In the race to automate everything with AI, few ideas captured investor imagination like the AI Sales Development Representative. A robotic SDR that could send personalized emails, qualify leads, and book meetings—what’s not to love?

Startups like Artisan, 11x, and dozens of others sprinted to market with bold promises: eliminate headcount, flood your pipeline, and scale outbound without limits. The pitch worked—at least for investors. Artisan raised $25 million. 11x scored a $50 million Series B led by Andreessen Horowitz. And hundreds of small and mid-sized businesses jumped on board.

Then came the churn.

As it turns out, AI SDRs were always building on shaky ground. The fundamental flaws weren’t just about hallucinations or bad prompts—they were structural, tied to the nature of outbound sales itself. In contrast, AI agents in customer service quietly gained traction, solving real business problems with deterministic outcomes and measurable value.

The divergence is stark—and it signals where AI agents are going next: B2B commerce automation.

AI SDRs: Great Pitch, Weak Foundation

The promise was seductive: outbound email is repetitive, tedious, and time-consuming. Why not hand it off to an LLM?

The problem is, outbound sales is also ambiguous, high-variance, and downstream from the result it seeks to drive. It’s not just about sending emails—it’s about knowing who to target, what to say, when to say it, and then nurturing the lead through an unpredictable process. Even seasoned human SDRs struggle with these nuances.

Startups like 11x and Artisan learned this the hard way.

“We had extremely bad hallucinations when we first launched,” said Artisan CEO Jaspar Carmichael-Jack. “I just cringe in pain” looking back at their early email output.

Customers churned. Some expected AI to replace their entire sales team. Others got flooded with low-quality leads that required more human cleanup than before. In some cases, the AI didn’t generate leads at all. At 11x, churn reached an estimated 70–80% of customers within months.

The AI wasn’t just inaccurate. It was also poorly aligned with how outbound actually works.

Outbound sales lacks structure. There's no definitive catalog of problems to solve, no deterministic system of record to query. Context is often missing or outdated. And most importantly, success is lagging and hard to measure—a meeting booked does not mean revenue closed.

As a result, many AI SDR systems were optimized for superficial signals (opens, replies) without delivering meaningful outcomes (qualified meetings, closed deals).

Tech That Doesn’t Stick

One key issue with AI SDRs is that they operate in an open-ended, generative environment—without reliable grounding data.

In customer service, AI agents like those built by Sierra can anchor their reasoning in product policies, CRM data, and predefined business rules. They know the return window. They can access the warranty terms. They’re connected to order systems and governed by guardrails.

AI SDRs, on the other hand, rely on scraped LinkedIn bios, fundraising news, and marketing sites to guess their way into a conversation. With no backend grounding and little feedback loop on success, these systems were built to sound good—not to be effective.

This lack of grounding led to inflated expectations. Some companies even falsified customer logos and ARR to appear more successful than they were. TechCrunch reporting revealed that 11x claimed companies like ZoomInfo and Airtable as customers, despite only offering short, failed trials.

Without a closed-loop system to verify impact, the illusion of traction was easy to manufacture—and even easier to unravel.

Why Customer Service Was the Right First Frontier

In contrast, AI agents in customer service are thriving—and it’s not by accident.

Customer service is the ideal first application for autonomous agents. The workflows are well-defined. The stakes are clear. The expected outcomes—ticket resolved, refund processed, shipment tracked—are measurable and discrete.

Sierra, one of the fastest-growing companies in this space, demonstrates what it takes to succeed:

• Deterministic architecture: Sierra’s AgentOS enables developers to set precise business rules—like ensuring an agent can’t process returns after 30 days. This avoids hallucinations and enforces compliance.

• System integration: Sierra’s agents plug into enterprise software stacks to pull real-time data and execute tasks like canceling subscriptions or modifying orders.

• Guardrails and observability: Through their Experience Manager and testing frameworks, Sierra ensures every agent interaction is logged, monitored, and continuously improved.

• Omnichannel deployment: Their agents operate over text, chat, voice—and eventually, avatars—making them adaptable to any support channel.

The result? CSAT scores up. Ticket resolution times down. Cost per interaction cut by 10x.

More importantly, these agents actually do the work—not just draft content, but take real actions on behalf of customers. They complete workflows end-to-end with a success rate exceeding 90% in some cases.

The Next Frontier: B2B Commerce Automation

The AI SDR market tried to solve the wrong problem at the wrong time. But the right problem may finally be coming into view: automating operational workflows in B2B commerce.

Much like customer support, B2B sales operations are plagued by slow, repetitive, rules-driven work. Every day, manufacturers, distributors, and suppliers waste hours handling:

• Inbound RFQs and product inquiries via email

• Bill of Materials (BOM) parsing

• Quote creation from spreadsheets or ERP systems

• Manual discounting and approval workflows

• Purchase order tracking and payment follow-ups

These are deterministic workflows, not creative ones. They follow rules, reference internal systems, and repeat endlessly. They’re ideal candidates for automation—but they’ve been ignored by most of the flashy AI startups.

Until now.

AI agents in this domain can function as autonomous inside sales coordinators—responding to inbound emails, building quotes, syncing to ERP and CRM systems, and looping in a human only when judgment or escalation is required. They don’t hallucinate because they don’t need to guess. They just follow the logic of the business.

And unlike AI SDRs, the success of B2B commerce agents is obvious: Did they complete the quote? Was the margin preserved? Did the customer place an order?

Why This Shift Matters

What separates successful AI deployments from failed ones is not the model—it’s the workflow. AI agents aren’t magic; they’re task-specific tools. The closer that task is to structured reasoning, defined rules, and measurable outcomes, the more effective the agent.

Outbound sales is inherently ambiguous, competitive, and hard to measure. It’s a poor match for first-generation AI. Customer service, by contrast, is structured, grounded, and feedback-rich—making it fertile ground for early success.

B2B commerce shares those same characteristics. And just like customer support, it’s a massive untapped labor cost buried under email threads, spreadsheets, and slow approvals. That’s the next big hill for AI agents to climb—and this time, the ground is solid.

Final Takeaway

AI SDRs weren’t a bad idea—they were just premature. They promised too much, too soon, in a domain too complex to automate effectively.

Meanwhile, AI customer service agents quietly demonstrated that structured, deterministic workflows are where AI can truly shine.

The next wave of agentic automation will focus less on pitching and more on performing—less on writing clever cold emails, more on doing real work in sales ops, quoting, and order management.

If AI is the future of work, then the lesson from the AI SDR wave is simple: know which jobs are ready. And start there.