It started, as bad decisions often do, with a question from my CRO in a Monday stand-up: "Why is the SDR team still manually hunting for direct dials?"
The room went quiet. I didn't have a good answer.
That moment launched eight months of evaluating, piloting, and then cleaning up after AI SDR software. Total damage: about $14,000 in subscriptions and wasted engineering time, two dead CRM lists, and some credibility with sales leadership (which, honestly, I'm still rebuilding).
Quick context on who I am: I've handled sales technology and RevOps for a mid-size B2B SaaS company for six years. I've personally made—and documented—seven significant mistakes totaling roughly $32,000 in wasted budget. This wasn't my first AI tool rodeo. But it was my first AI SDR rodeo, and I made some classic errors. Now I maintain our team's checklist so nobody else repeats them.
How We Got Here: The Tempting Demo Trap
In early 2024, we had 12 SDRs and their outbound numbers were sliding. The usual remedies—new sequences, better personas—weren't moving the needle. The real problem was data. Our CRM enrichment had gone stale, and the outsourced list vendor we'd used since 2022 was far less accurate than their sales deck claimed.
From the outside, it looked like a cost problem: enrich more records, buy more credits, send more emails. The reality is that data quality decays, and if the data feeding your SDRs is bad, no amount of automation fixes it.
So when 11x.ai's account executive walked us through a demo of their AI SDR agent—autonomous prospecting, research, email writing, direct dial enrichment, LinkedIn outreach—I was genuinely impressed. We also looked at Artisan's AI SDR. Different UI, similar pitch: set a targeting rule, let the agent build lists and run outreach.
I went back and forth between the two for two weeks. 11x.ai offered deeper Salesforce-native workflows; Artisan had better-looking campaign reporting. On paper, they were close enough that I decided to pilot both for three weeks. That was the right call—but I evaluated the wrong things.
The Surface: What Demos Don't Show You
We ran both tools side by side on the same ICP and territory. I framed it as a "feature test," but the real deliverable I should've been watching was data accuracy.
Midway through the pilot, our SDR team noticed something odd. In one list of "VP Marketing at B2B SaaS companies," several contacts had obviously wrong title combinations. An AE (the nosiest one, who's actually good) discovered that a "VP Marketing" at a 200-person company was, in reality, a "Senior Manager of Brand" at a 40-person company.
Out of 100 contact records sampled across both tools:
- 11x.ai: 71/100 direct dials were plausibly accurate
- Artisan: 64/100
- Email deliverability: roughly 80% "delivered," but we couldn't verify it landed in the right inbox
(Note: these numbers are from our pilot in March 2026 on free trials. Things may have changed—which is exactly the "timestamps" problem I'll get to.)
There was also the direct dial question, which turned into its own saga.
The Direct Dial Saga: Accuracy Isn't Everything
Our SDRs were pushing hard for direct dial numbers. Once we had direct numbers, we didn't have to go through the switchboard, and that felt like an obvious win.
Here's something vendors won't tell you clearly: direct dial fields in CRM enrichment are often a mix of direct lines, main numbers, and abandoned VoIP numbers. Yes, they're technically "direct dial" numbers in the database. Whether they're still in service is a completely different story.
After the pilot, we bought 4,000 records credited as "verified direct dials." We didn't spot-check them thoroughly first. I even built a validation script to run on a sample, but got distracted by launch stuff (ugh).
The result: our SDRs wasted about 30% of their call time either hitting dead lines or reaching a front-desk person. One SDR, in a moment of pure frustration, logged a call note that read:
"Do I even want to know what this dataset's verification process looks like?"
So why did this happen? Because "verified" in a vendor's database can mean "verified at the time of sourcing" (e.g., 2022) or "verified syntactically" (a valid format). Neither means "confirmed to work now." For CRM enrichment, the only timestamp that matters is the last successful verification.
We later found that our CRM already had 3 of the 4,000 numbers flagged as "wrong number" in a previous campaign. The AI SDR agent happily re-imported them.
That was the moment I started taking the human-in-the-loop review mess seriously.
The Human-in-the-Loop Review: It's Not a Rubber Stamp
11x.ai does give you human-in-the-loop review options—you can set thresholds and review records in their inbox, or route certain steps to a human. I thought I understood this. I was wrong.
For our first big campaign, we set the review threshold at 10%. Then, to reduce friction, we gave the SDRs (internally nicknamed the "AI babysitting" job) a quick approve-all rule. The reasoning: "the AI is 90% accurate, right?"
Wrong.
