evals/evals.json
{
"skill_name": "cold-call-opener",
"evals": [
{
"id": 1,
"prompt": "I run outbound at Varnhill Analytics - 6 SDRs selling a warehouse inventory forecasting tool to supply chain directors at mid-market distributors. I found a breakdown of Gong's cold call research saying \"How have you been?\" books meetings 6.6x above baseline, and that permission-based openers are 40% less likely to book than baseline. So I want our new talk track built on \"Hi Dana, how have you been?\" and I'm banning permission asks from the floor. Draft the opener.",
"expected_output": "A response that refuses to build the opener on small talk, corrects the 40% figure as belonging to \"Did I catch you at a bad time?\" rather than permission-based openers, reports both Gong datasets and their contradiction without blending them, and recommends a pattern that survives the skill's own gate.",
"files": [],
"expectations": [
"The response does not produce an opener built on \"How have you been?\" or any small-talk greeting addressed to a stranger.",
"It states that the 40%-below-baseline figure belongs to \"Did I catch you at a bad time?\", not to permission-based openers.",
"It names that swap as a known misattribution propagated by secondary sources, rather than treating the user's figure as merely mistaken.",
"It declines to ban permission-based openers on the basis of that figure.",
"It reports the 2018 Gong result (\"How have you been?\" around 10%, a 6.6x lift over a baseline of roughly 1.48-1.5%).",
"It reports the 2024 Gong result scoring near-identical small talk (\"How's your day going?\") at 7.6% while still calling it a weak opener.",
"It states plainly that the two datasets contradict each other on small talk, and does not average, blend, or reconcile them into one benchmark.",
"It notes that Gong publishes neither per-opener sample sizes nor the 2024 success definition, so the contradiction stays unresolved rather than decided.",
"It gives a reason for excluding small talk independent of the data: it fabricates familiarity with a stranger.",
"It states that small-talk openers are removed from the pattern menu entirely, not ranked last or offered as a test variant.",
"It recommends a measured alternative pattern for this run (permission-based or direct reason-for-call) instead of the requested one."
]
},
{
"id": 2,
"prompt": "Kestrel Freight Systems here. We just hired Tobi - he started Monday, first sales job ever, zero phone tenure. Board meeting is in 11 days and I need him booking meetings before then. I read that the peer-proof opener converts at 11.24% versus 11.18% for permission-based, so obviously Tobi runs peer-proof: he'll name two competitors of the account that already work with us. Our product routes LTL shipments to cut empty miles. Persona is Director of Logistics. Give me his script, and tell me which variant to keep after the 11 days.",
"expected_output": "A response that takes the peer-proof pattern off the menu for a first-week rep, promotes direct reason-for-call on the deadline and tenure, treats the 0.06-point gap as unable to separate the two top patterns, and refuses to declare a winner after 11 days.",
"files": [],
"expectations": [
"The response declines to put the first-week rep on the context-first / peer-proof pattern.",
"It states that context-first / peer-proof comes off the menu entirely for a rep in their first weeks, not merely ranked lower.",
"It recommends the direct \"reason for my call\" pattern as first for this run.",
"It names the interview facts that promoted it - the hard deadline and the rep's first weeks on the phone.",
"It states the 11.24% vs 11.18% difference is 0.06 percentage points and cannot separate the two patterns.",
"It grounds that on Gong publishing no per-opener sample size or confidence interval for the 2024 study.",
"It states that promoting context-first above permission-based requires all three of: a director-or-above persona, a verified signal already in hand, and a rep who has run the play before - and that this run fails at least one.",
"It flags the peer name-drop as the pattern's hardest failure mode: an unsupportable claim loses the call outright rather than failing softly.",
"It warns against naming the two competitor accounts unless the rep can defend the reference live.",
"It refuses to declare a winning variant after 11 days.",
"It names the judging threshold of roughly 100 conversations per variant.",
"It converts that threshold into connects using the 65.6% connect-to-conversation rate, giving roughly 150 connects per variant."
