How to Run a Campaign Retrospective: Diagnosing What's Not Converting

Your signals dashboard is full. Your team is responding. But pipeline isn’t growing. Demos aren’t booking. Something in the machine is broken, and nobody can point to what.

This is where most B2B teams spiral. They tweak everything at once: new keywords, new messaging, new follow-up cadence. Three weeks later, nothing has changed because they fixed the wrong thing.

A campaign retrospective gives you the discipline to diagnose before you prescribe. Instead of guessing, you look at the funnel, find the drop-off, and fix exactly one lever at a time.

What this article covers: The complete campaign retrospective framework for social selling. This is the manual version of what the /campaign-retro skill (available through CatchIntent’s MCP server) automates. If you want to understand the methodology first, keep reading.

The Social Selling Funnel

Signals (AI-qualified buying intent posts)
    |
    v  -- Action Rate (target: >40%) -->
Acted On (team engaged with the signal)
    |
    v  -- Reply Rate (target: >10%) -->
Replied (prospect responded)
    |
    v  -- Win Rate (target: >15%) -->
Won (became a customer or entered pipeline)

The Three Levers

Lever 1: Targeting

Symptom: Low action rate (below 40%).

Root cause: Listeners surfacing posts that look like intent to AI but aren’t relevant to your ICP.

Common patterns:

• Keywords are too broad (“project management” catches students, not buyers)
• Platform mix is wrong (monitoring subreddits where your ICP doesn’t hang out)
• AI scoring threshold is too low, letting marginal signals through
• Brand info in your listener doesn’t accurately describe what you sell

Diagnosis questions:

1. Pull your last 20 ignored signals. What do they have in common?
2. Which keywords generated the most ignored signals?
3. Are ignored signals clustered on a specific platform?

Fix: Listener optimization. The 5-phase listener optimization framework walks through the process.

Quick test: Have three team members independently rate your last 30 signals as “would act” or “would skip.” If skip rates are above 60%, targeting is your problem.

Lever 2: Voice

Symptom: Action rate above 40% but reply rate below 10%.

Root cause: Outreach not landing. Messages feel generic, salesy, or disconnected.

Common patterns:

• Outreach opens with a pitch instead of addressing the prospect’s stated problem
• Messages are too long (more than 3-4 sentences for first touch)
• Tone mismatch: prospect posted casually, your team replied like a corporate press release
• No evidence that you actually read their post

Diagnosis questions:

1. Pull your last 10 unreplied messages. Read them out loud. Do they sound like a human or a template?
2. Compare to your last 5 messages that got a reply. What is different?
3. Are you referencing the specific thing the prospect said?

Fix: The voice formula: (1) Reference their specific post, (2) Relate with a brief insight, (3) Offer something useful without asking for a meeting.

Lever 3: Reply Handling

Symptom: Reply rate above 10% but win rate below 15%.

Root cause: Team fumbles the handoff from conversation to sales.

Common patterns: Slow follow-up (>24 hours), no clear next step, jumping to hard-sell too fast, not qualifying.

Fix:

The Diagnostic Flowchart

Running the Retro: Step by Step

Block 45 minutes every two weeks.

Step 1 (10 min): Pull numbers. Signals generated, action rate, reply rate, win rate, average time to action. Use /weekly-report for automated data.

Step 2 (5 min): Identify the broken lever using the flowchart.

Step 3 (15 min): Root cause analysis. Pull actual examples. “Reply rate is 7% and 8 of 10 unreplied messages opened with a product pitch” is actionable. “Reply rate is 7%” is not.

Step 4 (10 min): Stop / Start / Continue.

Stop: Things hurting performance. (Stop using the generic template for Reddit signals.)

Start: New things to try. (Start referencing the prospect’s exact post in the first line.)

Continue: Things working. (Continue the conversational tone on X outreach, 18% reply rate.)

Step 5 (5 min): Assign and schedule. Each item gets an owner and deadline.

Before and After Example

Retro 1: Action rate 28%. Targeting problem. “Engineering tools” keyword matching physical engineering. Added negative keywords, removed noisy subreddits.

Retro 2: Action rate 51%. Reply rate 6%. Voice problem. Messages were 6-8 sentences with same opener. Rewrote to 3 sentences leading with specific pain point.

Retro 3: Reply rate 14%. Win rate 9%. Reply handling problem. Average response time was 38 hours. Implemented 4-hour SLA. Win rate moved to 19%.

Three retros. Three levers. Six weeks. Healthy funnel.

Common Retro Mistakes

Fixing everything at once. Change one lever per cycle or you can’t tell what worked.

Using averages instead of actual examples. Numbers tell you where. Examples tell you what.

Skipping retros when things go well. The “Continue” list matters as much as “Stop” and “Start.”

Running without the people who do outreach. Reps have context dashboards can’t capture.

Frequently Asked Questions

How often should I run a retro?

Every two weeks. Monthly is too slow for catching issues. Weekly is too noisy for patterns. A two-week window gives 50-200 signals to analyze.

What if action rate and reply rate are both below benchmark?

Fix targeting first. Bad signals make any message fail. Get action rate above 40%, then reassess reply rate.

How many signals do I need before a retro is useful?

At least 50 per period. Fewer and individual outliers swing percentages too much.

Should I track per platform or aggregate?

Both, starting with aggregate. Once you hit benchmarks overall, break down by platform to find which one drags the average.

What’s the difference between a campaign retro and listener optimization?

A retro covers the full funnel (targeting, voice, reply handling). Listener optimization zooms into just targeting. The 5-phase framework is the deep-dive for when the retro identifies targeting as the problem.