I Wasted 14 Months Chasing Intent Signals (And Almost Lost My Biggest Account Learning Why)

October 2022. I was sitting in front of a sales engagement platform dashboard that our team had been running for three months. 200 sequences. Roughly 4,200 emails sent. 72 meetings booked. Numbers that looked just fine in the weekly report.

Then I pulled the attribution view. Nine deals closed. That's one close per 466 emails. And about 60% of the "intent signals" we were chasing? Garbage. A competitor doing research on our pricing page. Two interns fat-fingering our homepage. One spike that turned out to be our own office IP hammering the blog because our CDN was misconfigured.

We were paying roughly $6,800 a month for those intent data feeds. That worked out to about $4.10 per actually-usable signal. The frustration wasn't the money — it was that I'd built a pipeline on top of data I couldn't trace.

I've been doing sales ops for nine years. I've stood up SDR teams four times. I say that because I want you to understand: I should have known better. But I didn't. And I documented the whole mess, so here's what it actually looked like.

The first mistake: confusing more signals with better signals

Background, so you can calibrate how much to trust me. Between 2020 and 2023 I ran a 12-person SDR org at a roughly $14M ARR company. I owned the outbound stack. I was the guy who said things like "let's layer in another intent source" without asking what 'another source' was actually measuring.

In early 2023 I got swept up in the classic rookie mistake: assuming that buying more data would automatically produce better targeting. It doesn't. It produces more meetings that don't convert, and more arguments between the SDR team and marketing about whose fault that is.

By Q1 2023 we had four tabs open, four intent platforms, and four different definitions of what a "high-intent account" looked like. The SDRs had stopped trusting all of it. Honestly, so had I — I just didn't want to admit that in the QBR.

What I kept telling myself was that we just needed better filters. But the actual problem wasn't filtering. It was that none of these platforms would tell me where the signal came from or how old it was. Everything was "proprietary." Everything was "AI-powered." Nothing was traceable.

The detour: okki-go and the uncomfortable question of data source transparency

Mid-2023 I started looking at replacements. That's when okki-go came onto my radar. What caught my attention first wasn't a feature list — it was the fact that they would show me the source layer for a signal. Not a score. Not a heatmap. The actual event, when it happened, and how stale the record was.

I'm not going to pretend that was love at first sight. I'd already been burned by two vendors who gave great demos. But the first call made me pause. The product lead — unprompted — told me I shouldn't connect every data source I had access to.

His exact framing stuck with me: "If your ICP coverage is narrow, more signals just manufacture noise. Decide what question you're trying to answer first."

To be fair, I was annoyed. I'd come into that call ready to talk about scale. I walked out realizing I wasn't ready to answer a simpler question — what was I actually trying to detect?

In my experience, vendors who can't tell you what they don't do are usually the same ones who overpromise on what they do.

That's the part that reframed the whole evaluation for me. If I'm being honest, I'd rather work with a specialist that draws a hard line around its capability than a generalist that says yes to everything. This is true in agencies, it's true in consultants, and it turns out it's true in intent data.

The real pivot: how sales engagement platform features fit into an agent-native prospecting workflow

Here's where I burned another two months, and where I think most teams still get it backwards.

When people ask how does sales engagement platform features fit into an agent-native prospecting workflow, they usually picture the agent as a filter at the front of the funnels: agent qualifies → human writes → platform sends. That architecture fails because the agent never learns from what happens inside the sequence. Email tracking, LinkedIn engagement, replies, bounces — all that signal dies at the human action layer.

The way okki-go structured it for us was different. Signals get normalized first. Then the agent reasons across them, including the outcome data flowing back from email tracking and LinkedIn tool features. It's not "here's a score, go be a hero," it's "here's what changed, here's why it probably matters, and here's what I'd do next."

That sounds fuzzy until you see it work. Two things changed for us almost immediately:

• The SDR team stopped treating every LinkedIn interaction as a buying signal. A profile view from a hiring manager is not the same event as a reply from the same person. The agent kept those separate.
• Email tracking data fed back into the intent layer instead of just sitting in dashboards. Open patterns started shifting which accounts got sequenced the following week.

I'd read plenty of posts saying intent data only works at the account level. In practice, for our motion, joining signal recency with engagement outcomes changed the picture more than any new source we added.

Where I slipped again (briefly)

October 2023 — I almost repeated the original mistake. I got excited about a new data enrichment category and quietly wired it into the workflow before we'd validated that the okki-go signals were stable. Three weeks later I had to unwind it. Same lesson, new costume.

The reason I'm writing that down publicly is because I don't think the pattern gets fixed once. It gets fixed by having a checklist. Ours now has four lines on it before any new data source goes near the pipeline:

1. Can I name the source event that produced this signal?
2. Can I tell how stale it is?
3. Do I know how the agent will use it, or am I just hoping it helps?
4. If it's wrong, how will I notice?

The fourth one is the one that saves the most money.

What actually changed over 14 months

By the end of 2023, the numbers were cleaner, but that's not the interesting part. The interesting part is that I stopped spending Fridays reconciling discrepancies between intent reports.

Reply rates moved from 1.1% to just under 3.4% on the sequences where the agent was doing the routing. Not because we wrote better copy — we didn't — but because we stopped sending high-effort touches to accounts the agent could tell weren't in-market. That's a sequencing change, not a persuasion change.

More than anything, okki-go's data source transparency is what made everything else possible. Without it, I was back in the same fog I started in — trusting a number I couldn't trace to an event I could verify. If I'd been honest about that requirement in early 2023, I would have saved my team about a quarter of bullshit work.

According to the FTC's advertising guidance, claims about data and targeting need to be truthful, substantiated, and non-misleading. That's a legal bar, but it's also a useful internal bar. If a provider can't tell you how their signal was generated, you're the one holding the liability for whatever you do with it.

What I'd tell someone starting this evaluation today

Don't compare feature lists first. Compare what each platform will let you verify first. Features are cheap to build. Verifiable data pipelines are not.

And when a vendor tells you something they don't do well, write it down. That's not a weakness to negotiate against. That's usually the only honest sentence in the entire demo cycle.

Fourteen months in, that's the lesson I keep coming back to. Not "buy better data." It's "buy data you can defend in a postmortem."