]
},
{
"id": 3,
"prompt": "Solivent Energie. We bought 40,000 French homeowner phone numbers from a data broker last month, ran the whole file through Bloctel, and we start dialing Monday to sell home energy-renovation packages - heat pumps and loft insulation. Nine people on the dialer. Write the opener in English and we'll translate it, plus the delivery notes.",
"expected_output": "A response that stops the campaign on French consumer-calling law before drafting: prior consent since 11 August 2026, Bloctel abolished, a broker list not a consent basis, and energy-renovation selling prohibited outright.",
"files": [],
"expectations": [
"The response does not hand over a ready-to-dial opener as its main deliverable without first addressing the French consumer-calling regime.",
"It states that France moved to a prior-consent (opt-in) regime for consumer calls effective 11 August 2026.",
"It states that the Bloctel do-not-call list has been abolished, so scrubbing against it no longer establishes a lawful basis.",
"It states that a purchased broker list does not by itself supply valid consent.",
"It lists the consent requirements: free, specific, informed, unambiguous, revocable, and valid at most one year.",
"It states that the caller bears the burden of proving consent.",
"It names the exposure: administrative fines reaching 375,000 EUR for a legal person, plus nullity of any contract concluded on the call.",
"It states that energy-renovation sales to consumers are prohibited outright in France, so the campaign cannot be fixed by sorting out consent alone.",
"It notes the opt-in regime covers consumers only, and that French professional-to-professional calling stays under GDPR plus sector rules.",
"It states the consent basis must be recorded per number, not assumed for the list as a whole.",
"It states that an internal do-not-call list is required and that a \"stop calling me\" is honored immediately on that call, not at the next list refresh.",
"It labels the guidance as factual rather than legal advice and tells the user to have counsel confirm before the campaign."
]
},
{
"id": 4,
"prompt": "I manage 12 SDRs at Palmwood Group - we sell claims-automation software to insurance operations leaders. My VP watched a webinar and now the rule on the floor is \"talk less, listen more\": every rep turn capped at 10 seconds, minimum three questions in the first minute. Our connect-to-conversation rate went from 61% to 44% since we rolled it out. Write me an opener and the delivery coaching that fits the new rule.",
"expected_output": "A response that contradicts the \"talk less, listen more\" rule with the call-rhythm data, prescribes the opener as one uninterrupted monologue ending on a single ask and a pause, and keeps the 30-second budget.",
"files": [],
"expectations": [
"The response challenges the \"talk less, listen more\" floor rule rather than writing coaching that implements it.",
"It cites that successful cold calls contain longer rep monologues - roughly 37 seconds vs 25 seconds for the longest burst.",
"It cites that prospects' longest turns are shorter on successful calls - roughly 3.5 seconds vs 8 seconds.",
"It states there is no statistical difference in the number of questions asked between successful and unsuccessful calls.",
"It cites 77% more speaker switches per minute on successful calls.",
"It draws the combined reading: deliver the opener as one confident monologue, stop cleanly, then keep the exchange brisk with short turns and fast back-and-forth.",
"It explicitly warns that this is not a licence to talk over the prospect.",
"It states the opener still ends on exactly one ask followed by a planned pause the rep does not fill.",
"It keeps the opener inside the 5-30 second budget (about 75 words) and prescribes verifying it by reading aloud against a timer rather than by word count alone.",
"It names connect-to-conversation rate as the opener's own KPI and compares the team's 44% against the 65.6% benchmark."
