Benchmarks · read from dialer exports · Sep 2026
What cold-calling homeowners actually returns.
Three call centers sent us their dialer exports this summer. We joined every row back to the file we delivered and counted. Every number below carries its sample and its month, and none is rounded up.
Read this first
Four things “match” can mean. We report two of them.
- CoverageA phone number exists in the file. This is what most “95%+” claims measure.
- MatchThe number belongs to the named owner on title.
- ActiveThe number is a live mobile today, not a landline or a disconnected line.
- ConnectSomeone picked up. Only a dialer export can show this.
A Scout Data record is 2 and 3 together. The connect numbers below are 4, from the floors that measured them.
Bay Area solar call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| wrong or dead numbers | 3.40% | 14,741 agent-tagged calls | Aug 31 – Sep 18, 2026 | Share of calls an agent tagged wrong number or disconnected, across three 25,000-record no-solar files. |
| already had solar | 0.98% | 14,741 agent-tagged calls | Aug 31 – Sep 18, 2026 | On a file built to exclude homes with panels — permits cross-checked against aerial imagery. |
| reached a renter | 1.43% | 14,741 agent-tagged calls | Aug 31 – Sep 18, 2026 | Owner-occupied filter, checked against what the agent heard. |
| agent connect per dial, first pass | 4.55% | 74,913 first dials | Aug 31 – Sep 18, 2026 | A predictive dialer with aggressive machine detection; across three files the second and third passes connected at 3.76% and 3.31%. |
| agent connect per dial on passes 1, 2 and 3 | 4.55% · 3.76% · 3.31% | 209,741 dials on three 25,000-record files | Aug 31 – Sep 18, 2026 | Three files worked three times each: 74,913, 72,389 and 62,439 dials. Connect falls with every pass; the appointments on those passes were 21, 23 and 24. |
| appointments on dials 1, 2 and 3 | 21 · 23 · 24 | 74,913 · 72,389 · 62,439 dials | Aug 31 – Sep 18, 2026 | Connect rate falls with each pass; appointments do not — they rose on every one. A record is not spent after one dial. Passes four to six added 31 more. |
| more appointments per dial from active mobiles | 3.65× | two 25,000-record files, pooled | Aug 31 – Sep 10, 2026 | Records whose mobile showed carrier activity in 5 or more of the last 12 months, against the rest of the file (95% CI 1.98–6.74). We rank on it now. |
| appointments in three weeks, from 75,000 records | 122 | 348,858 dials on 74,913 records | Aug 31 – Sep 21, 2026 | Every household the dialer export shows tagged Appointment Set on a delivered number, inbound callbacks included: 42 on the first file, 46 on the second, 34 on the third. One booking per household — a repeat tag is the agent ringing back to confirm, not a second appointment. |
| appointments in their first month | 139 | 419,627 dials on 97,450 records | Aug 31 – Sep 23, 2026 | The same count as the three-week figure, carried to the floor’s first month: every household tagged Appointment Set on a delivered number, inbound callbacks included — 42, 46 and 41 on the first three files and 10 on the fourth, delivered Sep 21. One booking per household. |
| bookings in the top 40% of a ranked file | 23 of 25 | 24,968 records, 25 bookings | Sep 8–10, 2026 | A five-trait score with weights learned on the first file only, applied to the second. Its top fifth booked 2.8 per 1,000 records; the bottom three fifths, 0.1. Superseded as the headline test by bay-area-ranked-forward, which scores a file before it is dialed. |
| more appointments per record once the file is ranked | +36% | 25,000 records each, 136,739 dials over three passes | Aug 31 – Sep 21, 2026 | The third file was built by scoring records on the traits that had booked across the first two, then keeping the top 25,000. Compared at the same dialing depth — the first three passes, as far as the third file has been worked — it booked 34 against the first file’s 25: 1.36 per 1,000 records against 1.00. The first prospective test of the score; the quintile test above was fitted and scored after the fact. |
| of bookings from the top 40% of a ranked file | 92% | 24,968 records, 25 bookings | Sep 8–10, 2026 | 23 of the 25 bookings on the second file landed in its top 40%, ranked by a five-trait score whose weights were learned on the first file only — the same read as the row above, as a share. |
| more agent connects per dial | +77% | 348,858 dials on Scout Data · 89,549 on their cold lists | Aug 12–13 and Sep 3–18 vs Aug 31 – Sep 21, 2026 | Agent connects per outbound dial, every attempt, from the same account’s dialer export: 3.47% on the three Scout Data files against 1.95% on the floor’s own cold lists — the four purchased lists it dialed in the two weekdays before our first file, plus the Southern California list it dialed beside ours through September. Its warm lists (prior inbound callers, prior agent connects, real-time web leads) are left out; with them in, the account connected at 3.16% before and 3.47% on ours. |
