cold-calling

Cold Calling

Overview

This skill produces a cold calling plan a team can run on Monday: a dated launch checklist with owners, a call list with every number typed (business line, direct dial, mobile), call blocks and an attempt cadence sized to a meeting target, a short script built on opener, reason, stake, question and ask, a typed objection bank, voicemail and gatekeeper scripts, CRM fields, and a scorecard where every metric has a target and an action threshold.

The mistakes it prevents: calling the switchboard when the owner's mobile was on file, dialing all week from a number that shows "Spam Likely", dialing between 12 and 2pm (the window Berman says almost everyone avoids), pitching something outside the offer, booking meetings with people who do not own the decision, promising emails nobody wrote, judging a script on 40 dials, and logging calls so loosely that you cannot tell whether the list, the opener or the rep is the problem.

When to Use

• "Write me a cold call script", "how do I handle 'send me an email'", "what do I say to the gatekeeper".
• "Get phone numbers for this list", "find mobiles for these VPs", "build a call list of roofers in Tampa".
• "Click-to-call or a parallel dialer?", "what hours get the most connects?", "our number shows as Spam Likely".
• "How many dials do we need for 20 meetings?", "our connect rate is 3%".
• Setting up dispositions and call logging in HubSpot, Pipedrive or GoHighLevel.

Use something else for nearby jobs:

Prerequisites

1. Market: B2B title list or local business category and geography.
2. Offer, in the user's words: what is sold and what is not. The plan never adds a product line the user did not name.
3. The meeting you are booking (length, who attends).
4. Dialer and CRM by product name, and how the dialer dials: click-to-call, one-line power dialer or a parallel dialer (step 3). If the user names none, ask; the plan names the product it is written for.
5. Reps, hours available per day, time zones they cover.
6. The company's own records: past one-time customers, lapsed customers, lost deals and quotes, demos that did not buy.
7. The company's internal do-not-call list (everyone who ever asked not to be called).
8. ScraperCity plan: Growth ($149/mo) or Scale for the Lead Database and Local Business Database; any plan for find_mobiles and the scrapers.

Authoritative Foundations

Detail and links: references/framework-notes.md.

Workflow

1. Choose the motion

2. Build the call list

0. Start with the company's own records: past one-time customers, lapsed customers, lost deals or quotes, demos that did not buy. They are Reisert's Priority bucket (he puts deals lost in the last 2 to 3 quarters there), get dialed first, and the reason line cites their own date ("we fixed your unit on March 3rd").
1. Pull the list (Lead Database, Local Business Database, Google Maps or Yelp) with hasPhone: true. Count first with limit: 1. Keywords cover every category of the market the user named (for "independent restaurants": full-service, bar and grill, cafe, each cuisine), not a sample. Add a filter the user did not ask for (a review band, a size band) only when the plan states in one clause what it changes for this offer.
2. Normalize every number to E.164 and keep phone (main line) and mobile in separate columns (references/list-quality-standard.md section 3c).
3. For named contacts with a work email or a LinkedIn URL and no mobile, run find_mobiles. You pay $0.25 only when a number is found.
4. Dedupe on E.164 number, then on business name plus postal code. One row per location for local lists.
5. Suppress: internal do-not-call list, open opportunities, customers, anyone who opted out of any channel.
6. Find the warmest public signal for this offer. When the offer fixes something customers complain about, pull review text (Maps and Yelp scrapes return reviews; read the header row first), match complaint words to the stake (for HVAC: "hot inside", "AC was out"), move those records to Priority and put the review date in the reason line. When the offer depends on who owns the asset (a tenant vs the landlord, a franchisee vs the franchisor), add that as the Qualify question in the script (Workflow step 5) and as a filter wherever the data carries it.
7. Add fresh openings. A minReviews floor drops businesses that opened in the last few months, and new owners are buying now. Pull a second list with maxReviews: 10, match it to free state and local permit, license, inspection, sales-tax or appraisal records, and cite the record's date in the reason for the call. These are non-ScraperCity sources. Source types, the trigger each implies and a verified Texas set: references/public-records-triggers.md.
8. Set the dial order by reachability: owner mobile, then direct dial, then main line (Berman: moving from switchboard to mobile can double the connect rate), and inside each, records with a fresh trigger or a past connect first.
9. With two or more reps, split the list so each rep owns their records: by territory (and the matching local caller IDs) or by tier. Write a "Rep split:" line with the counts.
10. Before the first block, hand-check a random sample of records: the number reaches the business, the named contact matches, the business is open. Write the sample size and the failure count that sends the filter back for a fix into the plan.