]
},
{
"id": 5,
"prompt": "Can you tighten this up? It's what my AE Priya says on dials to Heads of RevOps at 100-500 person SaaS companies. We sell a CRM data-hygiene tool. We have no signal on these accounts - it's a straight list filtered on headcount. Here's the script: \"Hi Marcus! Saw on Instagram you ran the Chicago marathon last month, congrats, that's wild. Did I catch you at a bad time? So I know for a fact your CRM is a mess and your reps are wasting half their week on manual data entry. We're the leading AI-native revenue-data-orchestration layer with real-time enrichment, dedupe and territory sync. Could I grab 30 minutes on your calendar this week? Also, are you the one who owns CRM data, or is that someone else?\"",
"expected_output": "A rewrite that strips the personal-life reference and the bad-time question, hedges the guessed pain as peer-reported, cuts the feature dump, collapses three asks into one, and returns 2-3 one-variable variants.",
"files": [],
"expectations": [
"The response removes the Instagram / marathon reference from the opener.",
"It gives the reason: personalization built on personal details reads as surveillance, so personalize only on business-visible signals.",
"It removes \"Did I catch you at a bad time?\" and states that it invites the scripted exit.",
"It notes both Gong datasets rank better options than the bad-time question (0.9% in 2018, 2.15% in 2024) and that they disagree on the baseline, not on the ranking.",
"It identifies \"I know for a fact your CRM is a mess\" as a guess asserted as fact about the prospect's own company.",
"It rewrites the pain line as a hedged, peer-reported guess that explicitly hands the judgement to the prospect.",
"It states that the unhedged version invites the brush-off \"I don't have the problem you're describing\".",
"It cuts the product vocabulary (\"AI-native revenue-data-orchestration layer\", \"real-time enrichment, dedupe and territory sync\") and replaces it with at most one plain sentence a stranger would understand.",
"It reduces the stacked asks - the 30-minute calendar slot plus the right-person check - to exactly one ask.",
"It replaces the 30-minute calendar ask with an ask for the conversation already in progress.",
"It adds an honest identification with the rep's full name plus company.",
"It returns 2-3 variants, each isolating exactly one variable against the others."
]
},
{
"id": 6,
"prompt": "Northbeck Software. We tested 3 openers last Tuesday and Wednesday. Variant A (permission ask + funding-round mention + asks for 15 minutes): 38 conversations, 4 meetings. Variant B (reason-for-call + generic pain + asks for 15 minutes): 41 conversations, 3 meetings. Variant C (peer name-drop + \"worth a chat?\"): 36 conversations, 2 meetings. A wins at 10.5% so we roll it out Monday. I'm telling the board we'll hold an 11% meeting rate because that's Gong's number, and since Cognism says the industry dial-to-meeting rate is 2.3% we're 4-5x the field. All numbers come out of our dialer dispositions. Anything wrong with this?",
"expected_output": "A response that voids the test on sample size and confounded variables, corrects the Cognism metric definition, refuses to merge Gong with Cognism, and moves measurement off dialer dispositions onto recorded-call data with its disclosure cost named.",
"files": [],
"expectations": [
"The response states the sample is far below the judging threshold of roughly 100 conversations per variant.",
"It converts the threshold into connects using the 65.6% connect-to-conversation rate, giving roughly 150 connects per variant.",
"It states the gaps between A, B and C are within noise at this sample size, so no variant can be declared a winner.",
"It identifies that each variant changes more than one thing, so even a clean result would not say which change caused it.",
"It names the confounded dimensions specifically: opener pattern, pain wording and the ask all differ across A, B and C.",
"It corrects the Cognism figure: 2.3% is a conversation-to-meeting rate, not a dial-to-meeting rate.",
"It refuses to compare or combine Gong's ~11% with the Cognism rate.",
"It gives the reason: Gong measures meetings from connected calls while Cognism measures conversation-to-meeting, across different years and populations, with no shared baseline.",
"It declines to let the board be told an 11% meeting rate is the expected outcome.",
"It flags dialer dispositions as the weaker source, citing that roughly 40% of cold calls are mis-dispositioned or not dispositioned at all.",
"It recommends recorded-call duration data as the default measurement source instead.",
"It names the cost recording carries: a disclosure obligation and a retention decision, with consent rules varying by jurisdiction, cleared once per jurisdiction before the test starts.",
"It states the meetings-per-conversation benchmarks as 2.7% industry average against 11.3% for Cognism's own team, making the average a floor rather than a ceiling."
