Most of the AI SDR pitches you hear sound the same. A founder gets on stage, shows a slick demo, and promises the death of the BDR. Then you ask one hard question and the whole thing falls apart.
Jaspar Carmichael-Jack, founder and CEO of Artisan, did the opposite at SaaStr AI Annual. He ran a live demo, named his own customer numbers, told a story about a customer who got “terrible terrible” results for two months, and explained exactly why his product still won’t cold call. That kind of specificity is rare, and it’s worth breaking down because the patterns here apply to anyone building or buying AI agents for go-to-market.
Artisan builds what it calls AI employees. The first is Ava, an autonomous AI BDR. You’ve probably seen the billboards. “Stop Hiring Humans.” The marketing is loud on purpose. The product underneath is more interesting than the billboards suggest.
The real pitch is accountability, not automation
The standard outbound stack is a pile of tools taped together. A sequencer, a data tool, an enrichment tool, a copywriting tool, plus a GTM engineer to hold it all in place. Jaspar’s argument against this isn’t that it’s annoying to manage. It’s that you can’t see ROI.
When everything is fragmented, no single tool owns the outcome. You don’t know what’s driving your response rate, and the contract sits there generating nothing while you guess.
Artisan built the opposite. In the UI you see cost per lead, cost per meeting generated, and the underlying credit cost to hit each outcome. Every campaign rolls up to a single number: what you paid in credits to get each positive reply or booked meeting.
That framing matters more than the automation itself. It’s the difference between buying software and buying a result.
The proof: 7,000 emails, 3.6% positive response rate
The customer example Jaspar led with was SaaStr’s own outbound. Thousands of emails in a single six-week period, 7,000 in total, landing a 3.6% positive response rate and generating hundreds of thousands of dollars in revenue.
And that was all on Ava 1.0. The 2.0 numbers come from a separate campaign Jaspar ran himself, targeting YC founders in San Francisco. He coached the messaging to read human. Lowercase, a couple of exclamation points, pulling personalization from social posts and mutual connections. That campaign hit a 4% response rate with no timing optimization at all.
Both campaigns cleared the band where most cold outbound lives. The 4% figure is the one to sit with: it came from a campaign that ignored timing entirely and still beat the benchmark, which is the whole argument for getting the who and the what right before you worry about the when.
The takeaway he kept returning to: outbound comes down to three things, and you can win on two of them.
Who, what, and when
Who is the data. Artisan runs two databases. One holds 280 million B2B contacts with waterfall email and phone enrichment from a couple dozen providers. The other covers every local business globally with a Google Maps profile. But the best-performing data is your own. When customers push CRM fields or data warehouse records in via webhook, the personalization gets sharper because Ava can use prior activity and existing relationships.
What is the message. This is where Artisan spends most of its energy. Dozens of enrichment sources feed an aggregate profile of each prospect. Ava then picks the best personalization angle per message based on context, not in isolation. If a prospect is hiring a bunch of BDRs, that’s a relevant hook for Artisan but useless for a different seller. You can also spin up custom research agents that crawl the web for whatever signal you define, down to what a CFO said in a recent 10K.
When is timing. Artisan de-anonymizes website visitors at the person and company level, so you can reach out while someone is actively showing intent. Funding data, hiring data, and social engagement signals layer on top.
Jaspar’s point with the YC campaign was that you don’t need all three. Get the who and the what right and the when becomes optional.
What “autonomous” actually means in 2.0
The 1.0 product wrote and sent outbound. The 2.0 product closes the loop. Turn on response handling and meeting booking, and Ava goes back and forth with the lead, handles objections, and drops the meeting on a rep’s calendar with no human in the loop. You can set escalation rules to pull a human in, but you don’t have to.
The vision Jaspar described goes past consolidation. The goal is removing you from the workflow entirely. Ava suggests campaign ideas, proposes audiences, writes the messaging from your website, and responds to leads. A full self-driving mode is coming in the next couple of months.
Why Artisan won’t cold call
Here’s the part worth sitting with. Artisan does not do AI cold calling, and the reasoning is the most credible thing in the talk.
It’s illegal for outbound, and humans are still better at it. So instead of forcing it, Artisan queues calls for reps inside a native dialer, generates per-prospect talking points, and drafts the follow-up email. They handle everything except the actual conversation.
Jaspar put a timeline on the technology too. He thinks AI voice is a couple of years from matching a human on a real call. You can already hear the tells: longer latency, slightly wrong answers, missed nuance. He’s not interested in shipping into a category until Artisan can do it flawlessly, and with outbound calling he can never legally get there. So they don’t.
A founder telling you what their product deliberately won’t do is a stronger signal than one promising it does everything.
Pricing built around the outcome
Artisan moved to a credit model because the product now spans the full GTM stack. One credit is roughly two cents. Waterfall email enrichment costs two credits. Everything bundles into a cost per lead enrolled in a campaign, usually 30 to 60 cents, down to about 20 cents for a bare-bones three-email sequence.
Two structural choices stand out. Lead discovery and list building are free. You only pay for enrichment, personalization data, and execution. And the analytics are framed entirely around cost per positive outcome, not cost per email sent.
The go-to-market is value-first. Sign up, get $300 in free credits, no credit card, get results before anyone asks for money. In a space full of skepticism, that’s a deliberate trust play.
Outbound market fit
The best concept from the talk was “outbound market fit.” Product market fit does not guarantee that cold outbound works. You can have great customers and a real product and still get no replies, because the people you’re emailing don’t want it.
Cook Unity is the example. Their first two months were, in Jaspar’s words, terrible. They stuck it out, kept iterating on messaging and targeting, and now sit under $50 cost per meeting booked. About 90% of the time, iterating on the who, what, and when gets you to a recipe that works. The other 10% is a signal that cold isn’t your channel and you should run warm, CRM-driven outbound instead.
That’s a more honest framing than most vendors will give you. Sometimes the answer is that outbound isn’t the problem to solve.
5 Unexpected Learnings
1. They deleted 178 million contacts on purpose. Artisan started with a 450 million contact database and trimmed it to 272 million to strip out the low-quality records. Everyone in data assumes bigger is better. Artisan made the opposite bet: fewer contacts, all verified and bounce-tested, beats a bloated list. The instinct to hoard data is usually wrong.
2. Lead discovery is free, and that’s a strategy. Finding leads and building the list costs nothing. You only pay for enrichment and execution. It sounds like a giveaway, but it’s an incentive design. If lead discovery were metered, customers would pay even when the leads were bad. By making it free, Artisan only earns when you actually move a prospect into a campaign, which forces their interests to line up with yours.
3. The restraint is the product decision. Refusing to do AI cold calling, in a market where competitors already do it, is a real position. Most founders chase feature parity. Artisan is betting that not shipping a mediocre voice agent protects the brand more than checking the box would. Knowing what to leave out is its own moat.
4. The controversial marketing outperformed the safe marketing, and they tested it. Artisan ran less-controversial campaigns between the big “Stop Hiring Humans” pushes. The tamer ones just didn’t work as well. The loud, polarizing creative is not a personality quirk. It’s the measured result of attention arbitrage in a crowded category, and they have the data to back the choice.
5. PMF and outbound fit are different things. The Cook Unity story rewires how you read a failing outbound program. A bad first two months isn’t proof the tool is broken or that you lack product market fit. It’s the normal onboarding curve. But if iterating on who, what, and when still doesn’t move the number, the honest read is that cold outbound isn’t your channel, and no amount of AI fixes that.
