Skip to main content

​
Campaign Feedback Loops

The best campaigns get better over time. This guide shows you how to import campaign results back into your context file to refine hypotheses, update proof points, and improve future campaigns.

​
The Feedback Loop

Campaign → Results → Context Update → Better Hypotheses → Next Campaign
Every campaign teaches you:
  • Which pain points resonate (hypothesis validation)
  • Which proof points drive replies (proof library tuning)
  • Which roles engage (ICP refinement)
  • Which email structures work (voice evolution)

​
When to Run Feedback Loop

Import results after your campaign has run for 1-2 weeks with at least 50 sends. Earlier than that: not enough signal. You need:
  • Open rate data
  • Reply data (positive, negative, neutral)
  • Bounces and unsubscribes
  • Which hypothesis/tier each email used

​
Data Sources

You can import from:
  1. Email sequencer exports (Instantly, Smartlead, Lemlist)
  2. Manual tracking (spreadsheet with opens/replies)
  3. Pasted reply threads (for qualitative analysis)

​
Instantly Export

Go to your campaign → Export leads → Download CSV with:
  • Email address
  • Campaign status (Completed, Active, Paused)
  • Lead status (Interested, Not Interested, No Reply)
  • Email opened (Yes/No)
  • Replied (Yes/No)

​
Smartlead Export

Campaigns → Select campaign → Export → Include:
  • Lead details
  • Sequence stats
  • Reply status

​
Workflow: Import Results

1

Trigger Feedback Loop Mode

Ask Claude to run context-building in feedback loop mode.
Update my context file with campaign results.
Campaign: marketplace-supplier-q1
Here's the export from Instantly [attach CSV]
Or paste results directly:
Import campaign results to my context file:

Campaign: marketplace-supplier-q1
Vertical: B2B marketplaces
List size: 250
Emails sent: 220
Opens: 132 (60%)
Replies: 18 (8%)
Positive replies: 12 (67% of replies)

Top hypothesis: #1 Stale supplier data (14/18 replies)
Bounces: 15
Unsubscribes: 3
2

Claude Extracts Metrics

Claude reads your data and extracts:
  • Campaign name, vertical, list size
  • Open rate, reply rate, positive reply rate
  • Which hypotheses got replies (from tier/hypothesis columns in your emails CSV)
  • Patterns in positive vs negative replies
  • Any new pain points mentioned in replies
3

Update Campaign History

Claude adds a new row to ## Campaign History in your context file.
| Campaign | Vertical | List Size | Reply Rate | Top Hypothesis | Learnings |
|----------|----------|-----------|------------|----------------|----------|
| marketplace-q1 | B2B marketplaces | 250 | 8% | #1 Stale supplier data | CEO/founders 2x reply rate vs VPs. "Profile decay" keyword drove engagement. |
4

Validate or Retire Hypotheses

Claude updates ## Active Hypotheses based on performance.Promote to Validated:
### Hypothesis 1: Stale supplier data ✓ VALIDATED

**Performance**: 8% reply rate across 220 sends (marketplace-q1)

**What worked**: "15-20% profile decay per year" stat resonated. 
CEO/founder replies mentioned rebrand/acquisition scenarios.

**Refinement**: Focus on 2+ year old marketplaces. Younger platforms 
don't feel this pain yet.
Demote to Retired:
### Hypothesis 3: Search latency ✗ RETIRED

**Performance**: 0 replies across 40 sends (marketplace-q1)

**Why it failed**: Too technical for sales/ops roles. May work for 
engineering audience but not current ICP.
5

Update Proof Library

If campaign results surface new proof points, Claude adds them.New win case from a reply:
| Proof Point | Best Audience | Best Hypothesis | Source |
|-------------|---------------|-----------------|--------|
| PS. BuildZoom (10K suppliers) uses our API to refresh profiles quarterly—cut support tickets by 30%. | B2B marketplaces | Stale supplier data | BuildZoom (won via marketplace-q1) |
Remove proof points that didn’t resonate:
~~PS. Fortune 500 companies use our search API.~~ 
(Too generic, 0 replies in 3 campaigns)
6

Extract New Hypotheses from Replies

If replies mention pain points you didn’t predict, Claude surfaces them.Example reply:
“Interesting timing—we’re actually struggling with duplicate suppliers in our database. Same company, 3 different profiles because of name variations. Do you handle entity resolution?”
Claude suggests:
### Hypothesis 5: Duplicate supplier profiles (NEW)

**Pain**: Marketplaces accumulate duplicate company records due to name 
variations ("Acme Inc", "Acme Corp", "Acme Corporation").

**Mechanism**: Manual onboarding + poor entity resolution = data bloat.

**Evidence**: Mentioned in 3 replies from marketplace-q1.

**Status**: TESTING - add to next campaign

​
What Gets Updated

Feedback loop mode touches these sections in your context file:
SectionUpdate Logic
Campaign HistoryAdd new row with metrics + learnings
Active HypothesesPromote (validated), demote (retired), or add (new from replies)
Proof LibraryAdd new proof points from wins, remove non-performers
ICPRefine role targeting based on reply patterns
VoiceUpdate hard constraints if replies indicate confusion

​
Analyzing Replies: Positive vs Negative

​
Positive Reply Patterns

Look for:
  • Pain confirmation: “Yes, we see this problem”
  • Detail requests: “How does your API work?”
  • Timing signals: “Interesting timing, we’re evaluating…”
  • Forward/intro offers: “Let me connect you with our VP Product”
What to extract:
  • Which hypothesis they confirmed
  • Language they used (add to hypothesis description)
  • Any new pain points they mentioned