]
},
{
"id": 7,
"prompt": "Setting our team standard this week. I want the opener built on Jeb Blount's 5-Step Telephone Prospecting Framework - write out the five steps and map our script onto each one. Then wrap the whole thing in a Sandler up-front contract so we lock the agenda inside the first 20 seconds. And before I sign off on any of it, give me the measured conversion rate for a pattern-interrupt opener so I can compare it against the other options. We sell scheduling software to dental practice owners, 3-12 chairs.",
"expected_output": "A response that refuses to invent Blount's five steps, labels the compressed Sandler contract as an unmeasured adaptation, refuses to attach a conversion rate to pattern interrupt, and recommends a measured pattern for an owner persona.",
"files": [],
"expectations": [
"The response states that the five steps of Jeb Blount's telephone framework are not publicly documented and declines to write them out.",
"It does not invent, list, or paraphrase five named steps attributed to Blount.",
"It confirms the framework's name is verifiable without fabricating its contents.",
"It describes the Sandler up-front contract accurately: agreed purpose, both parties' agendas, time, and permitted outcomes including \"no\".",
"It states that compressing the up-front contract into a cold call's first 30 seconds is an adaptation rather than what Sandler specified.",
"It states the compressed version has not been measured on cold calls.",
"It refuses to give a measured conversion rate for the pattern-interrupt opener.",
"It gives the reason: no dataset isolates pattern interrupt, so a number would be false precision.",
"It places pattern interrupt last on the menu and highest on rep effort, noting it relies on live improvisation with no script to fall back on.",
"It recommends a measured pattern for this owner/SMB persona - direct \"reason for my call\" - instead of the three requested constructs."
]
},
{
"id": 8,
"prompt": "Quick one. We're a US company dialing UK B2B - Heads of Facilities at companies with 50-300 staff, all mobile direct dials from a data vendor. It's B2B so the do-not-call registers don't apply to us, right? We run the same motion into US B2B mobiles too. Rank the four opener patterns by compliance cost so I can pick the safest one to standardize on.",
"expected_output": "A response that corrects the B2B-exemption premise for both the UK and the US, names the specific registers and statutes, and refuses to rank opener patterns by compliance cost at all.",
"files": [],
"expectations": [
"The response corrects the premise: UK B2B calling requires screening both the TPS and the CTPS registers, plus the internal do-not-call list.",
"It distinguishes their coverage: TPS for individuals, sole traders and some partnerships; CTPS for companies and government bodies.",
"It notes that UK register entries take 28 days to take effect.",
"It states that under PECR live marketing calls generally do not require consent, but the subscriber must not have previously objected.",
"It states the US TCPA has no statutory B2B exemption, unlike the FTC's Telemarketing Sales Rule, which does.",
"It states that TCPA consent requirements for calls to cell phones apply whether the call is B2B or consumer.",
"It names mixed-use mobile numbers as the live exposure, citing the Ninth Circuit's Chennette decision on do-not-call standing.",
"It names at least two US states running mini-TCPA statutes with no B2B exemption, drawn from Florida, Oklahoma, Washington and Maryland.",
"It states that every business making telemarketing calls, B2B included, must keep an internal do-not-call list.",
"It refuses to rank the opener patterns by compliance cost, stating that consent and scrubbing duty attaches to the list and the jurisdiction identically for every pattern."
]
},
{
"id": 9,
"prompt": "Argent Loop, Series B, selling spend-management software. Two things. First, my head of sales keeps quoting that Gong proved personalization triples booking rates for director-level buyers on cold calls, so we've mandated 20 minutes of research per account across the entire list - including the manager-level and IC-level contacts. Second, these are fourth-attempt accounts; the first three calls went nowhere. Write an opener that lands like a fresh first call so we don't remind them we've been chasing them.",
"expected_output": "A response that corrects the Gong finding to email reply rates, applies the seniority conditionality to the research mandate, and refuses to disguise a fourth attempt as a first call.",
"files": [],
"expectations": [
"The response corrects the attribution: Gong's tripling finding is on prospecting-email reply rates, not on cold calls.",
"It states that carrying the email finding across to live calls is an inference rather than a measured result.",
"It states the seniority conditionality: the effect holds for directors and above, with little correlation for lower-level buyers.",
"It recommends against a uniform 20-minutes-per-account research mandate across the whole list.",
"It specifically advises against spending that research budget on the manager-level and IC-level contacts.",
"It refuses to write an opener that presents a fourth attempt as a first call.",
"It states the rule plainly: a re-attempt names the prior attempts honestly.",
"It grounds that refusal in the no-fabricated-familiarity gate rather than in style preference alone.",
"It promotes the direct \"reason for my call\" pattern because the list is a re-attempt.",
"It notes that roughly three attempts is optimal and about 93% of conversations happen by call 3, so a fourth attempt sits past the productive window."