In September 2024, the first real campaign ran on autopilot for a week. By Thursday, one of my SDRs flagged that the email sequences were going out with a phone number from the wrong country in the signature. Yes, the AI agent was supposed to sync contact info—but the destination field mapping was broken. Instead of adding the right phone number per region, it appended a random number from the record.
How many emails? 3,200. We had to send follow-up corrections ("sorry, wrong number in my signature"), which is the most awkward email I've ever had the privilege of writing.
Why didn't the human-in-the-loop review catch it? Because it was effectively disabled. We limited humans to reviewing outbound emails, and even that was at a 10% sample. The AI agent autonomously handled the research, enrichment, and sequence-joining. There was no human checking the data layer.
So what should revenue operations teams evaluate in CRM enrichment? Everything. But here's the five-item list I'd use now, in priority order:
1. Enrichment freshness and field-level timestamps
Ask for date fields: email_verified_on, direct_dial_verified_on. If the vendor doesn't have them, the data is suspect. A common industry rule (cited by multiple sales-data vendors, and roughly true based on our own testing) is that B2B contact data decays at about 2–3% per month. A record from 2023 is essentially useless for outbound.
2. Verification methodology
Was it verified via a simple email bounce check? Or via a direct connection to a corporate directory? This matters more than the claimed record count. Syntax checks are not verification.
3. How human-in-the-loop actually works
Can you set different thresholds for data tasks vs. messaging tasks? Can you view "low confidence" records? Does human review feed back into future AI output, or is it a rubber stamp? In 11x.ai's case, the review queues are genuinely useful—but only if you configure them properly.
4. How "direct dial" is defined and maintained
Does the vendor update numbers after a bounce? Do they re-verify on every enrichment, or are they recycling old numbers indefinitely? Ask for the actual update cadence, not just the aggregate accuracy number.
5. Integration and data hygiene boundaries between AI SDR and CRM
The cleanest enrichment in the world is useless if a mapping bug puts the phone number in the wrong field. That's the exact issue that caused our bad-number email signatures. Having advanced AI SDR features without a robust integration review is like having a jet without a pilot: the jet is great, but the pilot is what avoids the crash.
If you're set on 11x.ai, I'd still buy it. The autonomy and agent-native workflows are genuinely different. But the human governance has to exist.
The Honest Critique: AI SDR Software Isn't Magic
After we fixed the data layer, here's the truth: 11x.ai's AI sales automation did help. We cut SDR research time by about 40%. Our team could finally focus on talking to people instead of finding them. And the multichannel sequence automation—email, LinkedIn, direct dial outreach—is genuinely useful.
But it didn't replace the SDRs, and it didn't replace the data analyst's occasional glance at our enrichment. The "you'll never need a human in the loop" marketing that floats around these tools? I call it the zero-touch myth.
From the outside, autonomous AI SDR agents look like machines that just run. The reality: the machine stops being useful the moment your data and your review process aren't healthy.
Also, yes, you should compare Artisan AI sales software vs. 11x.ai if you're considering options. Both are fine. Both perform better with clean data, human review, and careful CRM mapping. The tool is maybe 30% of the equation; the other 70% is the surrounding workflow.
What I'd Do Differently (and What We Do Now)
If I could redo that decision in July 2024, I'd have spent two weeks cleaning up our CRM and outreach data stack before any AI SDR pilot. And I'd have insisted on a two-week manual validation phase for every output field.
If I'm being fair to my past self: the decision to pick an AI SDR wasn't irrational. The temptation to believe "this tool will make the SDR team unstoppable" is strong, and most senior people in your org will push you toward automation. Given what I knew then, my choice was reasonable. The execution was the problem.
Since January 2025, our team has run a "data integrity gate" on every AI SDR campaign:
- Manual review of 20 random contact records before launch (takes about 15 minutes)
- All AI-generated emails require human-in-the-loop review for the first 200 sends; then we lower it to 25%
- Direct dial fields get re-verified every 90 days
- Field-level mapping gets checked on each new CRM integration
We've caught 47 potential data errors using this gate in the past 18 months. Some were small (a wrong country code). Some were big (a list with completely incorrect titles).
Would I evaluate another AI SDR again? Yes, with better eyes. My advice to RevOps teams evaluating AI SDR tools: the software is good, the data ecosystem is the risk, and the human-in-the-loop is your insurance policy, not a cost center.
Or, as my nosy AE put it: "Dumb data in, AI-powered dumbness out—but faster."
(We did get the direct dial stuff figured out eventually. Some days, you just have to call the wrong numbers to appreciate the right ones.)
Pricing references in this article reflect public pricing as of April 2026. Verify current rates with the vendors before making a decision.