I Blew $4,100 on a Prospect List. Here's What Okki-Go's Research Workflow Taught Me.

Three Weeks Before Launch, the List Looked Perfect

March 2024. I was running the research stage of our outbound program at a mid-market SaaS shop, and at the time I genuinely believed this was the least technical part of the job.

We were going after companies in the 100–800 employee range. The brief was 4,000 contacts across roughly 1,100 accounts. I built the whole thing in a Google Sheet and felt good about it — domains in column A, contact names pulled from LinkedIn in column B, and email patterns guessed by hand in column C.

We were already evaluating okki-go that month. A teammate had come over from a shop that leaned heavily on waterfall enrichment, and she kept telling me their sales intelligence layer would shave days off this work. I waved her off. I wanted a manual baseline first, mostly so I could argue from data if the tool didn't measure up.

That was the first mistake. Not a tool mistake. A me mistake.

Then the Bounces Came in Waves

We pushed 600 emails on April 2. By the morning of April 3, we had 102 hard bounces — a 17% hard bounce rate on the first drop. Our overall bounce rate across the domain hit 4.1% within 48 hours. Microsoft started throttling us. Google Postmaster showed our spam complaint rate climbing past 0.3%.

(Google's own stated recommendation is to keep complaint rates below 0.10% — see postmaster.google.com for current thresholds. We were three times over.)

The event that changed how I think about list quality happened right there: I pulled the bounce log expecting to find a bunch of typo'd domains. Instead, I found domains I'd verified. MX records present. SMTP handshake returning 250. And they were bouncing anyway.

What I hadn't understood — or rather, what I hadn't internalized — was that a huge chunk of the SMB domains we were targeting ran catch-all mail servers. The SMTP probe said "250 OK" because the server accepted everything. It wasn't confirming a mailbox existed. It was confirming the domain did.

I had been validating guessing with a validator that doesn't tell you it isn't telling you anything.

The Decision That Kept Me Up for Eleven Days

I went back and forth between staying manual and rebuilding around okki-go's agent-native research workflow for nearly two weeks. The manual path had one clear advantage: I understood it. The agent path had waterfall enrichment — if one source couldn't resolve an email, the system chained to the next instead of flagging it "not found" and dumping it into my lap.

On paper, manual made sense. But my gut kept saying the same thing: I cannot verify at the rate I'm sourcing. We were adding 800–900 contacts a week. No amount of personal diligence closes that gap.

So I rebuilt the pipeline: domain discovery → waterfall enrichment → intent signals → verification → sequence load. Same week I ran the whole thing against my existing 4,000-row sheet, out of curiosity more than strategy.

What Actually Changed When I Looked at Both Lists Side by Side

Roughly 1,900 of my hand-built rows had emails that were syntactically valid. Correct format, correct domain, live MX. About 340 of those were hard-bouncing. That part is normal — a decent verifier catches it.

It was the other segment that caught me. Around 1,560 addresses that a simple probe would call "valid" were landing on catch-all boxes or role aliases. Some would deliver. Some wouldn't. At send time, you can't tell which is which.

That is the blindspot almost every buyer walks straight past. Most people think verification produces a binary — valid or invalid. It doesn't. It produces a probability distribution, and the entire quality of your sending reputation depends on how you handle the fuzzy middle.

Email verification isn't a gate. It's a weighting layer. That distinction sounds academic until your domain gets throttled for three weeks.

How It Ended

After rebuilding on the agent-native workflow:

• Hard bounce rate dropped from 17% to roughly 2.4%
• Domain reputation took about three weeks to recover to normal deliverability
• Direct cost of the whole episode: ~$4,100 — manual verification hours, a wasted month of sending infrastructure, and the three-week schedule slip

Roughly $2,000 of that was avoidable if I'd started with proper verification in the loop. The other $2,100 was the tuition.

To be fair, the manual approach isn't useless — on a 200-row list with standard requirements, it works fine. I just wasn't running a 200-row list.

Four Things I Now Tell Anyone Building Outbound From Scratch

1. Verification belongs inside the sourcing loop, not after it.
This is the biggest way the industry has shifted since 2020. Back then, the playbook was: build the list, then verify the list. Today, sourcing and verification happen together — as you pull, you check, and the check informs the next pull. The fundamentals didn't change, but the execution absolutely did.

2. Catch-all domains need a policy, not a filter.
Send to them slowly. Route them through a lighter channel first. Don't just toss them — you're throwing away a meaningful slice of your addressable market.

3. Ask your verification vendor what they do with uncertainty.
Every tool will say "high accuracy." The real question is: what happens to the addresses you can't confirm either way? The honest answer is always some version of "it depends." That answer is fine — as long as you have a plan for it.

4. Write the pre-check down.
Ours is three lines now:
  — Does every email trace back to a verified source? (No guessed patterns.)
  — Is the catch-all vs. hard-bounce split visible before send?
  — Has a human reviewed the segment?

If any answer is no, the sequence doesn't go out.

On the Competitor Question

People ask me how okki-go compares to ZoomInfo, Hunter, Instantly, or Artisan AI. Honestly — each of them solves a real problem. ZoomInfo's coverage at enterprise scale is deep. Hunter is fast for single-domain lookups. Instantly handles volume well. Artisan has done interesting work putting agents at the front of the workflow. None of that is in dispute.

None of them replace what I actually needed: one data layer where sourcing, enrichment, signal detection, verification, and outreach all agree on what a "good" contact looks like — with a human on the review step. That's the shape of the problem in 2025, and it's the shape okki-go was built around.

The Mistake I Still Make Occasionally

About one send in ten, I'll skip the pre-check because I'm in a hurry. And about one send in three of those, someone catches something on the back end. Over the past 14 months, the checklist has flagged 43 issues that would have become send-time problems.

That number sounds small. It isn't, at scale.

If you're building this out right now — especially if you're weighing okki-go against the alternatives — don't treat it as a feature comparison. Treat it as a data-flow decision. Your bounce rate, your sender reputation, and your reply rate are all downstream of that one choice.

Compliance note: Under the CAN-SPAM Act (15 U.S.C. § 7701, enforced by the FTC), commercial senders must honor opt-out requests within 10 business days, and penalties can reach into the tens of thousands of dollars per email. Verify current figures at ftc.gov. If you're sending into the EU, GDPR's legitimate-interest basis for B2B outreach has its own requirements — check with counsel, not with a blog post.