outreach-strategy.md
1 --- 2 title: 'Outreach Strategy' 3 category: 'pipeline' 4 last_verified: '2026-02-15' 5 tags: ['outreach', 'strategy', 'testing', 'database', 'api', 'ai', 'llm', 'email'] 6 status: 'current' 7 --- 8 9 # Outreach Strategy Analysis - Template vs AI-Personalized 10 11 **Date:** February 13, 2026 12 **Status:** ASSUMPTIONS - NEEDS REAL-WORLD VALIDATION 13 14 --- 15 16 ## Executive Summary 17 18 **Current Plan:** Template-based cold email first, AI score only interested prospects (2% response rate) 19 20 **Key Question:** Would AI-personalized outreach (mentioning specific issues we see on their site) improve conversion enough to justify the cost? 21 22 **Answer:** **Unknown - need to run A/B test with real data** 23 24 **Recommendation:** Start with hybrid approach - programmatic personalization + template, then test AI enhancement once we have baseline data. 25 26 --- 27 28 ## The Two Approaches 29 30 ### Approach A: Template-Based Cold Email (Current Plan) 31 32 **How it works:** 33 34 1. Scrape SERP → find site domain 35 2. Send template email (no AI scoring upfront): 36 37 ``` 38 Hi [Business Name], 39 40 I analyzed your website [domain] and noticed it could be converting more visitors into customers. 41 42 Would you like a free 5-minute AI analysis showing exactly what's holding back your conversions? 43 44 Reply YES and I'll send it over (no obligation). 45 46 Jason 47 333 Method 48 ``` 49 50 3. If they reply YES → score their site with AI → send detailed proposal 51 4. 2% response rate assumed → 20% of those convert to customers 52 53 **Pros:** 54 55 - **Extremely low cost:** $0 AI cost upfront (just email delivery via Resend) 56 - **High throughput:** Can send 7,500 emails/month easily 57 - **Filters for intent:** Only score sites where owner is interested 58 - **Fast iteration:** Test different subject lines/copy quickly 59 - **Scales efficiently:** API costs only for qualified leads (2% of outreach) 60 61 **Cons:** 62 63 - **Generic:** Doesn't show we actually looked at their site 64 - **Lower trust:** Feels like mass email blast 65 - **May miss "show don't tell" effect:** Some prospects need proof upfront 66 - **Untested response rate:** 2% is industry average, not our actual data 67 - **Untested conversion:** 20% is assumption, not validated 68 69 **Cost per customer:** 70 71 ``` 72 7,500 emails/month → 150 responses (2%) → 30 customers (20% conversion) 73 API cost: 150 sites × $2.00 = $300/month 74 Cost per customer: $300 / 30 = $10 per customer 75 ``` 76 77 **Cost per outreach:** $0.04 ($300 API cost ÷ 7,500 emails) 78 79 --- 80 81 ### Approach B: AI-Personalized Outreach (Alternative) 82 83 **How it works:** 84 85 1. Scrape SERP → find site domain 86 2. **Score site with AI first** (capture screenshots, analyze with GPT-4o-mini) 87 3. Send personalized email mentioning specific issues: 88 89 ``` 90 Hi [Business Name], 91 92 I just analyzed your website [domain] and noticed 3 quick wins that could double your conversions: 93 94 1. Your "Contact Us" button is below the fold on mobile (87% of visitors never see it) 95 2. Missing trust signals - no testimonials or certifications visible 96 3. Page loads in 8.2 seconds (industry standard is <3 seconds) 97 98 These are easy fixes that could turn your 3% conversion rate into 6-8%. 99 100 Want the full analysis? Reply YES and I'll send the complete report. 101 102 Jason 103 333 Method 104 ``` 105 106 4. If they reply → send full report with payment link 107 108 **Pros:** 109 110 - **Highly personalized:** Shows we actually analyzed their site 111 - **Builds trust:** Provides value upfront (free insights) 112 - **"Show don't tell":** Proof we know what we're talking about 113 - **May improve conversion:** Harder to ignore when we point out specific problems 114 - **Potential viral effect:** Recipients may share email with peers 115 116 **Cons:** 117 118 - **Very expensive:** Score all sites upfront ($2/site × 7,500 = $15,000/month) 119 - **High risk if conversion doesn't improve:** Burning $15K/month for unknown lift 120 - **Slower throughput:** Enrich stage bottleneck at 14,400 sites/month (without VPS) 121 - **May not improve conversion:** Some prospects don't care about specifics, just price 122 - **Wasted scoring on unqualified leads:** Scoring sites that would