[AI SDRs in Action] How AiSDR Uses Its AI SDR

AiSDR runs its own outbound on AiSDR, and we publish the results of every experiment we run. This isn’t a case study or a customer story. It’s our own campaigns and our own numbers.

It’s a fair thing to ask any AI SDR vendor: “Do you use your own AI SDR for your outreach?” The easiest way to answer that question is to show the numbers, including the ones we’d rather not show.

Plenty of these experiments have contradicted the assumptions we started with, and we publish those write-ups exactly the same way we publish the wins.

What “AI SDRs in Action” is

AI SDRs in Action is an ongoing series documenting the outbound experiments we run across our own and client campaigns. Each completed experiment becomes a single post, published whenever the test finishes, with no fixed schedule and no predetermined count.

Tests fall into copy, structural, prospecting, and channel categories, and you can see what’s running, what’s queued, and what’s finished on the live experiment tracker. 

Every experiment in the series follows the same rules, regardless of category or campaign size:

• Only 1 variable changes per test.
• Both sets use matched ICPs.
• The primary metric is set before the test starts.
• Each test runs for a minimum of 4 weeks.
• Sources are our own campaigns plus anonymized client campaigns.

We publish losses the same way we publish wins, which is the entire point of the series and the reason the evidence here stays first-party. Market-wide survey data lives in our industry report.

[Report] State of AI SDR Industry 2026

88% of AI pilots stall before anyone sees value

Is AI worth it? Find out where adoption is taking off, where teams stumble, and how human + AI is rewriting the rules of sales development.

DOWNLOAD MY COPY

Which experiments have we done so far?

Here’s a rundown of the outreach experiments we’ve done up to now:

• Do AI memes get more replies in cold emails?
• How AI personalization beat our cold email templates
• How customer-owned domains slow email launches by up to 7 months

What have we learned from our outreach experiments?

These are the results that changed how we run outbound, measured directly on our own campaigns.

• Replies and pipeline move separately: Our memes test returned a 5.39% reply rate with memes against 5.32% without, while meetings per engaged lead stayed level at 0.625% and 0.718%.
• Owning the sending domain costs speed: Customer-owned domains added 1.5 to 7 months of delay against a 7-day managed baseline, and a single setup delivered only 2,557 emails across 9 months.
• Personalization pays off unevenly: Across campaigns of roughly 300 leads each, personalized emails outperformed templates on cold audiences in most markets, though Germany came out level.

The metric you choose decides the answer

Name the metric before the test starts, or the test will hand you whichever answer looks most flattering afterward.

We learned this with our memes test. For months we’d been dropping an AI-generated meme into the final email of a sequence, and the entire team expected it to win. The meme set generated a 5.39% reply rate against 5.32% for plain text, while meetings per engaged lead landed at 0.625% and 0.718%.

Both look healthy against industry benchmarks, which puts the platform-wide cold email reply rate at 3.43% and the top quartile at 5.5%. Strong replies produced zero movement on pipeline. If replies had been our primary metric, memes would have shipped everywhere.

The reverse case showed up in the personalization test, where personalized emails produced more of a reply we now think of as “no to the demo, yes to the reply.” 

The lead turns down the call and answers anyway, usually referencing the specific detail we mentioned. On a simple positive-or-negative count that reply registers as a loss, but it keeps a real person in play, so we track it as a separate outcome.

A tactic that shifts a single metric without moving the other is noise, and you only recognize the difference if you named the metric beforehand.

Control costs more than it saves

More control over your sending infrastructure usually buys less outbound, and the tradeoff is considerably bigger than teams expect.

We compared campaigns on AiSDR-managed sending domains against campaigns on customer-owned domains. 

Owned-domain launches ran 1.5 to 7 months behind the 7-day managed baseline. Almost none of that delay went into campaign setup, because internal IT, legal, compliance, and DNS approvals consumed it.

The most constrained setup delivered 2,557 emails across 9 months, roughly what a managed setup sends in a single month. At that volume you can’t meaningfully test sequences or audiences, and you can’t determine whether a weak offer or an undersized sample is the real problem.

The trust payoff we expected never materialized either. 

Leads still asked “What’s your website?” even when the email arrived from a corporate domain, because buyers rarely inspect the sender domain closely. They reply, ask, and form their judgment from there. Domain choice still matters for deliverability, though it mattered far less for credibility than we assumed.

So the recommendation is now hybrid. Launch on managed domains, and then build owned domains in parallel across the following 60 to 90 days. Campaigns go live immediately, and the slower domain work stops blocking the learning loop. Similar tradeoffs show up throughout our customer case studies.

Effort does not transfer across contexts

What wins in a given market or with a given persona won’t automatically win in the next one.

Across cold audiences in most markets, personalized outreach outperformed templated sends, and the gap wasn’t subtle. A cold email needs to demonstrate that it wasn’t distributed to a random list, and referencing something real accomplished that better than a merge field did.

Germany broke the pattern entirely. 

German leads replied to long, structured templates at roughly the same rate as they did to personalized emails. Our working hypothesis is cultural, because in markets where formal business communication is the default, a structured template may register as professional while a short personalized note may look like the lower-effort option.

This remains a testable idea rather than a conclusion, and we’ve already queued follow-ups in other formal-business markets.

Persona depth behaves the same way. Founders and CEOs respond to sharper signals rather than longer personalization, while sales leaders, marketing leaders, and individual contributors get considerably more value from added context. Identical tactic, different payoff, depending entirely on who’s reading.

How to run these on your own campaigns

The setup is simple enough to copy, and the discipline required to follow it is the difficult part.

1. Change 1 variable and hold everything else steady.
2. Split matched audiences across both sets.
3. Set the primary metric before you send anything.
4. Run for a minimum of 4 weeks.

The trap we keep hitting is selecting the metric that moves easily instead of the metric that matters. Reply rate is quick to read and quick to improve, while meetings, show rates, and pipeline take considerably longer to register. Those slower numbers determine whether the change was worth making, so choose them deliberately.

What we’re testing next

The queue keeps filling from the same categories: copy tests on message structure and framing, structural tests on sequence design, prospecting tests on targeting and signals, and channel tests on where outreach lands.

We won’t preview specific tests here, because the queue shifts continually as results arrive and new questions surface. The tracker carries the current state.

What won’t change is the publishing rule. Every completed test receives a write-up, whether it confirmed our expectations or dismantled them entirely. You can follow all of it through our outreach experiments.