Earlier I wrote about the first 375 cold emails. This is the longer retrospective — what I've concluded since, after the campaign's evolved and the data has matured. Some of the initial takeaways held up. Some didn't.

The Setup, Again

Johnny-5 (an Optiplex 7080 bare-metal Ubuntu box) runs the SiteLens outreach pipeline. The pipeline has:

The campaign has run for ~11 weeks. 375 emails sent. 349 unique recipients (some had variants A/B tested).

The Headline Numbers, Updated

Raw conversion: 0.3% reply rate, 0% close rate. This is below industry benchmark for cold B2B and significantly below my own optimistic projections.

What I Got Wrong

Three significant miscalibrations:

I overestimated how much personalized audit data moves the needle. My hypothesis: a cold email that leads with "your LCP is 5.3s on mobile and your meta descriptions are missing on 12 of your core pages" would be hard to ignore. Reality: most recipients either (a) don't understand what those things mean, or (b) understand, are told this by every SEO tool, and have learned to filter it as noise.

The audit data isn't useless — it generates some interest — but it doesn't cut through filters the way I hoped. The real thing that gets replies is whatever makes a human recipient believe there's a specific human on the other side who's worth responding to.

I overestimated how much AI personalization moves the needle. Claude is good at writing variant-personalized copy. We A/B-tested variants with and without deep personalization. The variant with Claude-written deep personalization did marginally better (0.4% reply rate vs 0.2%) — but the confidence interval on that is wide and the absolute numbers are tiny.

The insight is that deep personalization has diminishing returns past a baseline. Saying "hi, saw your site, noticed X" is worth something. Saying "hi, saw your site, noticed X, and also Y, and also Z, and also one very specific thing about your industry" is barely better than saying X.

I underestimated how dead "cold email to SMBs" is as a channel. SMB owners get hammered with outreach. They're adapted. They delete by default. The ones who do respond tend to be the most-engaged owners who are actively looking for help, which is a small fraction.

What Actually Works (Somewhat)

The 4 human replies fell into two groups:

Group 1 (2 replies): owners who were already looking for SEO/performance help. Our email arrived at the right moment. Nothing we did magical; right place, right time.

Group 2 (2 replies): owners who were specifically intrigued by the AI-readiness angle. The SiteLens score breaks down into SEO, performance, trust, content, and AI-readiness. The AI-readiness piece (does your site surface structured data for LLM retrieval? are your descriptions optimized for AI answers?) got responses that the traditional SEO angle didn't.

The second group is the interesting one. AI-readiness is a new framing. Nobody else is sending cold emails about it. It's novel, so it cuts through default filters that have trained on traditional SEO pitch patterns.

The Economics

375 emails × my founder hourly rate on the pipeline development = more hours than the campaign has generated in revenue. Measured narrowly, the campaign is a loss.

Measured more broadly — what did we learn, what did we build that we can reuse — it's a win. The DB, the scoring pipeline, the AI personalization infrastructure, and the audit-delivery mechanics are all durable assets. The campaign was the forcing function that caused them to exist.

SiteLens itself has pivoted based on what the campaign taught us. It's now positioned more as a SaaS tool with API-based usage (white-label agency API, programmatic audits) than a cold-outreach-lead-gen play.

What I Wish I'd Known About Deliverability

The 9.2% bounce rate sounds high; for cold B2B it's actually mid-pack. But the bounces hurt deliverability for the surviving sends. Mailbox providers see "lots of bounces" as a signal that you're a spammer, even when the bounces are from genuinely-stale addresses you scraped honestly. The fix is aggressive list hygiene before the first send: NeverBounce or ZeroBounce verification, drop anything that's not a confident pass, accept that the deliverable list is half the size of the scraped list.

We didn't do this on the first 200 sends. We did it on the next 175. Bounce rate dropped to under 4% on the verified subset. Open rates went up. Replies stayed about the same in absolute terms, which means the upgrade didn't move the conversion needle — but it preserved the sending domain's reputation, which matters for the next campaign.

What I'd Do Differently

I'd have tested channels beyond cold email earlier. LinkedIn outreach, agency partnerships, content marketing — all of which have different physics. Cold email is the most measurable but also the hardest channel right now.

I'd have built the AI-readiness scoring earlier in the pipeline. Once I realized that was the angle that got replies, I wished I'd led with it from week one.

I'd have had Jess review more of the outbound copy. She's a better writer than I am and her voice triggers fewer spam filters than my founder-engineer voice does.

The Agency API Pivot

The current version of SiteLens is the API-first agency tool. White-label audits, plan-based usage, admin key management. Agencies embed it in their own offerings. This is a better business than cold-email lead-gen because the agencies already have relationships with SMBs. We bring the audit, they bring the audience.

Early agency partners are coming in through organic channels (SiteLens GitHub repo visibility, Dangercorn blog posts, word of mouth). The deals are slower to close than cold-email would have been if it had worked, but they're larger and stickier when they do.

The Harder Truth

Cold email works for some people. Those people tend to be operating in markets where the pain is severe enough that recipients are actively hunting for solutions. SMB websites that are slow or poorly SEOed: not severe enough pain. The owner knows it's not perfect and has decided to live with it.

Our pivot is toward markets where the pain is severe. That looks less like cold SMB email and more like "here's a problem that's costing someone $10K/month, and our product fixes it." That's where outbound still works.

The honest lesson from 375 cold emails is that channel viability is domain-specific. A tactic that works brilliantly in one market is dead in another. Test the tactic; don't inherit it from a blog post about someone else's market.

Related

The original outreach post. The fleet that runs the pipeline. The channel we're trying next. The unit economics that survive despite the campaign.