never buy regardless 123 124 **Cost per customer (if conversion stays at 0.4% cold):** 125 126 ``` 127 7,500 emails → 30 customers (0.4%) 128 API cost: 7,500 sites × $2.00 = $15,000/month 129 Cost per customer: $15,000 / 30 = $500 per customer (UNSUSTAINABLE) 130 ``` 131 132 **Cost per customer (if conversion improves to 1.0%):** 133 134 ``` 135 7,500 emails → 75 customers (1.0%) 136 API cost: $15,000/month 137 Cost per customer: $15,000 / 75 = $200 per customer 138 Revenue per customer: $297 139 Gross profit: $97 per customer (33% margin - acceptable but not great) 140 ``` 141 142 **Cost per outreach:** $2.00 ($15,000 ÷ 7,500 emails) 143 144 --- 145 146 ## Hybrid Approach C: Programmatic Personalization (Recommended Starting Point) 147 148 **How it works:** 149 150 1. Scrape SERP → find site domain 151 2. **Use programmatic rules to extract basic info** (no AI): 152 - Business name from `<title>` tag 153 - Phone number from HTML (regex patterns) 154 - Location from footer/contact page 155 - Industry from keyword that found them 156 - Basic page load time (Playwright performance metrics) 157 - Mobile responsiveness (viewport meta tag check) 158 159 3. Send semi-personalized template: 160 161 ``` 162 Hi [Business Name from title tag], 163 164 I found your website while searching for [industry] in [city] and ran a quick check. 165 166 Your site loads in [X seconds] and [mobile-friendly or not mobile-friendly]. 167 168 Most [industry] websites in [city] convert at 3-5%. Would you like to see where yours stands and how to improve it? 169 170 Reply YES for a free 5-minute AI analysis. 171 172 Jason 173 333 Method 174 ``` 175 176 4. If they reply → score with AI → send detailed proposal 177 178 **Pros:** 179 180 - **Better than generic template:** Shows we visited their site 181 - **Much cheaper than AI:** $0 upfront (programmatic extraction is free) 182 - **Scalable:** Can process 7,500 sites/month easily 183 - **Filters for intent:** Only score interested prospects 184 - **Low risk:** If it doesn't work, we've wasted minimal resources 185 - **Builds foundation for testing:** Establish baseline conversion, then test AI enhancement 186 187 **Cons:** 188 189 - **Not as personalized as AI:** Can't identify specific CRO issues upfront 190 - **Limited to simple metrics:** Page speed, mobile-friendly, basic patterns 191 - **May still feel templated:** If everyone gets "your site loads in X seconds" 192 193 **Cost per customer:** 194 195 ``` 196 Same as Template-Based Approach A: $10 per customer 197 ``` 198 199 **Implementation complexity:** Medium (need to add programmatic extraction to SERP/Assets stages) 200 201 --- 202 203 ## Contact Extraction: AI vs Programmatic 204 205 **Current approach:** AI extracts contacts at Rescore and Enrich stages using Claude/GPT vision 206 207 **Alternative:** Programmatic ruleset for most contact types, AI fallback for edge cases 208 209 ### Programmatic Extraction Rules 210 211 **Phone Numbers:** 212 213 - Regex patterns for common formats (US, AU, UK, etc.) 214 - E.164 validation 215 - Common HTML patterns: `tel:`, `<a href="tel:">`, class="phone" 216 - Coverage estimate: 80-90% of sites 217 218 **Email Addresses:** 219 220 - Regex: `/[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}/` 221 - Common patterns: info@, contact@, hello@ 222 - Mailto links: `<a href="mailto:">` 223 - Coverage estimate: 85-95% of sites 224 225 **Contact Forms:** 226 227 - Look for `<form>` tags with name/email/message fields 228 - Common form action patterns: /contact, /submit, /inquiry 229 - FormSubmit, Typeform, Google Forms patterns 230 - Coverage estimate: 90-95% of sites 231 232 **Social Media:** 233 234 - Regex for Twitter/X URLs: `twitter.com/`, `x.com/` 235 - Regex for LinkedIn: `linkedin.com/in/`, `linkedin.com/company/` 236 - Facebook: `facebook.com/` 237 - Coverage estimate: 95%+ (well-structured URLs) 238 239 ### When to Use AI Fallback 240 241 **Use AI when:** 242 243 - No phone/email found via regex (might be in image, obfuscated) 244 - Unusual contact form patterns 245 - Need to determine business vs personal email 246 - Need to extract from images/screenshots 247 - Need to validate contact is still active 248 249 **Cost comparison:** 250 251 - Programmatic extraction: $0 (built into pipeline) 252 - AI extraction: ~$0.20 