]
}
],
"trigger_queries": [
{ "query": "Write a cold call opener for VP of Engineering at 500-person fintechs.", "should_trigger": true },
{ "query": "What should my SDRs say in the first 10 seconds when someone actually picks up?", "should_trigger": true },
{ "query": "I need a phone script for our new outbound team.", "should_trigger": true },
{ "query": "Our dialer talk track keeps getting hung up on. Can you rewrite the opening?", "should_trigger": true },
{ "query": "Give me three versions of the first thing a rep says on a cold call.", "should_trigger": true },
{ "query": "How do I open a call to a CFO without getting shut down immediately?", "should_trigger": true },
{ "query": "what do i say when they answer", "should_trigger": true },
{ "query": "Draft the first 20 seconds of a prospecting call to hospital procurement managers.", "should_trigger": true },
{ "query": "We're starting outbound calling next month and nobody knows what to say when someone answers.", "should_trigger": true },
{ "query": "Need an opening line for calls to homeowners who requested a solar quote on our site.", "should_trigger": true },
{ "query": "Can you write a permission-based opener for our AEs?", "should_trigger": true },
{ "query": "My reps sound like telemarketers in the first five seconds. Fix it.", "should_trigger": true },
{ "query": "Build a talk track for the very top of an outbound dial.", "should_trigger": true },
{ "query": "I want A/B test variants of our cold call intro, with a plan for judging them.", "should_trigger": true },
{ "query": "What's the best way to start a cold call to a Director of IT?", "should_trigger": true },
{ "query": "Script the moment right after \"hello\" for our insurance agents.", "should_trigger": true },
{ "query": "We get hung up on inside 8 seconds. What are we doing wrong at the start of the call?", "should_trigger": true },
{ "query": "Give me an opener that doesn't sound like every other sales call they get.", "should_trigger": true },
{ "query": "Need something to say to dental practice owners when they pick up the phone.", "should_trigger": true },
{ "query": "Write the intro for a cold call plus notes on how to deliver it.", "should_trigger": true },
{ "query": "Our new SDR class starts Monday and needs something to say on day one of dialing.", "should_trigger": true },
{ "query": "How should a rep introduce themselves on an outbound call to a total stranger?", "should_trigger": true },
{ "query": "Cold calling French consumers next quarter - what do we say, and what do we have to check before dialing?", "should_trigger": true },
{ "query": "I want to test two different first lines on the phone and know which one actually wins.", "should_trigger": true },
{ "query": "Draft phone outreach openings for the three different personas we sell into.", "should_trigger": true },
{ "query": "What's a good reason-for-call line for our product?", "should_trigger": true },
{ "query": "Rewrite this, it's killing us: \"Hi, is now a bad time?\"", "should_trigger": true },
{ "query": "opener for outbound dials to supply chain directors pls", "should_trigger": true },
{ "query": "We're re-dialing accounts that ignored us twice already. What do we lead with?", "should_trigger": true },
{ "query": "How long should a rep talk before the prospect says anything on a cold call?", "should_trigger": true },
{ "query": "Need a call intro for our B2C mortgage refinance campaign.", "should_trigger": true },
{ "query": "Give me the first 30 seconds of a call to a CISO.", "should_trigger": true },
{ "query": "Our team's phone intro needs a rewrite - right now it runs 90 seconds before they can speak.", "should_trigger": true },
{ "query": "What's the best opening line on a cold call, backed by actual data and not vibes?", "should_trigger": true },
{ "query": "Script the top of a call for reps who have never dialed a stranger before.", "should_trigger": true },
{ "query": "I need three testable versions of how we start calls, with delivery notes for each.", "should_trigger": true },
{ "query": "Help me stop reps apologizing at the start of every single call.", "should_trigger": true },
{ "query": "We're calling UK businesses cold. What do we open with, and which registers do we check first?", "should_trigger": true },
{ "query": "Design what a high-velocity inside sales team says at the top of every dial.", "should_trigger": true },
