B2B outbound no longer has a signal problem. It has a speed-to-conversation problem. Modern cold calling needs three layers working together: intent data to know who to call, AI dialing to actually reach them while the signal is warm, and a human to build trust once someone picks up. Remove one layer and the other two are worth very little.

What we learned building an intent platform

At Sonik, since 2025, we have been building an intent signal platform to tell sales reps exactly who to call. Website visitors, job changes, closed-lost deals, dormant leads, people engaging with competitors' content. The idea came from Google Ads. In B2C, intent triggers action instantly: you look at a pair of shoes, and seconds later you are retargeted.

Then we discovered the uncomfortable truth. Finding the right prospect was only a third of the problem.

In B2B, a prospect shows intent today. Sales acts on it one to three days later on average. The rep calls once, hits voicemail, and moves on. The signal was correct. The action never happened. A signal you do not act on is just data.

The gap is already visible in most CRMs: reps use only about 20% of the signals that are sitting there. The problem is not detection. It is what happens in the hours after detection.

Layer 1: know who to call

The first layer replaces the cold list with a warm one. A cold list is a data problem, and you fix the list before you fix the pitch.

• Input: CRM history, website visitors, job changes, closed-lost deals, dormant leads, custom intent.
• Method: only dial contacts carrying a signal. Never a raw export.
• Goal: a warm list every morning, built for the rep, not by the rep.
• KPI: an identified intent signal on 100% of the leads dialed.

The payoff is in the conversion numbers. Trigger-based openers reach a 7-10% meeting rate against 2-3% for pitch-first openers. A call to a dormant contact that opens with "How have you been?" succeeds about 10% of the time, against roughly 1.5% for a standard cold open. And 50% of Sonik's own pipeline comes from signals that were already in the CRM.

Layer 2: actually reach them

The second layer is where most teams leak. Across the 1,000,000 B2B cold calls we analyzed with Katana, the average connect rate is 11.3%. Around 90% of calling time is not spent talking: it is rings, voicemails, IVRs and wrong numbers. AI's job here is to delete the waiting.

• Input: the warm list of the day, already prioritized.
• Method: several lines ring at once; voicemails, IVRs and dead numbers are skipped; the rep only hears a human.
• Goal: turn dial time into talk time while the signal is still warm, the same day, not three days later.
• KPI: 5x more conversations per hour than manual dialing. In the benchmark, 16.2 conversations an hour on a parallel dialer versus 3.2 by hand.

Timing belongs to this layer too. 4pm connects at 15.8% across personas, 10am at 14%, 8am at 9.8%. Founders pick up more around 10am, marketing around 4pm, HR around 1pm. In our experience, most connects happen within the first three attempts, which is an argument for retrying fast, not for giving up after one voicemail.

Layer 3: build trust

The pickup decides the call. The stack buys 20 seconds. The rep has to win them.

• Input: LinkedIn context, past interactions, and the signal that triggered the call, on screen before the "hello".
• Method: open on the signal, ask "how" rather than "do you", listen before answering objections. Train it three times a week.
• Goal: earn the meeting in the first 20 seconds.
• KPI: 100% of connected calls prepared with AI, so the human can focus on the human part.

This is where the benchmark is most striking. Discovery questions starting with "how" lift conversion by 190%. Asking "What do you mean?" before answering an objection lifts it by 197%. Neither is something AI does for the rep. Both are things AI can make possible by giving the rep context and time.

Common mistakes when assembling the stack

• Buying intent data without changing how you dial. You get better lists and the same voicemails. The signal cools while the rep waits for a pickup.
• Treating speed as an email problem. A sequence triggered in 5 minutes is fast. It is also ignored. Speed to conversation means a phone call, the same day.
• Automating layer 3. AI SDRs that "handle" the conversation remove the only layer that builds trust. Sales is a transfer of trust, and you cannot delegate that to a bot.
• Measuring signals captured instead of conversations held. A dashboard full of intent events means nothing if the conversations-per-day number did not move.
• Fixing them in the wrong order. Targeting first, reach second, conversation quality third. Coaching a rep who talks to 3 people an hour is a slow way to improve.

How Sonik approaches the 3-layer stack

Sonik started as a signal platform and became a dialer because of the second layer. Today the product covers all three: signal agents build a warm list every morning from the CRM, a parallel dialer turns it into 5x more conversations, and AI call prep gives the rep the reason to reach out before each connect.

The design principle is right person, right moment, right conversation. The CRM tells us who. The dialer decides when, which is now, while the signal is warm. The rep handles the conversation.

Our north-star funnel is 20 conversations per day per SDR, with 2x more meetings as the outcome. You can see how the layers connect on the signals and dialer pages, and the dormant-lead playbook in how to reactivate dormant leads.

Conclusion

Intent data told sales who to call. It did not solve the one-to-three-day delay or the single voicemail. The stack that works has three layers: signals, AI dialing, human trust. This week, take the last 20 signals your CRM produced and check how many turned into a conversation within 24 hours. That number is your real bottleneck.