Two lessons we’ve learned the hard way running AI agents across sales, marketing, and customer success at SaaStr in 2026.
If you’re rolling out your first AI SDR, internalize these now. They will save you a lot of frustration.
Lesson #1: Budget at Least 2 Weeks of Ramp Time. Always.
Regardless of vendor. Regardless of whether they give you a forward-deployed engineer. Regardless of how slick the demo looked.
Two weeks is the floor. Here’s why.
Outbound agents have built-in warm-up periods. If you’re using dedicated IPs, domains, or email addresses, you’re looking at 2-3 weeks just for the warming process. That isn’t the vendor being slow. That’s how email infrastructure works.
Nothing is set-and-forget. Some vendors are guilty of pitching it that way. They shouldn’t. Even with a forward-deployed engineer doing the heavy lifting, you still need to be in the back end every day. Some days you’ll have more time, some days less. But every single day, even if it’s just a 15-minute speedrun at the start or end of the day, you should be checking on the agents. It’s basically a short one-on-one with a human counterpart, just compressed.
The prep work alone takes 2 weeks. Before you even deploy, you need to:
- Figure out what copy works
- Test subject lines
- Test time of day
- Hyper-segment your base
- Multiply all of the above by each segment
That’s 2 weeks of work right there, before any agent is actually running.
Then comes the integration question. Do you want this AI SDR flowing back into Salesforce? Does it live in a silo? Once you’ve deployed an outbound agent, do you add an inbound one? What about customer success? These questions snowball, and they all need answers.
The implementation itself isn’t actually the longest part. The mental overhead of setting all this up is what eats your calendar.
Even Monaco, which is genuinely great at refilling itself and keeping the pipeline up to date, took us about a week and a half to get fully set up. And that’s our fastest deployment.
If a vendor tells you their AI SDR is instant, “roll it out today,” be skeptical. Be very skeptical. A lot of the prep work is stuff only you can do, no matter how much help they offer on their end.
Lesson #2: Most People Still Prefer Chat Over Voice or Video
We’re one of the few B2B + AI companies running a multimodal voice agent alongside chat and email agents.
Amelia AI has been live since November with chat and video. Digital Jason on Delphi has been live in chat and voice for almost a year, with 2.75M+ conversations.
After running these agents for that long, here’s the surprise:
Most customers / prospects still prefer to chat. Not voice. Not video. Chat.
A lot of it is comfort. Just yesterday on a speaker call, the person on the other end pulled up SaaStr.ai while we were talking. Amelia AI popped up and they asked, “Wait, is this you?” Yes, that’s the AI version. They wanted to interact through chat, not through video, even though video was right there.
Some people do prefer video because they don’t have to type. They feel like they’re actually talking to someone. A handful of attendees at our London event last December came up and said they’d talked to Amelia AI before the event. That’s exactly the point. She’s there 24/7 to answer questions, route people, and be helpful in a way no human can match. No human is on 24/7.
The lesson: let people pick their format.
You don’t need to be chat-only. You don’t need to be voice-only. You don’t need to be video-only.
If they want to chat with Amelia AI, they can chat. If they want a video call with her, they can have one. If they just want a text-based conversation, that works too.
Forcing your audience down one channel because that’s the channel you built first is a mistake. Build the optionality. Let the user pick.
What This Means If You’re Deploying AI Agents
Two things to internalize before you sign that contract:
The ramp is real. Budget for it. Plan for it. Don’t believe anyone who tells you it’s instant. The vendors who tell you it’s instant are also the ones who will leave you holding the bag in week three when nothing’s working.
Channel preference is real too. Most of your audience still wants to chat with your agent, even when voice and video are available. Build all three. Don’t force any of them.
Pick a leading tool, budget the time, train it for real, have a strong human manage it. And success will come.