per site (LLM call to analyze HTML/screenshots) 253 254 **Recommendation:** 255 256 1. Extract 80-90% of contacts via programmatic rules (free) 257 2. Use AI fallback for remaining 10-20% (reduces AI cost by 80%) 258 3. Monitor extraction success rate, refine rules over time 259 260 --- 261 262 ## A/B Test Plan (When to Validate) 263 264 **Timeline:** Start after Month 2-3 (once we have baseline data) 265 266 ### Test Setup 267 268 **Control Group (50%):** Template-based (Approach A/C) 269 **Treatment Group (50%):** AI-personalized (Approach B) 270 271 **Sample size:** 1,000 emails per group = 2,000 total 272 273 **Metrics to track:** 274 275 1. **Response rate:** % who reply to initial email 276 2. **Conversion rate:** % who purchase after receiving proposal 277 3. **Cost per customer:** API + email costs ÷ customers acquired 278 4. **Time to close:** Days from first email to purchase 279 5. **Sentiment:** Positive/neutral/negative tone in responses 280 281 **Success criteria for AI-personalized:** 282 283 - **Minimum 2x improvement in response rate** (2% → 4%+) OR 284 - **Minimum 2x improvement in conversion rate** (0.4% → 0.8%+) OR 285 - **Cost per customer <$200** (to justify higher upfront cost) 286 287 **If AI-personalized wins:** Scale to 100% of outreach, optimize prompts 288 **If template wins:** Stick with template, invest savings into more outreach volume 289 **If marginal difference (<20%):** Stick with template (not worth cost increase) 290 291 --- 292 293 ## Real-World Validation Requirements 294 295 **CRITICAL:** All assumptions below are **UNTESTED** and need real data: 296 297 ### Untested Assumptions 298 299 | Assumption | Source | Risk Level | How to Validate | 300 | ------------------------------------------ | --------------------------------- | ---------- | ---------------------------------------------------------------------- | 301 | **2% email response rate** | Industry average for cold email | Medium | Track actual response rate over first 1,000 emails | 302 | **20% conversion of interested prospects** | Assumption (no data) | **HIGH** | Track actual conversion from interested → paid over first 50 responses | 303 | **0.4% cold conversion rate** | Conservative estimate | Medium | Test AI-personalized approach with 1,000 emails | 304 | **Template is "good enough"** | Assumption | **HIGH** | A/B test template vs AI-personalized (Month 3) | 305 | **$2.00 API cost per site** | Based on current token usage | Low | Monitor actual API costs per site scored | 306 | **Programmatic extraction works 80%+** | Estimate based on common patterns | Medium | Measure extraction success rate vs AI fallback | 307 308 ### Required Metrics to Track (Starting Month 1) 309 310 **Email metrics (tracked via Resend):** 311 312 - Sent, delivered, bounced, opened, clicked 313 - Response rate (% who reply) 314 - Unsubscribe rate 315 - Spam complaints 316 317 **Conversion metrics (tracked in database):** 318 319 - Responses by intent (interested, not interested, spam, opt-out) 320 - Proposals sent (from interested responses) 321 - Purchases completed 322 - Time to close (days from first email → payment) 323 324 **Cost metrics (tracked weekly):** 325 326 - API cost per site scored 327 - API cost per customer acquired 328 - Email cost per outreach sent 329 - Total CAC (customer acquisition cost) 330 331 **Quality metrics (tracked monthly):** 332 333 - Customer satisfaction (NPS survey after delivery) 334 - Refund rate 335 - Testimonial/referral rate 336 337 ### Decision Points Based on Real Data 338 339 **After 100 emails sent:** 340 341 - If <1% response rate → revise subject line/copy 342 - If >5% spam complaints → soften approach, review CAN-SPAM compliance 343 344 **After 50 responses received:** 345 346 - If <10% convert to customers → revise proposal quality 347 - If >30% convert → response rate assumption was too conservative, can be more aggressive 348 349 **After 30 customers acquired:** 350 351 - Calculate real CAC (total costs ÷ 30) 352 - Calculate real conversion funnel (sent → response → customer) 353 - Update business plan projections with actual data 354 355 **After Month 3:** 356 357 - Run A/B test: Template vs AI-personalized 358 - Make definitive decision on outreach strategy 359 - Update Year 1 forecast based on real conversion