The design. An illustrative worked example, not a client result. The 8-rep series a B2B sales team and every number here are assumptions. The design is a research-plus-drafting layer that pre-populates draft outbound emails from each rep's call list. Reps edit and send.
Assume the team is 8 account executives who each spend the first hour of the day researching prospects and writing cold emails from scratch. With this design, that hour becomes mostly editing drafts that already cite the prospect's recent funding, hiring, or product news.
What matters to a revenue lead is reply rate and rep time, and this example makes no claim about either. A team in this position would record both before the change and compare after. Volume stays flat at a human range of 30 to 80 prospects per rep, which is an assumption. This is not a volume play.
#Can AI outreach lift reply rate without hurting deliverability?
Does using AI to draft outbound hurt deliverability?
It can, if you use it to send more. It need not, if volume stays the same. In this design each rep stays at a human range of 30 to 80 prospects, sending from their own warmed mailboxes. The AI changes what goes in each email, not how many are sent, so there is no reason for spam complaints to rise.
The deliverability fear is real. Blast 500 near-identical emails a day from a fresh domain and Google and Microsoft will route you to spam fast. So the design does the opposite. Same human-range volume, same warmed mailboxes the reps already use in HubSpot, just better-researched copy per send.
#What does the design consist of?
Three pieces, none of them exotic. A research step that pulls each prospect's recent signals. A drafting step that writes a first-person email referencing one of those signals. And a sync that drops the draft into the rep's HubSpot sequence as an editable task.
- Research: pulls funding, headcount, and job-post signals for each account on tomorrow's call list, using Clay enrichment plus a Claude pass to pick the single most relevant hook.
- Drafting: writes one short email per prospect in the rep's voice, citing that hook. No merge-tag soup.
- Sync: writes the draft back to HubSpot as a task the rep reviews by 9am. Nothing sends automatically.
We are
A research-and-draft layer that hands each rep a better first version to edit. A human reads and sends every email.
We aren't
A high-volume spam blaster that auto-sends thousands of templated emails and burns your domain reputation to hit a send quota.
That human-in-the-loop step is the whole point. Reps read every draft and rewrite or reject the ones with weak hooks. The ones they keep should land better because the hook is specific, which is the claim to test, not to assume. This is the same principle behind our AI implementation work: the system drafts, the operator decides.
#How would you know it works?
Not on day one, and anyone who says otherwise is selling. In a design like this the research step can surface weak hooks early, and reps stop trusting the drafts. The first tuning job is deciding which signals count and cutting the ones that produce generic openers.
The lift, if it comes, shows up once the hooks are specific enough that reps trust them. Edit time drops and reply rate is the number to watch. If a sales lead tells you AI outreach pays off immediately, ask to see the data from the start of the rollout.
The risky part is the start. Reps who stop trusting the drafts stop using them. The fix is tuning the hooks, not rewriting every draft by hand.
We deliver layers like this two ways. Sometimes we build the pipeline and hand it over. Sometimes we sit inside the revenue team and tune it against live reply data. More write-ups like this live in the journal.
If you run a sales or revenue team and you're weighing whether AI can lift reply rate without torching deliverability, the honest answer is: yes, if you keep volume human and keep a person on every send. That's the version we build. tell us what you're building.
