Why AI SDRs take 2 weeks to deploy

SaaStr ran AI SDRs across sales, marketing, and customer success through 2025 and into 2026. They now have enough deployment data to say something useful: most vendor timelines are wrong, and most buyers are wrong about which channel their prospects actually prefer. Both findings matter if you're about to sign a contract for an AI sales agent.

The 2-week floor is not a vendor problem

When an AI SDR vendor tells you their product deploys in a day or two, they are technically correct about one thing: the software can be running in a day or two. What they're not counting is everything that has to happen before the software can do anything useful.

Start with email infrastructure. If you're using dedicated IPs, domains, or email addresses for outbound, you're looking at 2-3 weeks of warm-up time just to avoid spam filters. That's not the vendor being slow. That's how email infrastructure has worked for years, and no AI wrapper changes it. Domain reputation is earned through volume and time, not through a better product.

Before warm-up even starts, there's a preparation phase most vendors undercount in their pitch decks. You need to figure out what copy works for your segments. You need to test subject lines. You need to test send times. You need to hyper-segment your list and then multiply every copy variant across every segment. That's not a day's work. SaaStr's experience puts it at about 2 weeks of work, done before a single agent is live.

Then comes the integration layer. Does this AI SDR flow back into your CRM, or does it live in a silo? Once the outbound agent is up, do you add an inbound one? What about customer success handoffs? These questions don't resolve themselves during deployment -- they need answers before deployment, because the architectural decisions you make early are hard to undo.

The implementation itself is rarely what consumes the calendar. The mental overhead of making all those decisions upfront is what does.

SaaStr's fastest deployment was Monaco, which they describe as genuinely good at keeping pipeline up to date and refilling itself. It still took a week and a half to get fully running. That's the floor on a good deployment with a good vendor.

If a vendor is telling you their AI SDR is live today, be skeptical. The prep work that determines whether the agent succeeds or fails is prep work only your team can do. No amount of forward-deployed engineering support changes that.

The daily check-in that nobody budgets for

One expectation worth resetting before you deploy: AI SDRs are not set-and-forget tools, regardless of how the demo presented them.

SaaStr found that even with a forward-deployed engineer doing heavy lifting, someone internal needed to be in the back end every day. Not hours every day -- sometimes 15 minutes, a fast check at the start or end of a workday. Think of it as a compressed one-on-one with a team member: quick, specific, focused on what went sideways overnight.

This matters for planning. If your team is already at capacity, an AI SDR doesn't reduce that capacity -- it adds a lightweight but real daily task. Budget for it before you sign, not after you're wondering why the agent is underperforming.

The operators we work with who get the most out of LinkedIn engagement face a similar dynamic: the tools can surface the right posts and draft sharp comments, but someone still needs to check the output and stay in the loop. Automation compresses the time commitment; it doesn't eliminate judgment. That's as true for AI SDRs as it is for building inbound through LinkedIn comments.

Most people still prefer chat

SaaStr is one of the few B2B organizations running a genuinely multimodal AI agent setup: chat, voice, and video, live simultaneously, with real volume behind each channel. Amelia AI has been live since November with chat and video options. Digital Jason on Delphi has been running in chat and voice for close to a year, with over 2.75 million conversations.

After that much usage, the data surprised them: most prospects and customers prefer chat. Not voice. Not video. Chat.

Part of it is comfort. During a speaker call, someone pulled up SaaStr.ai mid-conversation, saw Amelia AI appear, and immediately reached for the chat box -- even though a video option was right there. The preference wasn't about capability; it was about habit and control.

Some people prefer video because they don't have to type and it feels more like talking to someone. A handful of attendees at SaaStr's London event mentioned they'd interacted with Amelia AI before showing up, which is exactly the point of a 24/7 AI presence. But "some people prefer video" and "most people prefer chat" can both be true at the same time, which is the actual lesson.

Build optionality, not a single channel

The mistake most teams make isn't choosing the wrong channel. It's building only one and assuming preference will conform to what they built.

Forcing prospects into chat because you built chat first, or into voice because your vendor is voice-first, costs you the conversations that would have happened on a different modality. SaaStr's approach is explicit: if someone wants to chat with Amelia AI, they chat. If they want video, they get video. Text works too. The agent adapts; the human doesn't have to.

For B2B founders thinking about this in the context of their own sales motion, the same logic applies to where you show up on LinkedIn. The founders who build the most durable inbound aren't forcing their buyers to find them on one specific type of content. They're in comment sections, in posts, in DMs, visible across multiple contexts. Buyers discover them through whatever they happen to be reading. That breadth is what LinkedIn's top creators have in common when you look at the actual engagement data.

The channel mix question matters for AI SDRs for a concrete reason: your segments probably don't share channel preferences. Enterprise buyers who've been in B2B sales for 20 years may prefer a methodical email thread. Younger buyers who grew up in Slack and Discord may chat with your AI agent the way they'd message a colleague. Building a single-modality agent and hoping it fits everyone is the same mistake as sending one cold email template to your entire list.

What to take into your next vendor call

Two things worth having clear before you get on the demo call.

First: ask the vendor to walk you through the preparation work your team will need to do before the agent goes live. If they skip past it or minimize it, that's a data point. The vendors who've actually deployed at scale will be specific about copy testing cycles, warm-up periods, and segmentation work. The vendors who haven't will tell you it's mostly handled.

Second: ask what the data shows about channel preference in your specific segment. If the vendor only has one modality, that's fine -- just go in knowing you'll need to build or buy optionality separately as your usage grows. The worst outcome is deploying a voice-first agent to a segment that overwhelmingly prefers chat, then attributing the underperformance to the technology rather than the channel mismatch.

The 2-week floor and the chat preference finding are both correctives to vendor narratives that have gotten ahead of what the deployment data actually shows. Both are useful to know before you commit budget, not after.

For founders who are building pipeline through content and engagement while the AI sales stack matures, the underlying dynamic is the same: the preparation work is the work. Whether it's warming a domain, segmenting a list, or showing up consistently in the right comment sections, the part that drives results is the part vendors and shortcuts can't do for you.