The January batch that broke my patience
January 2024. I opened our weekly outbound quality report, and the reject rate was sitting at 38.6%. Not a typo. Roughly four out of ten emails our six-person SDR team was about to send, I'd flagged as unfit to send.
I run quality and brand compliance for an 80-person B2B SaaS company. Part of my job is reviewing outbound deliverable before it hits a prospect's inbox—about 180 to 220 emails a month hit my desk directly, and our SDRs self-review the rest against a checklist I built in 2022. When I say I rejected 38.6% in one week, that's not a feeling. That's a spreadsheet column.
The failures weren't copy problems. The copy was fine. Grammatically tight. Reasonably personalized. The problem was the data underneath it.
Wrong titles. Stale employees. Personalization tokens pointing at people who'd left the company seven months ago. One email greeted a CTO who, according to LinkedIn, had been at a different company since spring 2023. We sent it anyway. That one stung.
I remember texting our RevOps lead at 8:40 p.m. on a Thursday: we can't keep shipping this.
The 'cheaper' vendor that cost us a quarter
Everything I'd read about fixing outbound data said the same thing: if your enrichment is bad, switch providers. Get better email verification. Add another waterfall layer. So that's what we did first—and honestly, that's what most teams do.
We moved off our $0.10-per-record contact vendor to a $0.04-per-record one. On paper, we cut our prospecting data spend by 60%. Finance was thrilled. I was cautiously optimistic.
Three weeks in, the reject rate went up, not down. It hit 44%.
Here's what we didn't account for. The cheaper provider was cheaper because it ran fewer verification passes. Emails bounced at roughly 8.2%. Our sending domain's reputation started tanking. By early March, two of our sending IPs were on a widely-used blocklist, and our DNS configuration needed a full rebuild—which took our deliverability vendor two weeks to sort out. Meanwhile, our SDRs were spending an estimated 40% of their working hours manually cleaning records before they could even load them into the sequence tool.
That's when I started doing what I now do before any vendor comparison. Total cost of ownership math.
Base price per record: $0.04. Time cost of manual cleanup: roughly $0.31 per record in SDR hours. Deliverability damage from bounces: hard to price, but we lost about three weeks of pipeline. Rebuild of sending infrastructure: two vendor invoices and one very tired DNS engineer. Effective cost per usable record: well north of $0.50.
The expensive vendor wasn't expensive. The cheap vendor was. I've said this to our CFO twice since—she now asks for TCO on every lead-gen line item.
The turn: data isn't a list, it's a flow
Around April, I had a conversation that shifted how I think about this whole problem. We were evaluating how a prospecting agent could sit on top of our existing stack—not replace our SDRs, not replace our CRM, just handle the parts of the pipeline that were breaking.
The thing that clicked for me: all the tools I'd been comparing treated company data as a static list. Buy a file, load a file, send against a file. But our actual problem wasn't that we had a bad file. It was that the file was stale the moment it landed, and nothing in our workflow triggered a refresh.
That's where the concept of an agent-native prospecting workflow made sense to me—not as a buzzword, but as an architectural fix. In an agent-native setup, a prospecting agent doesn't just pull records once. It queries multiple enrichment sources on demand, checks intent signals, verifies emails at send time rather than import time, and routes anything with an anomaly to a human reviewer (that's me, in this case) before it goes out. Data enrichment sales automation stops being a one-shot purchase and becomes a continuously-running process.
We ended up piloting okki-go's agent workflow for one of our SDR pods. Setup took about nine days—longer than the okki go setup guide suggests, in our case, mostly because our CRM schema was a mess (note to self: fix that before we scale this to the other pods). The okki go ai sales agent layer handles the enrichment waterfall and intent-based routing; my team handles the final brand-quality review before send.
A specific thing worth calling out, because it confused me at first: when people ask how does api company data fit into an agent-native prospecting workflow, the answer isn't 'plug in an API and forget it.' It's that API-sourced company data becomes a live input the agent reasons over—real-time headcount, funding events, tech-stack signals, leadership changes—rather than a column in a spreadsheet that ages out. The agent decides when to re-query, based on what it's about to do with the record.
What actually changed
By July 2024, my reject rate on the pilot pod was down to 5.9%. Not because the agent is magic—it isn't, and anyone who tells you an agent can replace human quality review is selling you something. It's down because the failure modes moved upstream. Stale titles get caught at query time. Bounce-risk emails get flagged before sequencing. Anomalies get routed to me with the source and timestamp attached, so I can reject in 30 seconds instead of 4 minutes.
Bounce rate on the pilot pod: 1.4%, down from 8.2%. SDR hours spent on manual cleanup per week: down from roughly 12 hours to under 2. Domain reputation recovery, per our deliverability vendor, took about six weeks from the fix date.
I'll be honest about what we still do manually. Human-in-the-loop outreach isn't a nice-to-have in our world—it's the whole reason I have a job. Every email still passes a brand-tone check. Every claim about a prospect's company still gets a second look if it's going to appear in writing. The agent handles the volume; we handle the judgment call. That split feels right for us. Teams with different risk tolerance will draw the line somewhere else, and that's fine.
The lesson I keep relearning
It took me about 14 months and three vendor migrations to understand something that sounds obvious in hindsight: in lead-gen data, the cheapest per-record price is almost never the lowest total cost. It just moves the cost to a different line item—SDR hours, deliverability damage, brand exposure, or a quality auditor staring at a spreadsheet at 8:40 on a Thursday night.
If you're evaluating prospecting tools right now, price the TCO before you price the seat. Ask what happens when the data is wrong, who catches it, and how long that catch takes. Ask whether the enrichment is a one-time pull or a live input. Ask who reviews the anomalies.
The vendor that answers those questions honestly is usually the one that ends up cheaper—even when it isn't.
Verify current pricing and product capabilities directly with any vendor before signing; our specifics are as of mid-2024 and tools change fast.
