The Pre-AI Infrastructure Most Companies Ignore Before Deploying AI SDRs

Everyone’s talking about AI SDRs like they’re magic bullets. They’re not. They’re amplifiers. And if your sales operations are a mess, AI will just amplify the mess faster and at scale.

Our data at PainSignal tracks 2,586 operational problems across 92 industries. When we look at the Workflow Automation category—where AI SDRs live—we see 18 distinct problems with an average severity score of 3.4 out of 5. The most common ones? ‘Disconnected data sources prevent effective automation,’ ‘poor CRM data quality undermines AI training,’ and ‘manual follow-up tasks are tedious and often skipped.’ These aren’t AI problems. They’re pre-AI infrastructure problems. And they’re what kills most AI SDR rollouts before the first email gets sent.

Jason Lemkin over at SaaStr recently shared a detailed piece on what they learned deploying 20+ AI agents. It’s packed with tactical advice—segment ruthlessly, read every message early, budget real ramp time. Solid stuff. But it assumes you’ve already done the hard work of fixing your sales fundamentals. Our data suggests most companies haven’t.

Take his point about the tool only being 20% of the outcome. He’s right that training, ownership, and process matter enormously. But our data challenges that 20% figure for companies without SaaStr’s resources. We track problems like ‘choosing the wrong automation tool wastes months of effort’ and ‘poor tool integration creates more work than it saves.’ For a small team or an indie hacker, tool selection isn’t a minor detail—it’s make-or-break. Pick something that doesn’t integrate with your CRM, or that requires a PhD to configure, and you’re sunk before you start. The 80/20 split might hold for a well-oiled SaaS company, but for everyone else, the tool is a bigger piece of the puzzle.

Where Lemkin’s advice really resonates is on the ‘work humans refuse to do.’ Our data backs this up hard. Those 18 Workflow Automation problems include exactly the tedious, repetitive tasks that get skipped: follow-up sequences that fall through the cracks, inbound leads that come in at 11 PM, re-engagement of churned customers. These aren’t glamorous, but they drive revenue. And they’re perfect for AI. We’ve seen 1,265 app ideas in our database targeting these gaps—everything from automated lead responders to AI-driven re-engagement bots. The opportunity is real, but you have to identify those tasks first. Most companies don’t even have them documented.

Another area where our data diverges: communication channels. Lemkin says 85% of prospects prefer chat, 15% voice. That’s based on SaaStr’s experience, which is valuable but not universal. Across our 92 industries, preferences vary wildly. In some B2B sectors—like manufacturing or logistics—phone calls still dominate because buyers are on the road or in noisy environments. In regulated industries like healthcare or finance, email might be the only compliant channel. Blindly following a ‘chat first’ rule could mean missing your best prospects. You need to know your industry’s norms, not just SaaStr’s.

And that’s the bigger point the article misses: industry context. SaaStr sells to SaaS founders and operators. Their playbook works for warm audiences like conference attendees. But what about a construction company trying to automate bid follow-ups? Or a restaurant owner re-engaging catering leads? The compliance requirements, sales cycles, and buyer expectations are completely different. Our data shows problems like ‘AI sales communications violating privacy regulations’ and ‘lack of transparency in AI decision-making’ cropping up in regulated verticals. Deploy an AI SDR without considering GDPR or CCPA, and you’re risking fines, not just low response rates.

So what should you actually do? Start with the fundamentals Lemkin mentions—proven messaging, clear ICP, compelling offer—but dig deeper. Our data suggests three pre-AI steps most companies skip:

1. Audit your data infrastructure. If your CRM is a graveyard of stale contacts and incomplete fields, AI training will fail. We track this as a high-severity problem across industries. Clean your data first. Integrate your sources. Without this, AI has nothing to work with.
2. Document your actual sales process, not the ideal one. Not what the playbook says, but what your best rep actually does. As Lemkin notes, clone what works. But our data shows most companies haven’t even mapped this out. Use tools like our Workflow Automation problems to identify gaps—like missed follow-ups or manual data entry—that AI could fix.
3. Assess compliance and ethical risks. Especially if you’re in a regulated industry. Our data includes growing concerns about AI and privacy. Build transparency and opt-out mechanisms into your AI SDR from day one.

Only then should you think about tools and training. Because AI scales what exists. If your fundamentals are broken, AI will break them faster. The companies seeing real results—like SaaStr’s 50%+ pipeline from AI agents—are the ones who did this groundwork first.

For builders and indie hackers, this is an opportunity. The market isn’t just for AI SDR tools themselves. It’s for the pre-AI infrastructure: data cleaners, process mappers, compliance checkers. We have app ideas in our database targeting exactly these gaps. If you’re looking to build, start there—where the pain is sharpest and most ignored.

Explore more operational problems and ideas at PainSignal.