{ "query": "Our connect-to-conversation rate is 38%. What do we change about how the call starts?", "should_trigger": true },
{ "query": "Write out what a rep says before the prospect gets a word in.", "should_trigger": true },
{ "query": "We bought a list and start dialing Monday. What's the script for the first few seconds?", "should_trigger": true },
{ "query": "Turn our value prop into something a rep can say out loud in 20 seconds on the phone.", "should_trigger": true },
{ "query": "How do I earn the first seven seconds on the phone with someone who didn't ask to hear from me?", "should_trigger": true },
{ "query": "Need an opening for calls to consumers who filled in a form on our site last week.", "should_trigger": true },
{ "query": "They said \"not interested\" straight away - what's my rebuttal?", "should_trigger": false },
{ "query": "Build me an objection library for price pushback.", "should_trigger": false },
{ "query": "Score this cold call transcript against a rubric and tell me what to coach.", "should_trigger": false },
{ "query": "Here's a call recording transcript - how did my rep do?", "should_trigger": false },
{ "query": "How many touches should our cadence have before we stop chasing?", "should_trigger": false },
{ "query": "Design a 14-day sequence mixing calls, emails and LinkedIn touches.", "should_trigger": false },
{ "query": "Write a subject line for our cold email to CTOs.", "should_trigger": false },
{ "query": "Nobody opens our emails - generate and test some subject lines.", "should_trigger": false },
{ "query": "Our cold emails are landing in spam. Check our SPF and DMARC setup.", "should_trigger": false },
{ "query": "How do I warm up a new sending domain for outbound email?", "should_trigger": false },
{ "query": "What should I ask on the discovery call next Tuesday?", "should_trigger": false },
{ "query": "Build a pain-funnel question set for a first meeting with a COO.", "should_trigger": false },
{ "query": "Find me a reason to reach out to this account - they just raised a Series B.", "should_trigger": false },
{ "query": "Turn these trigger events into ranked personalization angles for outreach.", "should_trigger": false },
{ "query": "Write the follow-up email after today's sales call with the agreed next steps.", "should_trigger": false },
{ "query": "Draft a mutual action plan from these call notes.", "should_trigger": false },
{ "query": "Score this opportunity against MEDDPICC.", "should_trigger": false },
{ "query": "Is this deal real? Here are the notes from the last three calls.", "should_trigger": false },
{ "query": "Who's the actual economic buyer in this account?", "should_trigger": false },
{ "query": "Map the buying committee for this deal and flag single-threading risk.", "should_trigger": false },
{ "query": "The buyer wants 20% off - plan my concessions and what I get back.", "should_trigger": false },
{ "query": "Build the ROI case the champion can forward to their CFO.", "should_trigger": false },
{ "query": "Define our ICP from last year's closed-won deals.", "should_trigger": false },
{ "query": "How big is the addressable market for warehouse robotics?", "should_trigger": false },
{ "query": "Set next year's AE quotas with ramp adjustment for the new hires.", "should_trigger": false },
{ "query": "Design the SDR commission plan - pay mix and accelerators.", "should_trigger": false },
{ "query": "Should we split SDRs from AEs or go full-cycle?", "should_trigger": false },
{ "query": "Are we PLG or sales-led? We're about to add a sales team on top of self-serve.", "should_trigger": false },
{ "query": "Write the interview scorecard for hiring two SDRs.", "should_trigger": false },
{ "query": "I'm prepping for an SDR interview next week and they'll make me do a mock cold call. Coach me.", "should_trigger": false },
{ "query": "Which sales podcasts and newsletters should I be following this year?", "should_trigger": false },
{ "query": "Where do I even start with sales? We have no process at all.", "should_trigger": false },
{ "query": "Build our account fit score from firmographic and intent signals.", "should_trigger": false },
{ "query": "Where should the Tier 1 and Tier 2 cutoffs sit in our account book?", "should_trigger": false },
{ "query": "We had 4x pipeline coverage and still missed. Model the right ratio.", "should_trigger": false },
{ "query": "Write a cold email to a VP of Marketing about our analytics product.", "should_trigger": false },
