything new right now.

  3. You

    Makes sense. I'll come back in September. Want me to send the one page I'd have walked you through so it's on file?

  4. Prospect

    Sure, send it.

Log that exchange as a warm lead with a date. If it goes in the loss column, next week's review will tell you to change a message that already worked.

Change one variable per week

The common mistake is the total rewrite: new target, new offer, new script in the same week. If the numbers move, you've learned nothing about which part did it.

Treat outbound as one experiment at a time. Change a single variable, run it for a week, and let the next review judge that change against the weeks before it.

Rewrite everything at once

  • Switching industry, offer and script in the same Monday
  • Asking the assistant to make the whole sequence sound more professional
  • Starting over with a new value proposition whenever a week is slow

One variable, one week

  • Changing only the first sentence of the connection request and watching acceptance
  • Shortening the call to action and watching how many replies turn into bookings
  • Testing one answer to the most common objection in the second follow-up

Judge a change against your own last few weeks, not against a benchmark you read online. Change one line, run it for a week, then compare.

Doing this with Kinlead

Copying data out of LinkedIn every Monday is the part that gets dropped first. Kinlead has a REST Agent API and an MCP server, so an assistant like Claude or ChatGPT can find leads, check replies and draft answers without you exporting anything. Chat apps connect with OAuth and scripts use API keys.

Replies from every campaign land in one inbox, classified by intent with interested replies first, so the raw material for the review is already sorted. Leads sit in lead groups and score 0 to 100 against the profile you described, which makes a targeting problem visible before you blame your copy.

  • Ask the assistant to pull last week's interested replies and group them by the objection in each one.
  • Compare two lead groups by score to see whether a slow week came from the list or from the message.
  • Keep campaigns inside your send windows and daily limits while you test, so the experiment runs at a normal pace.

Campaign messages wait on the Actions page until they send, so you can fix a line mid-week instead of waiting for the next review.

Frequently asked questions

Related

Get started with Kinlead

Prefer markdown? How to run a weekly outbound review with an AI assistant.md