| more appointments per dial | +95% | 122 appointments on 348,858 dials · 14 on 78,184 before | Aug 12–13 vs Aug 31 – Sep 21, 2026 | Households booked per 1,000 outbound dials on the same account, every list included: 14 on 78,184 dials in the two weekdays before our first file (0.18), against 122 on 348,858 dials on the three Scout Data files (0.35). On its cold lists alone the floor booked 9 in 89,549 dials, 0.10 per 1,000. |
| fewer wrong or dead numbers | 2.94× | 14,741 agent-tagged calls | Aug 31 – Sep 18, 2026 | 3.40% of agent-tagged calls were wrong or dead numbers, against the 10% line a call center’s own scorecard calls strong. |
New Jersey solar call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| calls per live pickup | 4.58 | 107,574 calls | Aug 1–22, 2026 | Calls ÷ (connects + abandoned), the dialer’s own definition. Inside the floor’s “strong” band (under 7) on all 22 days. |
| answered, machines included | 80.0% | 107,574 calls | Aug 1–22, 2026 | Answered = calls − no answer. Voicemail counts as answered in this report, so read it with the machine rate. |
| of answered calls hit a machine | 72.7% | 107,574 calls | Aug 1–22, 2026 | Strong on 16 of 22 days by the floor’s own bands; the two weak days were low-volume weekends. |
| more live pickups per call than the vendors it already dials | +29% | 71,958 calls on 15 list segments | Aug 2026 | The floor’s own segment tracker, same dialer and month: 26.59 live pickups per 100 calls on the four Scout Data segments (12,825 calls) against 21.30 on the eight segments from one list vendor it already buys (46,197 calls) and 17.46 on the three from a second (12,936 calls); the two pooled, 20.46. Pickups = calls ÷ the dialer’s calls-to-connect, per segment. |
Michigan solar call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| more agent connects per dial than the list it already buys | +22% | 641 dials on Scout Data · 387,712 on that list, same week | Sep 15, 2026 · the other list: week to Sep 15 | The floor’s own dialer performance report, human answers per outbound dial: 4.83% on the 641 dials it put into our Detroit test file on Sep 15 (31 answers, first and second attempts, inside its regular campaign with rotating caller IDs), against 3.95% on the 387,712 dials that week into its own Michigan list from another vendor (15,328 answers). That vendor’s contacts in every state answered 4.54%. A one-day, one-file read. |
Central New Jersey solar call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| more agent connects per dial than its own lists, same hour | 2.63× | 992 dials on Scout Data · 1,305 on its own lists | Sep 21, 2026 | Same agents, same 14:00 ET hour, both first pass: 40 human answers on 992 dials into our 1,000-record test file (4.03%), against 20 on 1,305 dials into the floor’s own numbers (1.53%). Its own rate across the whole day was 1.62%, so the hour is not the reason. The mechanism is dead air: 1.8% of our dials rang out, failed or were busy, against 25.7% of its own. A first-day read, with too few conversations yet to read appointments. |
California battery call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| more appointments per record than its own lists in the same cities | 2.83× | 281 test records · 3,790 on its own lists | Sep 10–21, 2026 | 41 of the 281 records in our San Diego and Sacramento test file were set as appointments (14.59%), against 195 of 3,790 (5.14%) on the permit lists the floor builds itself in the same two cities — its best-booking source, and one it had been dialing for weeks longer. A small test, cut from the top of a ranked file; it stopped moving on day seven. |
Dallas–Fort Worth roofing call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| more conversations per dial than its own lists | +58% | 1,992 dials on Scout Data · 2,788 answered calls on its own lists | Sep 18–21 vs Sep 5–11, 2026 | Conversations of 30 seconds or more per outbound dial: 4.22% on 1,992 dials into our 1,000-record test file, worked by one agent, against 2.67% on the floor’s own lists the week before; answered per dial, 12.3% against 7.8%. Appointments per dial came out level with its best caller on its own lists — the lift is in connection, not yet in bookings, on two appointments. |
Florida roofing call center
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| appointments in their first four weeks, from 15,000 records | 36 | 22,852 dials on 9,658 records | Aug 27 – Sep 22, 2026 | Households the floor tagged Appointment Set on the four files we delivered, one booking per household. A third of the records had not been dialed yet when it was read. |
