What we're learning in public.
Build-in-public notes from inside the product. Agentic decision logs, deliverability benchmarks, ICP-discovery experiments — the kind of thing we wish someone had written when we started. Long-form, citable, no listicles.
-
Multi-source lead discovery vs. a contact database
A contact database sells you a static list everyone else also bought. Multi-source discovery routes your ICP across many live sources and finds the B2B companies you actually sell to — including the long tail one database misses. Here's the difference, and why it changes who you can reach.
-
Deliverability for founders: warmup, bounce-pause, and SPF/DKIM/DMARC
Cold email lands in spam for boring, fixable reasons: unauthenticated domains, cold mailboxes sending too fast, and bounce rates that torch your reputation. Here's the founder's version of deliverability — what actually protects your domain, and what Overwise enforces by default so you can't shoot yourself in the foot.
-
How AI qualifies a B2B lead: absence signals and buying triggers
Finding companies is the easy half. Qualification is what turns a list into leads worth emailing. Here's how configurable absence signals — what a company doesn't have — plus hiring and funding triggers score each lead 0–100 against your product, not a generic firmographic filter.
-
Cite-or-discard as a product primitive: how the MessageVerifier earns its keep
The MessageVerifier is a binary cite-or-discard gate on every draft the agent produces. Every factual claim has to point at evidence from the lead's context — or the draft is discarded and re-drafted. Why we built it as a primitive instead of a score, and what it costs.
-
Cross-tenant cooldown: 7 days because we'd want it ourselves
When two of our customers independently find the same lead, the second one to send is silently held for seven days. Thirty days if the lead already replied. Neither customer sees the other. Here's why we built it that way, what it costs each tenant, and why the alternative — both customers firing in the same week — is the failure mode we refuse to ship.
-
The reply-rate trap: why we don't promote A/B winners by reply rate
Every other outbound tool ships an A/B winner-picker that promotes the variant with the higher reply rate. The trap: reply rate collapses positive replies, negative replies, and unsubscribes into one number. The 'winning' variant is often the one with more 'stop emailing me'. Here's what we built instead.
-
Replacing a fixed sequence with an agentic loop
Notes from rebuilding Overwise's outreach pipeline as an agentic system: a Sonnet executor that decides per-lead what to send, an Opus advisor for edge cases, and a MessageVerifier that discards drafts whose claims can't be cited. What we threw away, what we replaced it with, and why.
More posts soon. If there's something specific you'd want us to write about, email tobias@overwise.ai.