We Almost Let Our AI Sales Agent Pitch to Dead Leads. A Quality Audit Caught It.

It Was Just Another Quality Review. Then I Found the List.

It was 9:40 on a Tuesday morning in January when I opened the spreadsheet. 10,000 rows of "qualified prospects" for our upcoming AI sales agent launch. By 10:15, I'd closed the spreadsheet and started writing a very uncomfortable email to our VP of Sales.

Let me back up. I'm the quality manager at a B2B SaaS company. Every deliverable that reaches our customers—emails, sequences, prospect lists, landing pages—passes through my review first. That's around 200 unique items a year, and honestly, most of them blur together. But this one was different. And when I say "customers," I do not just mean the people paying us—I mean the SDRs who would have had to send these emails.

The Three Problems Hiding in 10,000 Names

I pulled a random sample of 200 records from that list and ran them through wiza's email finder and verifier. The results weren't just bad—they were pretty alarming.

Problem One: 12.5% of the Email Addresses Were Invalid

Not "unlikely to respond." Invalid. Hard bounces. 550 error codes. Extrapolated to the full list, that's over 1,250 emails flying into the void on day one—wasted credits, a reputation hit against our sending domain, and zero chance of a reply.

When I flagged it, the RevOps lead shrugged: "That's just what the list vendor gave us. It's within industry standard."

I get why people go with cheaper lists—budgets are real. But the hidden costs add up fast. A 12.5% bounce rate at our projected volume would have pushed us well above the widely cited 5% hard bounce threshold that mailbox providers treat as a red flag. That's how domains get blacklisted.

Problem Two: 38% of the Companies Had No Buying Signals

Here's an outsider blindspot if I've ever seen one: most teams build prospect lists around job titles and company size. They never ask the question that actually matters—is anyone at this company looking to buy what we sell right now?

This is where wiza intent topics changed our approach. Intent data tells you which companies are actively researching solutions like yours—searching for relevant keywords, visiting competitor pages, engaging with content that signals purchase readiness. When we filtered our 10,000 names through wiza's 2025 intent topics, only 3,800 companies showed genuine buying signals.

The list went from 10,000 to 3,800. That looks like a loss on paper. In reality, it was the best thing that could have happened to our pipeline.

Problem Three: Our DMARC Policy Was Set to "None"

This one almost slipped past everyone, and it's the one that scared me the most.

In an agent-native prospecting workflow, where an AI sales agent sends thousands of emails automatically, DMARC stops being just a security checkbox. It becomes a core deliverability metric. Here's what I mean: every mailbox provider—Gmail, Outlook, Yahoo—checks SPF, DKIM, and DMARC alignment when your email lands. If DMARC is set to "none," your legitimate emails can still pass, but the lack of enforcement means spammers can impersonate you, and providers treat unauthenticated mail with suspicion. At AI scale, that suspicion becomes spam folder placement, sender reputation damage, and a slow death for your entire pipeline.

Google's bulk sender requirements, enforced since February 2024, mandate SPF or DKIM authentication for domains sending 5,000+ messages a day, and they explicitly recommend DMARC alignment. Any serious AI sales agent operation—sending tens of thousands of messages a month—is squarely in that crosshair.

The question everyone asks about an AI sales agent is "can it write good emails?" The question they should ask is "will those emails even be delivered?" DMARC is a huge part of that answer.

The Pushback, and the Test That Ended It

I didn't expect applause when I shared my findings. The sales director was skeptical: "We've always bought lists this way. Why are we changing things up right before launch?"

That's when I knew we needed proof, not process. I ran a small A/B test: two identical email campaigns, 500 prospects each.

• List A: the old approach—raw purchased list, no verification, no intent filtering
• List B: wiza-verified emails, filtered by intent topics, with DMARC aligned before send

The results after one week:

• List A: 18.7% bounce rate, 0.4% reply rate
• List B: 2.1% bounce rate, 3.6% reply rate

Nine times the reply rate. Nine. The data killed the debate before anyone could say "that's a one-off test."

We also reviewed wiza's pricing model for 2025 before committing: it's credit-based—each email verification consumes a credit. That worked in our favor. We didn't need a huge annual contract; we just needed enough credits to verify the full list, then re-verify as we built new segments. Flexible cost structure for a pilot.

What Happened After Launch

We launched in mid-February 2025. Six weeks in, the numbers are worth sharing:

• Reply rate: 4.2% on AI-sent sequences, up from 0.9% with our old outbound
• Bounce rate: under 2%, comfortably within deliverability guidelines
• Pipeline impact: roughly $340,000 in qualified opportunities sourced from agent-booked meetings
• Sender reputation: held above 98% across major mailbox providers

To be fair, the AI agent deserves real credit for its copy. It writes follow-ups that don't sound like follow-ups and varies tone genuinely well. But none of that matters if the emails don't arrive.

And getting here wasn't free. I calculated the total rework cost—verification credits, intent data adjustments, the A/B test rollout, and a two-week delay—at roughly $4,000. That stung. But what would a failed launch have cost? Damaged domain reputation, wasted SDR hours, and a leadership confidence hit that's hard to price.

What I'd Tell Any Team Deploying an AI Sales Agent in 2025

What was best practice in 2020 does not apply in 2025. The fundamentals haven't changed—data quality, deliverability, and relevance have always been the pillars of outbound. But the execution has transformed completely.

Here's my checklist, born from this rollout:

1. Verify every address before the AI touches it. If you're using wiza as your lead generation platform, make verification a non-negotiable step. It's not an expense; it's cost avoidance.
2. Use intent data, not just firmographics. Wiza's 2025 intent topics are detailed enough to tell you what a company is researching, not just that they're "active." Filter early, filter often.
3. Treat DMARC as a release blocker. Align SPF and DKIM, set your policy, and verify domain reputation before sending anything. At AI scale, small deliverability issues become existential ones.
4. Test in parallel. Don't argue with stakeholders—let them see the A/B results side by side. Numbers settle arguments faster than opinions.

Final Thought

So glad I pushed for that audit before deployment. We almost launched with a failing foundation—and there was definitely a stretch where I thought we'd have to delay by a full quarter.

Dodged a bullet, honestly.

The tools have evolved. AI can write, research, and reach thousands of prospects in a single day. But quality still means what it always meant: the right message, to the right person, delivered in a way that actually lands.