| more conversations per first dial than its own lists, same hour | +43% | 5,787 first dials on Scout Data · 8,644 on its own lists | Aug 27 – Sep 22, 2026 | First dials only, compared inside each day and hour both sources were dialed: 12.20% of ours reached a person against 7.07% of its own lists before the hour is matched, 1.43× after. Appointments ran about twice as often, on too few bookings to quote as a figure. |
Pennsylvania, Massachusetts and California solar floor
What the floor saw
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| wrong numbers, Massachusetts vs Pennsylvania | 3.7% · 9.2% | 13,038 rep-tagged calls | Jul 20 – Aug 19, 2026 | The honest version: it varies by market. Pennsylvania’s rate is ours to fix and is why we re-verify before a repeat order. |
| live human answer per dial | 4.45% | 2,224,746 dials | 60 days to Aug 25, 2026 | The floor number every call center already knows. 90.54% of dials went to machine detection; 3.70% reached an agent. |
across delivered files
How fast numbers go stale
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| of records go dead within six weeks | 1.4% | re-enriched paid orders | Jul–Aug 2026 | Why a repeat order is re-verified instead of re-shipped. |
suburban Dallas–Fort Worth pool
How many homes yield a record
| Measure | Value | Sample | Window | Note |
|---|---|---|---|---|
| of homes yield the owner on title | 79.7% | 18,309 single-family parcels, Texas | Aug 18, 2026 | A returned contact whose name matches the deed. |
| of homes yield the owner with a live mobile | 54.6% | 18,309 single-family parcels, Texas | Aug 18, 2026 | The deliverable number. About half of owner-occupied homes produce a record; the rest are dropped rather than filled with a relative or a landline. |
Method
How each number was counted.
- Dialer exports are joined to the delivered file on phone number, then on owner name, before any rate is computed. A batch that matches zero rows is a scoping question for the floor, not a finding.
- Attempts are stratified. A first dial and a third dial are not the same event; connect rate falls with each pass and appointments do not. Batches are only compared at the same attempt depth.
- Dialer reports use the dialer’s own definitions and say so. ReadyMode counts a voicemail as answered, so its answer rate is read next to its machine rate.
- Yields are stated with the denominator: an 18,309-parcel pool where 54.6% produced an owner with a live mobile means the other 45.4% were dropped, not filled.
- Nothing is rounded up, no floor is named without written approval, and nothing about a floor’s own dialer settings or deal terms is published.
Do the math on your own numbers: dials per appointment and cost per dialable contact.
Case studies
Call centers that measured it. In their own dialer.
Solar · Bay Area · CallTools122appointments in three weeks, from 75,000 recordsA California solar call centerEleven agents, three 25,000-record files, three weeks, 122 appointments — and a call log that changed how the next file was picked. The third was ranked on what had already booked, before anyone dialed it, and out-booked the first by 36% on the same number of passes.Read the case study
Solar · New Jersey · ReadyMode$200Kin one month, 18 deals, every one from a cold callA New Jersey solar call centerEighteen months of cold calling into a Florida market that had stopped paying, then the floor moved its dialing to New Jersey. August: 18 power-purchase agreements, about $200,000 in commission, every one from a cold call — on a file whose answer rate and calls to connect graded strong on all 22 days in their own dialer.Read the case studyReading a match rate
The questions a buyer should ask.
Answered the way we answer them on a call.
What can a vendor mean by a “95% match rate”?
One of four things. Coverage: a phone number exists in the file for 95% of addresses. Match: the number belongs to the named owner. Active: the number is a live mobile today. Connect: someone picked up. Most published match rates are the first. A call center needs the second and third, and only a dialer export can show the fourth.
Which of the four does Scout Data report?
Match and active together, as the definition of a record — one home, the owner on title, a live mobile — and connect from customers’ dialer exports. A coverage number is never presented as accuracy.
Why do the numbers vary by market?
Because phone data does. On one floor’s export, wrong numbers ran 3.7% in Massachusetts and 9.2% in Pennsylvania on the same month’s files. Publishing the spread is more useful than publishing the best market.
Are these numbers rounded?
No. 3.40% is 3.40%. Every figure carries the sample it was counted on and the month, and the derivation file is named in the page source.