# Diagnose a campaign by its bottleneck

A step-by-step checklist for finding why a Victoria AI campaign underperforms, starting at the first weak stage of the funnel and fixing that first.

A campaign is a funnel: contacted, accepted, replied, positive, booked. Each stage feeds the next, so the first weak stage limits everything after it. Always start with the bottleneck, and fix it before touching anything downstream.

## The campaign diagnosis checklist

Work through these in order and stop at the first one that fails. The metric names are the ones on **Campaigns → your campaign → Analytics**.

1. **Is it sending?** **Contacted** should be growing. If not, the cause is operational, not copy. See "Rule out operational causes first" below.
2. **Is there enough data?** Wait for about 100 contacted leads, contacted at least two to three weeks ago, before judging rates. See [Outbound benchmarks](https://docs.versionseven.ai/playbook/benchmarks).
3. **Find the bottleneck.** Compare each stage with the benchmarks, in funnel order: **Accept rate**, then replies, then **Positive reply rate**, then **Meeting rate**. The first stage below its range is the bottleneck.
4. **Low acceptance on LinkedIn?** Fix the sender's LinkedIn profile, and send requests blank.
5. **Accepted but few replies?** Fix message relevance: personalize to a real problem, find a way to give, and tighten the opener.
6. **Email gets few replies?** Fix deliverability first (dedicated domain, warm-up, DNS), then the copy.
7. **Replies but few positive?** Fix the offer and the targeting.
8. **Positive replies but few meetings?** Make the meeting more valuable, frame it as a valuable next step, and give first.
9. **Change one thing at a time**, ideally as an [A/B test](https://docs.versionseven.ai/help/ab-testing), and re-measure once each arm has enough contacted leads.

Each bottleneck below has its own checks and fixes.

## Rule out operational causes first

Before judging copy, make sure the campaign is actually running as intended. A campaign that sends little looks like one that gets no replies.

- A red **This campaign isn't sending** banner, a paused switch, or a disconnected sender.
- **Not yet contacted** at zero: the campaign has run out of leads. The **Dashboard** shows "is out of leads" for an active campaign with nothing left to send.
- Daily limits, sender caps or a sender still in its warm-up ramp holding volume down.
- A credit balance at zero or a failed payment.
- AI Personalization fields holding leads because they have no working fallback.

See [Leads aren't moving](https://docs.versionseven.ai/help/troubleshooting-leads-not-moving) for every cause and its fix. You can also ask the Copilot "why isn't this campaign sending?".

## Bottleneck: low LinkedIn connection acceptance

Signs: **Accept rate** below the benchmark range, so few leads reach the **Yes** path where LinkedIn messages go out.

Check, in order:

1. **The sender's profile.** The request shows only a photo, a name and a headline. Is the headshot clear, professional and recent? Does the headline say in a few words who the sender helps? See [Optimize your LinkedIn profile](https://docs.versionseven.ai/playbook/linkedin-profile).
2. **The request itself.** Send it blank. A note that pitches gets ignored or declined. See [LinkedIn connection requests](https://docs.versionseven.ai/playbook/linkedin-connection-requests).
3. **The acceptance window.** If the wait after the request is short, people who would have accepted end up on **No**. About 10 days works well. See [Step delays](https://docs.versionseven.ai/playbook/step-delays).
4. **The targeting.** Very senior or very large-company leads accept less. If the profile and request are right, check that the list matches who you help.

## Bottleneck: accepted connections but few replies

Signs: a healthy **Accept rate** but few people reply on LinkedIn (**Replied** divided by **Connected**, on a LinkedIn-only campaign).

Check, in order:

1. **Relevance.** Does the first message name a real problem this person has? Move up the personalization hierarchy: a buying signal beats a company or role detail, and both beat a personal detail. See [Personalization that works](https://docs.versionseven.ai/playbook/personalization-instructions).
2. **The give.** Does every message offer something useful, or does it ask? Turn asks into gives. See [Give before you ask](https://docs.versionseven.ai/playbook/give-before-you-ask).
3. **The opener.** Use "I help X do Y by Z" or a specific, true observation followed by one question. See [Openers that get replies](https://docs.versionseven.ai/playbook/openers).
4. **Length and slop.** Two or three short lines, no em dashes, nothing that reads like a template. See [Slop](https://docs.versionseven.ai/playbook/slop-language).

## Bottleneck: cold email gets few replies

Signs: plenty of **Emails sent** but a **Reply rate** below the email benchmark, especially when LinkedIn on the same list does better.

Check deliverability first, because no copy works in the spam folder:

1. **Sending domain and mailbox.** Is the campaign sending from a dedicated sending domain, not the mailbox the team uses every day?
2. **Warm-up.** Was the mailbox warmed up with a warm-up service before sending? Victoria ramps volume but doesn't warm mailboxes.
3. **DNS.** Does the mailbox's chip on **Sender Accounts** read **DNS ok**? Fix SPF, DKIM and DMARC first. See [DNS setup](https://docs.versionseven.ai/help/dns-setup).
4. **Bounces and links.** Use verified emails, and keep links and attachments out of sequence emails.

Then check the copy the same way as for LinkedIn: relevance, the give, the opener and length. See [Sending mailbox best practices](https://docs.versionseven.ai/help/sending-mailbox-best-practices) and [Cold email principles](https://docs.versionseven.ai/playbook/cold-email-principles).

## Bottleneck: replies but few positive replies

Signs: a healthy reply rate, but a low **Positive reply rate**: replies are mostly negative or neutral.

Check, in order:

1. **Read the replies.** In the **Inbox**, filter by the campaign and read the last 20. "Not relevant" and "wrong person" point at targeting; "not interested" across the board points at the offer.
2. **The targeting.** Tighten titles, seniority, company size and industry so the list is people who own the problem. See [Who to target and on which channel](https://docs.versionseven.ai/playbook/targeting-and-channels).
3. **The offer.** Is the give something this buyer wants? Is the "I help X do Y by Z" line specific enough that they recognise themselves? See [Framing your value proposition](https://docs.versionseven.ai/playbook/value-proposition).

## Bottleneck: positive replies but no meetings booked

Signs: **Positive replies** coming in, but a low **Meeting rate** (**Meetings booked** divided by **Positive replies**).

Check, in order:

1. **The Appointment Setter is on**, in **Autonomous** mode, with a working **Booking link**, and conversations marked **Needs you** are answered the same day.
2. **Give first.** Does the reply deliver what was offered before asking for the meeting? See [After a positive reply](https://docs.versionseven.ai/playbook/after-a-positive-reply).
3. **Make the meeting more valuable.** Is it framed as a valuable next step, with what they'll see or get on the call, rather than a donation of their time?
4. **Unanswered questions.** Read the conversations for questions the agent couldn't answer, and add those facts to its **Knowledge base**.
5. **Bookings not recorded.** **Meetings booked** counts only bookings the prospect confirmed or you marked. If prospects book without saying so, select **Mark meeting booked** in the **Inbox**.

## After the fix: measure the change

Change one thing, at the bottleneck, and give it enough leads to show a result. An A/B test makes the comparison clean: Variation B differs from A at the bottleneck, such as a new opener or a new give, and the verdict needs 50 contacted leads per arm before it calls anything. Then start the checklist again: fixing one stage often reveals the next.

The weekly campaign review follows the same order and proposes changes for you to approve in the Copilot. See [Your first month with Victoria AI](https://docs.versionseven.ai/help/first-month).