​
Negative Reply Patterns

Look for:
  • Irrelevant: “We don’t have this problem”
  • Wrong persona: “You should talk to [different role]”
  • Timing: “Not a priority right now”
  • Competitive: “We built this in-house”
What to extract:
  • Hypothesis mismatch (refine targeting)
  • Role mismatch (update ICP)
  • Positioning issues (update voice/value prop)

​
Example: Full Feedback Loop

​
Campaign Results

Campaign: procurement-vendor-intelligence-q1
Vertical: Procurement software
Sent: 180 emails
Opens: 95 (53%)
Replies: 11 (6%)
Positive: 7 (64%)

​
Breakdown by Hypothesis

HypothesisSendsRepliesReply Rate
#1 Vendor coverage gaps (APAC/LATAM)80810%
#2 Vendor onboarding efficiency6023%
#3 Compliance data gaps4012.5%

​
Breakdown by Role

RoleSendsRepliesReply Rate
VP Product7068.6%
Chief Product Officer5048%
Head of Data3512.9%
VP Engineering2500%

​
Context File Updates

Campaign History: 
| procurement-q1 | Procurement software | 180 | 6% | #1 Vendor coverage gaps | Product leaders (VP, CPO) 8%+ reply rate. APAC/LATAM angle resonated. Engineering roles didn't engage. |
Active Hypotheses: 
### Hypothesis 1: Vendor coverage gaps ✓ VALIDATED
Performance: 10% reply rate (procurement-q1)
Refinement: Focus on companies with international customer base. 
Domestic-only platforms don't feel this pain.

### Hypothesis 2: Vendor onboarding efficiency → NEEDS TESTING
Performance: 3% reply rate (procurement-q1)
Note: Only 60 sends, not enough data. Test again with 100+ sends.

### Hypothesis 3: Compliance data gaps ✗ RETIRED
Performance: 2.5% reply rate, 1 reply across 40 sends
Why it failed: Compliance is a top-down mandate, not a product pain. 
Wrong buyer persona.
ICP Refinement: 
| Procurement Software Leadership | 100-1000 | US/EU | VP Product, Chief Product Officer | Enterprise SaaS, Procurement |
Remove: Head of Data, VP Engineering (poor reply rates) Proof Library: 
| PS. SAP Ariba integrated our vendor search and saw 40% more APAC matches in Q1. | Procurement software | Vendor coverage gaps | SAP Ariba partnership (won via procurement-q1) |

​
Multi-Campaign Analysis

After 3-5 campaigns, patterns emerge across verticals.

​
Cross-Vertical Learnings

## Campaign History

| Campaign | Vertical | List Size | Reply Rate | Top Hypothesis | Learnings |
|----------|----------|-----------|------------|----------------|----------|
| marketplace-q1 | B2B marketplaces | 250 | 8% | Stale supplier data | CEO/founders 2x VPs |
| investment-q1 | Investment platforms | 180 | 12% | Deal flow discovery | "Emerging" keyword |
| procurement-q1 | Procurement software | 180 | 6% | Vendor coverage gaps | Product leaders 8%+, engineering 0% |
Patterns:
  1. Founder/CEO/CPO titles > VP titles across all campaigns
  2. Specific keywords (“emerging”, “APAC”, “decay”) drive engagement
  3. Engineering roles don’t reply to GTM pitches (even for technical products)
  4. Larger companies (500+ employees) have higher reply rates (better data, clearer pain)
Actions:
  • Update ICP to prioritize Founder/CEO/CPO
  • Build a “high-signal keywords” library
  • Remove engineering roles from people search
  • Increase list-building employee filters (100+ → 200+)

​
Hypothesis Library Evolution

​
Hypothesis Lifecycle

Untested → Testing (1 campaign) → Validated (2+ campaigns, 6%+ reply rate)
                   ↓
               Retired (0-2% reply rate)

​
After 10 Campaigns

You’ll have:
  • 5-7 validated hypotheses (work across verticals)
  • 10-15 vertical-specific hypotheses (validated per segment)
  • 20+ retired hypotheses (learned what doesn’t work)
This becomes your hypothesis library. New verticals start with validated templates, not blank slates.

​
Call Recording Integration

If a reply turns into a call, import the transcript to extract deeper signals. 
Update my context file with this call recording.
Prospect: Jane Doe, VP Product at Acme Procurement
Context: Reply from procurement-q1 campaign
[Paste transcript or attach file]
Claude extracts:
  • ICP signals: What Jane cares about, her workflow, team structure
  • Win case data: If they become a customer, capture their use case
  • Proof point candidates: Specific results or quotes
  • Hypothesis validation: Which pain points Jane confirmed
  • Voice feedback: Reaction to positioning or language
This creates a tight loop: email → reply → call → context update → better emails.

​
Automation: Export → Import

For high-volume campaigns, automate the export → import step.

​
Instantly Webhook → Claude Code

  1. Set up Instantly webhook for campaign completion
  2. Trigger a script that downloads the CSV export
  3. Call Claude Code API with: 
    Update context file with campaign results.
Campaign: {campaign_name}
[Attach CSV]
This runs feedback loop mode automatically 2 weeks after each campaign launches.

​
Best Practices

Import early and often: Even partial results (50 sends) give signal. Don’t wait for campaign completion.
Tag replies by sentiment: Positive, Negative, Neutral. This helps Claude differentiate hypothesis validation from objections.
Don’t over-rotate on small samples: 1-2 replies for a hypothesis isn’t validation. Wait for 5+ replies before promoting to validated.
Track seasonal patterns: If a hypothesis works in Q1 but not Q4, note it. Budget cycles affect reply rates.
Preserve negative learnings: Retired hypotheses are just as valuable as validated ones. They tell you what NOT to try in the next vertical.

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us