3. Set up the dialer and caller ID

Settings in references/dialer-and-caller-id.md. Short form:

Write each setting as a value the rep types into the named dialer (calls per number per day, attempts per number per day, maximum attempts, voicemail mode, which numbers each rep holds), not as a description.

4. Size the call blocks from the meeting target

Work backward from meetings with the funnel in references/call-blocks-and-metrics.md:

dials needed = meetings / (connect rate x conversation rate x meeting rate)

With Belkins' 2025 rates (9.9%, 58%, 4.6%) one meeting needs about 379 dials (1 / (0.099 x 0.58 x 0.046) = 378.6). Treat those rates as the starting hypothesis and replace each with your measured rate once the segment has the sample the plan sets (Berman reads rates over 50+ dials at minimum).

Blocks: phone only, no email or CRM cleanup inside the block; block length and count come from the reps' available hours. Berman's page puts a trained rep at 40 to 60 targeted calls a day; the funnel decides whether that is enough, and dials per hour are measured in week 1.

Block policy (Blount, Fanatical Prospecting): blocks sit in the Golden Hours when buyers answer, go on the calendar as busy, and survive a good week. Under his Law of Replacement every opportunity that closes or is lost has to be replaced, so next week's block count does not drop after a win. Research, email and CRM cleanup go in the admin slots.

Work order inside the day (Reisert's buckets): confirm Scheduled meetings first, then Priority callbacks (Activated, Not Now due, no-shows), then Working records in cadence, then new records. Detail and CRM filters: references/call-blocks-and-metrics.md sections 4a and 4b.

5. Write the script

Structure (from Blount and Berman, tuned with Gong's data):

Name the stake. Before writing the reason for a segment, decide what the most expensive thing is that can break, sit idle or go wrong for this buyer this month that the seller's offer fixes, and put it in the reason or the question. The stake comes from inside the offer as the user stated it: an HVAC maintenance seller calling restaurants names a rooftop unit failing over a full dining room on a 100 degree Friday, not the walk-in cooler, unless the user said they service refrigeration. Confirm it on the first 10 conversations. Examples by seller and buyer: references/call-scripts-and-objections.md section 2h.

Write every script line, voicemail and email with merge fields from the list columns ({owner_first}, {business}, {rating}, {review_count}, {city}, {trigger_date}), so reps never rewrite a sample record. Name the column each field comes from in the "Source of the fact" cell.

Before each dial, 3x3 research (Steve Richard): three insights in three minutes covering the account, the person and why now, written into the call notes. On local lists the record already carries the account facts.

If they engage, follow Pessar's beats: confirm the right person, then a Qualify line: a present-tense closed question that confirms this person owns and pays for what the offer covers ("Are the rooftop units on you or the landlord under your lease?", "Do you sign off on the office software?"); a no gets "who does?" and a Not Me or Referred log, not a booking. Then one or two open questions with a pause after each, then sell the meeting the way Slocum describes it: repeat the problem in their words, say what 15 minutes will show, offer two times, and confirm the invite while they are on the line.

Tonality: Pessar sets tone by role, warm for people-facing roles and confident and crisp for technical ones (Mixmax interview, 2023). Berman gives the opener 8 to 10 seconds before the prospect checks out, so name, company and reason come first.

Library and variants: references/call-scripts-and-objections.md (section 0 for 3x3, section 2g for discovery).

6. Handle objections, voicemail and gatekeepers

Use the bank in references/call-scripts-and-objections.md: acknowledge, ask one question, then ask again or exit. Each objection carries a type (value, timing, competition, authority, status quo, trust, access, stop) and the outcome to log if they hold firm, so the weekly review can count them. Once one objection type leads the count, answer it inside the reason or the voicemail and test that as one lever. Voicemail under 30 seconds with the callback number said twice (Berman). Gatekeepers get your name, company and the reason in one line; the rep asks the gatekeeper's name and when the owner is in, logs both, and books the callback at that time.

