Phone Propensity Scoring: Predicting Who Will Pick Up
Phone propensity scoring ranks contacts by how likely they are to answer a call, not by how valuable they are as customers. It combines inputs such as number type, past call outcomes, persona and hour, and sorts a list into tiers so that the best slots go to the numbers most likely to produce a conversation. You can build a workable version from your own call log in an afternoon.
Phone propensity is not lead scoring
A lead score answers "is this account worth winning?". It is built on fit and intent. A propensity score answers "will this number be answered?". The two often disagree.
The perfect-fit CFO may never pick up an unknown number. The office manager of a so-so account answers every time. If you sort by lead score alone, reps spend the day in voicemail. If you sort by propensity alone, they spend it talking to people who will never buy.
So use both, as a simple grid.
High value, likely to answer: call first, at the best hour, fully prepared.
High value, unlikely to answer: keep dialing, in parallel, and add other channels.
Low value, likely to answer: filler for quiet slots.
Low value, unlikely to answer: do not call.
We say it often: connect rate is an input, not an outcome. A propensity score that raises connects and lowers meetings is a bad score.
The inputs that predict a pickup
Some of these come from our benchmark of 1,000,000 B2B cold calls in France. The others come from what we see in sessions, and we label them as such.
History on this number. In our experience the strongest input. A number that was answered once will be answered again more often than a number never tested. Three no-answers at three different hours pull the score down. A wrong-person outcome sets it to zero.
Number type. A mobile reaches the person. A direct line reaches a desk. A switchboard reaches a colleague. Score the first two, and keep switchboards out of the score: they inflate it without producing the conversation you wanted.
Freshness. When was the number last verified or last answered? People change jobs, and a number goes stale with them.
Persona. The average connect rate in our benchmark is 11.3%, with a range of 11 to 20% depending on persona and slot. Sales and leadership personas sit above 15% overall.
Hour. 4pm connects at 15.8%, 10am at 14%, 8am at 9.8%. Founders pick up more around 10am, marketing around 4pm, HR around 1pm.
Relationship. Known contacts behave differently. A call that opens with "How have you been?" to someone you already spoke to reaches about 10% success, against roughly 1.5% for a standard cold open.
A recent signal. In our experience, someone who visited your site yesterday or replied to a message this week is reachable this week.
Two things follow. First, propensity is not a fixed property of a contact. It is a contact at a given hour: the same marketing lead is a poor dial at 8am and a good one at 4pm. Second, one input sits on your side of the line. If your caller ID is flagged as spam, everyone's propensity drops, and no score will fix that.
Build a three-tier score from your own call log
Export three months of calls. Contact, number, number type, persona, hour of the call, outcome.
Define the outcome you care about. A conversation with the intended person. Not a pickup, not a voicemail, not a receptionist.
Compute the conversation rate by number type, by persona and by slot. This shows you which inputs separate good dials from bad ones in your market.
Give points. As a starting grid, to be corrected by your data: answered before, plus three. Mobile verified in the last 90 days, plus two. Persona above your average, plus one. Known contact, plus one. Three no-answers at different slots, minus two. These weights are an example, not a benchmark.
Cut into three tiers. A, B and C. Three tiers a rep trusts beat a score out of 100 that nobody reads.
Test it. On the next 500 dials, does tier A convert to conversations clearly more often than tier C? If not, the weights are wrong. Change them.
Feed it back. Every call outcome updates the score. A tier C contact who answers moves to tier A that day.
Use the tiers to run the session
Tier A gets the best slot for its persona, and a prepared rep. If the contact is also high value, the rep reads the brief before the dial, because the conversation is likely to happen.
Tier B fills the normal blocks.
Tier C goes to parallel dialing, where a dial that rings out costs the rep no waiting time, or starts with another channel.
The score orders the list. It does not delete anyone. A high-value account in tier C still gets called, with more lines and more patience.
The limits of a propensity score
It predicts a pickup, not an identity. The wrong person can answer a high-scoring number. Verification is a separate job.
It is an average. A tier that answers one time in four still does not answer three times in four. Reps should not read a tier A as a promise.
It only learns from the numbers you dial. Untested contacts stay unknown forever if the score always sends reps to the known ones. Keep a share of every session for untested numbers.
Small samples lie. A persona and slot combination with 30 dials tells you nothing. Wait for a few hundred.
Keep the inputs explainable. In Europe especially, build the score on professional data and your own call outcomes, and be able to say what goes into it.
How Sonik approaches pickup prediction
We start from the benchmark, not from a black box. Sessions in the Sonik dialer are built by persona and run at the hour that persona answers: founders mid-morning, HR at lunch, marketing in the late afternoon.
The Sonik dialer during a session: parallel calls, statuses and the running count of meetings booked.
The reason to call comes before the odds of a pickup. The signal engine reads website visitors, job changes, closed-lost deals, dormant leads and social intent, and those contacts go to the top. Reps use only about 20% of the signals already in their CRM, and that is where the contacts who both answer and convert tend to be.
Parallel dialing then changes what the lower tiers cost. Reps get about five times more conversations than by dialing one line at a time, so a high-value contact who rarely answers keeps being dialed without eating the rep's hour. Our full numbers on pickup are in the real cold call connect rate in B2B sales.
Conclusion
Phone propensity scoring is a way to spend your best hours on the numbers most likely to answer, without forgetting who is worth a conversation. Past outcomes, number type, persona and hour do most of the work. This week, export three months of calls, compute your conversation rate by persona and by slot, and move one persona to the hour where it answers.