Two conversations this week landed in the same place.

One was with a founder who has enormous senior contacts and almost no customers. He spends his time talking to execs in his network. In theory they're his ICP. In practice none of them are in pain — they take the call because he's a friend.

The other was a GM building a new business line. He doesn't care about his network. He cares about signals, and he's building an outbound motion around them instead.

I know which one wins.

Almost every outbound program that's actually working right now runs on the same three steps: signal, contacts, outreach. Find a reason to reach out. Find the right person to reach out to. Reach out in a way that doesn't read like a machine wrote it.

That's the whole motion. The hard part is that each step fails differently, and a failure in step one stays invisible until step three — when nobody replies, or worse, when the replies never turn into customers. Most teams are building this today in Claude Code, Cowork, or Clay. The tool matters less than getting each step right, and the companies pulling ahead are spending human time on these three steps, not just tokens.

Step one: the signal

A signal is a change in a company's world that makes your product more relevant this week than it was last week. It doesn't have to be causal — it just has to correlate closely enough that you're not reaching out after the buying window's already closed.

The good ones are specific and recent: a new head of sales or a first RevOps hire, a first AE or a new AE req posted, a funding round, a new office or product line, a pricing change or rebrand, even a leadership departure, which tells you almost as much as an arrival does. The bad ones are static — company size, industry, and tech stack are filters, not signals. They tell you who to talk to, not when. For Skyp, a new senior sales leader, a first AE hire, or a new SDR manager are the signals that matter. A funding round can work too, though not always.

Signals come from a few places: press for funding and layoffs, LinkedIn activity for job changes and launches (the richest source and the least structured), public data sets like job boards and filings, and your own site — a target-account visitor spending real time on your pricing page after ten other pages is a legitimate signal, treated as probability, not fact.

On an overused signal, speed beats accuracy. Every vendor watches the same funding feed, so by the time a round is verified and published, a few hundred sellers already have the same list. A source getting that data a few days early, before it's verified, is worth more than one that's always exactly right. Back in 2011, when my old company announced its Series A, someone from Morgan Stanley found our office and dropped off a cake the same day. Speed was the whole point. Rare signals are the opposite — you get one attempt, and the prospect will notice if you're wrong, so accuracy matters more once you're working with something nobody else has.

Most teams build these workflows in Claude Code or Clay, or buy something like White Whale off the shelf. Start off the shelf if you can — you'll be live immediately — then bring it in-house once you know which signals actually work for you.

Step two: contacts

Once a company shows a signal, you need the humans behind it. Start with who actually feels the pain and who signs — a company-level signal is useless if it's routed to the wrong title. Work this out from your own closed deals and call recordings, not from an agency or a consultant. Test different titles if you're not sure.

One CRO replied to me specifically because I'd connected with his entire team the day before, and most of them had already accepted. That's the kind of detail that actually gets read.

Once you know the title, go find them. ZoomInfo is still the best single dataset, priced like it. Waterfall tools like FullEnrich stack cheaper providers together automatically. Clay or Claude Code let you build your own waterfall with your own logic, at lower cost and more control. We've stopped recommending Apollo — in some samples, 40% of its contacts had already left the company it listed for them. LinkedIn shows your current employer at the top of your profile, so messaging someone about the job they left last quarter reads as lazy, not personalized. Stale data doesn't just waste a send. It poisons the relationship before it starts. Better to skip the personalization than get it wrong.

Step three: outreach

Once you know who to reach and why, you have to actually reach them — and the goal is a reply, not a send.

LinkedIn gets better response rates than email, largely because the platform rate-limits how much of it you can do — scarcity is the whole reason it works. Email scales, and scale is exactly what ruined it; it still works with tight targeting and clean infrastructure, but the entire stack has to be right for a message to reach a human inbox. Phone is the one channel AI can't touch: the FCC has ruled AI-generated voices illegal on cold calls, and TCPA penalties run in the hundreds of dollars per call, tripled for willful violations. Nobody's putting AI on the phone. That makes it expensive — you need real humans and a parallel dialer like Nooks, Orum, or ConnectAndSell to make them productive — but it works.

We built Skyp to run email and LinkedIn from one platform, wired into whatever orchestrates the rest of your system, with an easy handoff to a dialer — open an email, follow up with a call and a LinkedIn request, without stitching five tools together by hand.

What you actually say matters more than which channel you're on. Would you ever follow "what do you do for work" with "what's your biggest challenge generating leads" in an actual elevator? No. So on LinkedIn: connect, send a short thank-you note, and don't lean on the signal or pitch the product in the same breath. Every specific detail you add is a chance to be wrong, and a wrong detail reads as AI slop faster than a generic one does. Let your confidence in the data set how personalized you get, then follow up later, once LinkedIn's already surfaced a few of your posts.

It's never right on the first try

Nobody gets this right on the first configuration. If it's not working, check each step before you blame the channel — bad contact data or a weak signal will sink good copy every time.

Keep tracking which signals produce meetings, which produce customers, and which titles actually convert. Reply rate and meeting rate by signal type are your leading indicators long before revenue shows up. Give it real time before you judge it — a sales cycle is long enough that three weeks of data tells you nothing.

Expect two things. It'll work differently for you than it does for whoever's posting their framework on LinkedIn — one of my best-performing posts was satire, and I still get messages asking for the playbook behind it. And outbound leads convert on a different timeline than your inbound channels, so judging them against your search or social benchmarks will make a working program look broken.

Outbound also works better when it isn't the only thing running. Brand and content make your name familiar before the email lands, and that raises the reply rate on everything else. Do the three steps well, but don't expect them to carry the whole motion alone.