Every email, text and confirmation the plan mentions is written out in full with a subject and merge fields (E1 to E3 for the cadence, C1 and C2 for confirmations), in the same table as the voicemails. A cadence cell names the message ID; it never says "send an email" without the copy. Templates: references/call-scripts-and-objections.md section 4a and references/call-blocks-and-metrics.md section 10.

7. Run attempts and log every call

Cadence per number: up to 5 attempts over about two weeks (Berman), one attempt a day per person at most (Kixie, June 2023), rotated across morning and afternoon blocks, rep-left voicemail on some attempts with a same-day email. Write the attempt days into the plan (references/call-blocks-and-metrics.md section 4c). Cognism and WHAM found 93% of conversations by attempt 3 and 98.6% by attempt 5. Log every dial with the disposition list in references/call-blocks-and-metrics.md, and move any "don't call me" to the internal do-not-call list the same day.

Every conversation also gets one outcome, using the labels from Gilkey's and TitanX's framing: Meeting Scheduled, Activated, Not Now, Not Me, Referred, Not Interested. Referred names go through find_mobiles or find_emails and into list v2. Every Meeting Scheduled gets the invite confirmed while the prospect is on the line (Pessar), then Reisert's confirmations: a call about a week out, an email 12 to 24 hours before, and a call 2 to 3 hours before if the email got no reply. The plan writes the confirmation text and email (C1, C2) in full. After attempt 5 with no conversation, the record closes for the cycle (Berman: past five attempts, fresh targets pay better); if the user wants an email-only follow-up, the plan writes those emails too.

8. Run the dated 2-week pilot

Every plan starts with a 2-week pilot on calendar dates. Week 1 runs the plan as written as the baseline. Week 2 changes exactly one lever (list, opener, reason or stake, ask, or calling window). Before day 1, the plan writes down the conversation count each week needs before a verdict and the lift in conversation-to-meeting that keeps the variant. On the last day, apply those rules (references/call-blocks-and-metrics.md section 7d); if connects run far below the Belkins 9.9% planning rate, fix the list or number type before judging any script change.

9. Review weekly

Compare each stage rate against the hypothesis. Low connects point at the list, the number type or a caller ID label; low conversations at the opener; low meetings at the reason or the ask. Change one lever per test. A large share of Not Me plus Referred is a list or title problem (TitanX read about 32% in one Director-level example that way, April 2026): change the title filter and feed referrals into list v2. Fill the Rep Scorecard every Friday against the targets in references/call-blocks-and-metrics.md section 9a, cut by segment, rep, number type and block. The same Friday, the manager scores a sample of recorded calls per rep, gives one keep and one fix, and the rep role-plays the fix before Monday (section 13).

10. Prove calling works before hiring

Before hiring an SDR, the founder or first seller runs the motion and measures cost per held meeting = (hours x loaded hourly cost + data + dialer) / held meetings. Hire when conversation-to-meeting is at or above the Belkins 4.6% planning rate and cost per held meeting is at or below cold email's (outbound-unit-economics). Below that, keep email primary and use the phone for Priority callbacks and replies that ask for a call. Detail: references/call-blocks-and-metrics.md section 11.

ScraperCity Implementation

MCP calls:

{"tool": "query_local_business_database",
 "keyword": ["roofing contractor"], "location": ["Tampa FL", "St. Petersburg FL", "Clearwater FL"],
 "hasPhone": true, "hasContact": true, "minReviews": 15, "limit": 1}

{"tool": "find_mobiles",
 "inputs": ["https://www.linkedin.com/in/example-owner", "[email protected]"]}

CLI: npx scrapercity db-local --keyword "roofing contractor" --location "Tampa FL; Clearwater FL" --has-phone --has-contact --limit 100, then npx scrapercity mobile-find --file inputs.txt, npx scrapercity poll <runId>, npx scrapercity download <runId> -o mobiles-roofers-tampa-20261013.csv. REST: POST /api/v1/scrape/mobile-finder with {"inputs": [...]}, then GET /api/v1/scrape/status/{runId} and GET /api/downloads/{runId}.

Mobile Finder returns mobile_number, status and message per input (scrapercity.com/mobile-finder); read the header row before mapping.