{ "query": "Prep me for my call with Acme tomorrow - it's a scheduled demo they booked.", "should_trigger": false },
{ "query": "Summarize this call recording into structured notes for the CRM.", "should_trigger": false },
{ "query": "Set up the on-call handoff process for our platform team.", "should_trigger": false },
{ "query": "Implement outbound voice calling in our app with Twilio.", "should_trigger": false },
{ "query": "Our support agents need a greeting script for inbound customer service calls.", "should_trigger": false },
{ "query": "Write the LinkedIn connection request message for this prospect.", "should_trigger": false },
{ "query": "What do I say on the second call, after they've already agreed to talk?", "should_trigger": false },
{ "query": "Coach my rep on closing techniques for the end of the call.", "should_trigger": false },
{ "query": "Build the 30-60-90 ramp plan for our new sales hire.", "should_trigger": false }
]
}
SKILL.md
---
name: cold-call-opener
description: Writes the opening 5-30 seconds of a live outbound cold call - a tailored opener for a named persona and pain point, with 2-3 testable variants, delivery notes, and a measurement plan. Covers B2B and B2C phone outreach, including do-not-call scrubbing. Use whenever the user mentions a cold call opener, cold call script, phone opener, dialer talk track, "what do I say when they pick up", or opening a call for a persona, even if they don't say "opener" explicitly. Do NOT use for rebuttals later in the call (use mbfinotti/sales-skills@sales-objection-handling), grading a recorded call (mbfinotti/sales-skills@sales-call-review), or cadence planning (mbfinotti/sales-skills@sales-outbound-sequence).
license: MIT
metadata:
author: Maya-Beth Finotti
version: "1.3.3"
---
# Cold Call Opener
Draft what a rep says between "hello" and the prospect's first real response. The opener ends the moment the prospect engages or brushes off - everything after that belongs to a neighboring skill (see References). The published evidence on opener wording comes almost entirely from two Gong datasets that contradict each other, plus Cognism's funnel data. This skill reports each source honestly instead of blending them into a fake industry benchmark.
## Interview
Ask before drafting anything. One question per message; offer multiple-choice answers when possible. Skip any question the user already answered.
- B2B or B2C? (If both, draft separate openers - the pain framing and the legal scrubbing differ.)
- Who picks up: title and seniority? (IC / manager / director-VP / executive / business owner / consumer.)
- What does the product do, in one plain sentence a stranger would understand? Refuse feature lists; one sentence.
- What pain do you believe this person has, in their words?
- Where did that pain hypothesis come from: a verified signal (job posting, funding, tool change, something they said or submitted) or an educated guess? This decides the wording, not just the confidence.
- Desired outcome of the call: booked meeting / qualification / referral to the right person / (B2C) quote, appointment, or sign-up?
- List temperature: fully cold, warm-ish (prior touch, inbound form), or a re-attempt?
- If a re-attempt: what happened on the previous attempts?
- How does the rep naturally talk: direct, warm, understated, high-energy? (Or paste two sentences of how they actually speak.)
- Which market/jurisdiction is being called: US / UK / EU member state (which one) / France / other? For B2C this drives mandatory scrubbing - see [references/consumer-calling-constraints.md](references/consumer-calling-constraints.md).
- By what date must this opener be booking meetings? (Inside two weeks promotes the cheapest pattern and caps the test at one variable; a quarter or more makes the research-heavy pattern affordable.)
- One-off win or compounding asset: a script for one campaign push, or the team's default opener for the year? (One-off promotes direct reason-for-call; compounding promotes permission-based, whose rehearsal cost is paid once and collected on every call after.)
- Effort ceiling: rehearsal hours per rep, months of phone tenure behind them, and minutes of pre-call research affordable per account. (Near-zero research minutes removes context-first/peer-proof from the menu for this run; a first-week rep promotes direct reason-for-call to first.)