rates 360 361 --- 362 363 ## Recommendation Summary 364 365 ### Month 1-2: Programmatic Personalization (Hybrid Approach C) 366 367 **Rationale:** 368 369 - Lowest risk (minimal upfront cost) 370 - Gathers baseline data for future testing 371 - Better than pure template, cheaper than AI 372 - Can extract 80%+ of contacts programmatically 373 374 **Implementation:** 375 376 1. Add programmatic extraction to SERP/Assets stages: 377 - Business name from `<title>` 378 - Phone/email via regex 379 - Page load time from Playwright metrics 380 - Mobile-friendly check 381 382 2. Create semi-personalized template using extracted data 383 384 3. Track response and conversion rates religiously 385 386 4. Score only interested prospects (2% response rate assumed) 387 388 **Expected results:** 389 390 - 2-3% response rate (better than pure template due to personalization) 391 - 15-25% conversion of interested prospects (assumption - needs validation) 392 - 15-20 customers in Month 2 (vs 8 in business plan forecast) 393 - API costs: <$200/month (only scoring interested prospects) 394 395 ### Month 3: A/B Test AI-Personalized vs Template 396 397 **Rationale:** 398 399 - Have baseline data from Months 1-2 400 - Can calculate whether AI lift justifies cost 401 - Low risk (only 50% of outreach uses expensive approach) 402 403 **Test parameters:** 404 405 - 1,000 emails per group (2,000 total) 406 - Track response rate, conversion rate, cost per customer 407 - Make data-driven decision 408 409 ### Month 4+: Scale Winning Approach 410 411 **If template wins:** 412 413 - Scale to 7,500+ emails/month 414 - Optimize copy, subject lines, timing 415 - Invest savings into more outreach volume 416 417 **If AI wins:** 418 419 - Upgrade to VPS ($70/month) for Enrich throughput 420 - Increase API budget to $1,000-1,500/month 421 - Scale to 3,000-5,000 AI-personalized emails/month 422 - Monitor CAC closely (target <$100/customer) 423 424 --- 425 426 ## Cost Breakdown Detail (for Business Plan) 427 428 ### Current "~$2 per customer" Claim 429 430 **Source:** PIPELINE-CAPACITY.md line 115-117 431 432 **Actual breakdown per site scored:** 433 434 1. **Scoring (GPT-4o-mini vision):** 435 - Input: 6 screenshots (cropped) = ~100,000 tokens 436 - Cost: 100K tokens × $0.15/1M tokens = **$0.015** 437 - Output: ~500 tokens @ $0.60/1M tokens = **$0.0003** 438 - **Scoring subtotal: $0.015** 439 440 2. **Rescoring (GPT-4o-mini, 60% of sites):** 441 - Same cost as scoring, but with below-fold screenshots 442 - **Rescoring subtotal: $0.015** 443 444 3. **Proposals (Claude Sonnet 4.5):** 445 - Input: ~50K tokens (screenshot analysis + prompt) @ $3/1M = **$0.15** 446 - Output: ~2K tokens @ $15/1M = **$0.03** 447 - **Proposals subtotal: $0.18** 448 449 4. **Enrichment (Claude Haiku for contact extraction):** 450 - Input: ~20K tokens (HTML + contact patterns) @ $0.80/1M = **$0.016** 451 - Output: ~500 tokens @ $4/1M = **$0.002** 452 - **Enrichment subtotal: $0.018** 453 454 **Total API cost per site (full pipeline):** 455 456 ``` 457 Scoring: $0.015 458 Rescoring: $0.015 (60% of sites) 459 Proposals: $0.18 460 Enrichment: $0.018 461 ────────────────── 462 TOTAL: $0.228 per site 463 ``` 464 465 **Note:** Original estimate of $2.00/site was **too high**. Actual cost is **~$0.23/site** (10x cheaper!). 466 467 **Updated cost per customer (selective scoring strategy):** 468 469 ``` 470 7,500 emails → 150 responses (2%) → 30 customers (20%) 471 API cost: 150 sites × $0.23 = $34.50/month 472 Cost per customer: $34.50 / 30 = $1.15 per customer (not $2.00!) 473 ``` 474 475 **This is MUCH better than projected!** 99.6% gross margin instead of 99.3%. 476 477 --- 478 479 ## Sources for Forecasts 480 481 ### 2% Email Response Rate 482 483 **Source:** Industry benchmarks for cold B2B email outreach 484 485 - **Mailshake 2024 study:** 1-3% response rate for cold email (average: 2%) 486 - **Woodpecker 2023:** 1.5-2.5% response rate for well-targeted cold email 487 - **HubSpot 2024:** 2-3% for personalized cold outreach 488 489 **Note:** These are ASSUMPTIONS from industry averages, not our actual data. 490 491 **Validation plan:** Track actual response rate starting Month 1, adjust forecast monthly. 