Worked cost estimate (B2B title list)

Growth plan. Target: VP Sales and Head of Sales at 50 to 200 employee US software companies.

Over $5: call check_wallet, show this table, and wait for a yes (an agent step before the run; it stays out of the plan). The hit rate is not published, so run a first batch, measure found / inputs, and re-estimate the rest as remaining inputs x hit rate x $0.25.

Worked cost estimate (local business list)

Output Format

One Markdown call plan from templates/output-template.md. It opens with one header line (company and offer in the user's words, market, reps, dialer and CRM by name, plan period) and the Program Summary, so the answer comes first. Sections in this order:

Add a Cost Estimate section only when the user asked about cost, budget or cost per meeting; otherwise the paid data line is priced in Call List Build and cost per held meeting is a measured Funnel Targets row.

When the user asks only for a script, return the Call Script, Objection Bank and the Voicemail, Gatekeeper and Email sections.

What the reader receives is the deliverable only:

• The offer exactly as the user stated it. The stake, script, voicemails and emails sell only that.
• ScraperCity products by name (Local Business Database, Mobile Finder) with filters in plain words. Tool ids, API paths, CLI commands, file paths, skill names and this skill's section numbers stay out; where another routine would run, write its steps.
• Instructions without the names of trainers or frameworks (no "Berman's window", "Reisert's buckets", "3x3"). Published datasets are cited by name and year in a Basis cell (Belkins 2025, Cognism 2025); vendor recommendations as recommendations (Readymode recommends 75 calls per number per day).
• No agent process: no wallet checks, approval gates, "not run" or "not checked" lists. One "Estimates:" line covers every estimate. Absence of data is written as "none found", never "none exists".
• Each fact once: targets live in Funnel Targets only, the scorecard holds actuals, symptoms and fixes live in the "Act if" cells, no second diagnosis table, no restated dates.
• A full plan runs 900 to 4,000 words.

Quality Check

python3 scripts/check-output.py call-plan.md enforces:

• All twelve sections present, no [fill] left, at least 30 table rows, 900 to 4,000 words.
• No tool ids, CLI commands, file paths, skill names, trainer or framework names, agent process lines (wallet, approval, "not checked"), or "no ... exists" claims; exactly one "Estimates:" line.
• Every Launch Checklist row has an owner and a YYYY-MM-DD date; a plan with two or more reps has a "Rep split:" line in Call List Build.
• Dialer Setup names the dial mode, caller ID numbers with a test schedule, the window with hours, and the attempt cap.
• The plan carries no legal content.
• Funnel Targets is one table; dials, connects, conversations and meetings each have a count and a basis, and every stage after dials has an "Act if" with a number.
• The Call Script has opener, reason, stake, question and ask rows, at least two merge fields and a Qualify beat.
• The Pilot Plan spans 7 to 16 calendar days, every row has a date, and each row changes at most one lever and carries a decision rule.
• At least 8 objections, each response 45 words or fewer and containing a question.
• Every voicemail script is 75 words or fewer: Berman's 30 seconds at the 150 words a minute the National Center for Voice and Speech gives for US conversation (cited by Voices.com).
• Every email the cadence names is a written row with a subject; a confirmation message (C row) is written.
• Mobile Finder, wherever used, is priced at $0.25 per number found; a Cost Estimate, if present, has dollar figures and a total.
• The Rep Scorecard has every required column, no "target" rows, and any filled row's rates match its counts within rounding.

Also check by reading: the stake and every message sell only the offer the user named; the reason cites a fact from the data; the keywords cover every category of the market the user named; the Qualify line tests who owns the asset or budget; the ask names a time; every number is typed before it is dialed; caller ID numbers show no label on the latest test; every price is current on the plan date.

Common Pitfalls

Execution Artifacts

Related Skills

• multi-channel-outreach places calls inside an email and LinkedIn cadence.
• mobile-appender runs find_mobiles at CRM scale.
• local-business-list and scrapercity-local-business-database build the local call list.
• decision-maker-finder and scrapercity-lead-database find the title holder to call.
• signal-triggered-outbound ranks which accounts get called first.
• outbound-unit-economics turns the call funnel into cost per meeting.
• reply-handling and cold-email-strategy cover the email side of the same accounts.