Data Preparation
Clean your CSV before upload
Clean your CSV before upload
- Remove empty rows
- Trim whitespace from cells
- Fix obvious typos in company names
- Standardize job titles where possible
Use consistent column naming
Use consistent column naming
Keep column names consistent across files if you reuse templates.
Company and Company Name are different variables.Deduplicate before upload
Deduplicate before upload
While Smelt detects duplicates, removing them beforehand saves processing and usage.
Include rich data fields
Include rich data fields
The more context you provide, the better the personalization:
- Job title + seniority level
- Industry + sub-industry
- Company size + growth stage
- Location (city, not just country)
- Recent news or funding info
Use UTF-8 encoding
Use UTF-8 encoding
When exporting from Excel or Google Sheets, always choose UTF-8 for best compatibility.
Template Writing
Be specific about output format
Be specific about output format
Tell the AI exactly what you want:❌ “Write a hook”✅ “Write a 1-sentence opening hook under 150 characters. Start with their pain point, not a compliment.”
Include examples in your prompt
Include examples in your prompt
Show the AI what good output looks like:
Specify what NOT to include
Specify what NOT to include
Use forbidden words and explicit instructions:
Set appropriate length limits
Set appropriate length limits
- Hooks: 100-200 characters
- Pain points: 150-300 characters
- Full emails: 500-1000 characters
- CTAs: 50-100 characters
Use all relevant variables
Use all relevant variables
Reference your data in the prompt:
Test before bulk runs
Test before bulk runs
Always test on 5-10 rows before running thousands. It’s worth the small usage cost to catch prompt issues early.
Cost Optimization
Use BYOK for high volume
Use BYOK for high volume
If processing 10,000+ rows/month, BYOK saves significant money. You pay ~0.02-0.10 on per-credit platforms.
Test on small row ranges first
Test on small row ranges first
When creating a new job, start with 10-50 rows to verify output quality. Then scale up.
Reuse templates across jobs
Reuse templates across jobs
Build a library of tested templates. Reuse them rather than recreating from scratch each time.
Monitor usage regularly
Monitor usage regularly
Check Settings → Billing to track your usage against limits. Avoid surprises.
Be strategic with template chains
Be strategic with template chains
Each template in a chain multiplies usage:
- 1,000 rows × 1 template = 1,000 usage
- 1,000 rows × 3 templates = 3,000 usage
Batch similar campaigns
Batch similar campaigns
If running multiple campaigns with the same template, combine into one job when possible.
Quality Optimization
Review flagged outputs first
Review flagged outputs first
Filter results by “has flags” and review those first. They’re most likely to need edits.
Use forbidden words strategically
Use forbidden words strategically
Add words that don’t fit your brand:
- Competitor names
- Overused phrases (“synergy”, “leverage”)
- Words that trigger spam filters
Iterate on your prompts
Iterate on your prompts
If outputs are consistently off, revise the prompt:
- Run small test batch
- Review outputs
- Identify patterns in bad outputs
- Adjust prompt
- Repeat
Add context to your templates
Add context to your templates
Help the AI understand your product/offer:
Workflow Tips
Start Simple
Begin with 1 template (hook or first line). Add more templates after you’ve validated quality.
Document Your Templates
Use clear names: “SaaS Hook v2 - Pain Focus” not “Template 1”
Version Your Templates
When iterating, duplicate and rename: “Hook v1” → “Hook v2” rather than editing in place.
Export Regularly
Download results after each job completes. Don’t rely on Smelt as long-term storage.