492 493 ### 20% Conversion of Interested Prospects 494 495 **Source:** Assumption based on warm lead conversion benchmarks 496 497 - **Gartner B2B sales:** 20-30% of qualified leads convert (2024) 498 - **HubSpot inbound:** 15-25% conversion of "interested" prospects (2023) 499 - **Our assumption:** 20% of people who reply "YES" will buy the report 500 501 **Note:** This is an ASSUMPTION, not validated data. Actual conversion could be 5% or 40%. 502 503 **Validation plan:** Track actual conversion rate from first 50 interested responses, update forecast. 504 505 ### 0.4% Cold Conversion Rate (AI-Personalized Approach) 506 507 **Source:** Conservative estimate based on cold outreach benchmarks 508 509 - **Industry average:** 0.5-1.0% for cold B2B sales 510 - **Our assumption:** 0.4% (conservative) if scoring all sites upfront 511 512 **Note:** This is an ASSUMPTION. We chose template-based approach to avoid testing this expensive hypothesis. 513 514 **Validation plan:** A/B test in Month 3 to measure actual conversion with AI-personalized approach. 515 516 ### $0.23 API Cost Per Site (Corrected) 517 518 **Source:** Calculated from actual API pricing as of Feb 2026: 519 520 - **OpenRouter GPT-4o-mini:** $0.15/1M input, $0.60/1M output tokens 521 - **Anthropic Claude Sonnet 4.5:** $3/1M input, $15/1M output tokens 522 - **Anthropic Claude Haiku:** $0.80/1M input, $4/1M output tokens 523 524 **Token counts:** Measured from actual pipeline runs (see PIPELINE-CAPACITY.md lines 98-195) 525 526 **Validation plan:** Monitor actual API costs weekly, compare to forecast. 527 528 --- 529 530 ## Business Plan Updates Needed 531 532 ### Missing from Business Plan 533 534 1. **Token-level cost breakdown** for $0.23/site API cost 535 2. **Sources for 2% and 20% assumptions** (currently just stated as fact) 536 3. **Template vs AI-personalized analysis** (not discussed) 537 4. **Programmatic vs AI contact extraction** (not discussed) 538 5. **A/B testing plan** (not mentioned) 539 6. **Real-world validation requirements** (critical gaps) 540 541 ### Statements That Need Caveats 542 543 **Current (line 1071):** 544 545 > "Score only interested prospects (2% response rate to cold email)" 546 547 **Should be:** 548 549 > "Score only interested prospects (**assumed** 2% response rate based on industry benchmarks for cold B2B email - requires validation with real data)" 550 551 **Current (line 1096):** 552 553 > "Conversion rate improves from 0.4% (cold) to 20% (interested prospects only)" 554 555 **Should be:** 556 557 > "Conversion rate **estimated** to improve from 0.4% (cold, industry average) to 20% (interested prospects, based on warm lead benchmarks - **requires validation**)" 558 559 ### Recommended Additions 560 561 **Add to Financial Spreadsheets:** 562 563 1. API Cost Breakdown table (Scoring, Rescoring, Proposals, Enrichment) 564 2. Sources and Assumptions table (with validation status) 565 3. Real-world validation checkpoints (Month 1, 2, 3 decision points) 566 567 **Add to Risk Management section:** 568 569 1. Risk: "Conversion assumptions too optimistic" 570 2. Mitigation: "Track actual metrics from Month 1, update forecasts monthly, run A/B tests by Month 3" 571 572 --- 573 574 ## Conclusion 575 576 **We should NOT bet the farm on any approach until we have real data.** 577 578 **Starting strategy: Programmatic Personalization (Hybrid Approach C)** 579 580 - Lowest cost, lowest risk 581 - Better than template, cheaper than AI 582 - Establishes baseline for future testing 583 - Extract 80%+ contacts programmatically (free) instead of AI ($0.20/site) 584 585 **Validation timeline:** 586 587 - **Month 1:** Gather baseline metrics (response rate, conversion rate) 588 - **Month 2:** Refine based on early data 589 - **Month 3:** A/B test template vs AI-personalized 590 - **Month 4+:** Scale winning approach 591 592 **Most important action:** **Start the pipeline and track real metrics.** All forecasts are currently based on industry averages and assumptions, not our actual performance. 593 594 --- 595 596 **Last Updated:** February 13, 2026 597 **Status:** Recommendations based on analysis - NEEDS REAL-WORLD VALIDATION 598 **Next Steps:** 599 600 1. Update business plan with cost breakdowns and sources 601 2. Add validation plan and decision points 602 3. Implement programmatic extraction in SERP/Assets stages 603 4. **START RUNNING THE PIPELINE** to get real data