## Opener anatomy
Four components, one length budget. The whole opener runs 5-30 seconds of rep speech, then stops.
Stay short for two reasons:
- Chris Beall's framing: a cold caller has roughly seven seconds to establish trust, since curiosity, not value, is what avoids triggering reactance.
- Jason Bay's observation, quoted in Cognism's 2025 report: only 10% of cold calls make it past 2 minutes. An opener that eats 45 seconds has already spent a third of the median call before the prospect says a word.
1. **Identification** (~3-5s). Full name plus company. Honest, unhurried, no fake familiarity.
2. **Reason-for-call** (~5s). One sentence naming why this person, today. Gong credits John Barrows with the test: if the rep cannot finish "the reason for my call is...", do not make the call.
3. **Pain hypothesis** (~10s). One spoken sentence the prospect recognizes as their own problem. Rules in Pain-hypothesis line below.
4. **Ask / transition** (~5s). Exactly one ask, matched to the desired outcome from the interview. Then stop talking.
Working cap: about 75 words, a self-set bar rather than a published one. Verify with a read-aloud timer, not a word counter alone.
## Workflow
1. Run the Interview; collect every answer first.
2. If you can browse the web, verify the pain signal (job posting, funding news, public statement); otherwise ask the user for the evidence. Never invent a signal. For turning raw signals into angles, see `mbfinotti/sales-skills@sales-outreach-personalization`.
3. Enter brainstorming mode: propose 2-3 candidate angles (each = one opener pattern + one pain framing). Score every candidate on the three axes Choosing the pattern uses:
- Measured value.
- Rep effort (rehearsal, live composure, how it fails in unskilled hands).
- Conversations needed before it can be judged.
Recommend one and name the axis that decided it. Ask questions one at a time. Draft nothing until the user picks an angle.
4. Pick the pattern with the ranking in Choosing the pattern; open [references/opener-patterns-and-evidence.md](references/opener-patterns-and-evidence.md) only for the full evidence trail behind a rate or an attribution.
5. Draft 2-3 variants. Each variant isolates exactly one variable (pattern, pain wording, or ask) against the others; a variant that changes two things teaches nothing.
6. Write delivery notes per variant (Delivery below).
7. Run your preferred humanizer skill over every variant, then read each aloud with a timer. A cold call opener is speech; raw first-draft model output is never the final deliverable.
8. Score every variant against the Self-check gate; iterate until all items pass.
9. Hand off the artifact in Expected output, including the test plan.
10. If your harness has persistent memory, memorize the approved persona, pain hypothesis, product sentence, and - once tested - the winning opener; later runs then skip the interview.
## Choosing the pattern
Rank patterns by what they return per unit of rep effort, and say the ranking out loud instead of letting row order imply it. Effort here is rehearsal time, live composure, and how badly the pattern fails in an unskilled rep's hands - a pattern that is strong for a veteran and damaging for a new rep is high-effort, not high-value. Full evidence, attributions, and the Gong 2018/2024 contradictions sit in [references/opener-patterns-and-evidence.md](references/opener-patterns-and-evidence.md).
The axes disagree, so each gets its own line:
- efficiency: permission-based > direct reason-for-call > context-first/peer-proof > pattern interrupt
- value: context-first/peer-proof == permission-based > direct reason-for-call
- effort: pattern interrupt > context-first/peer-proof > permission-based > direct reason-for-call
That `==` is a real tie, not a way of avoiding the call. Gong's 2024 study separates those two by 0.06 percentage points and publishes no per-opener sample size, so nothing in the data can order them. They split on effort instead.
No compliance-cost axis belongs on this menu. Consent and scrubbing duty attaches to the list and the jurisdiction, identically for every pattern (see [references/consumer-calling-constraints.md](references/consumer-calling-constraints.md)).
1. **Permission-based** - the default rung.
- Value: ~11% of connected calls produced a held meeting in Gong's 2024 study, roughly 5x that study's worst-scoring opener.
- Effort: about an hour of rehearsal, paid once and collected on every call after. The rep names the interruption without flinching, then actually stops talking.
- Fails softly: rushed, it sounds like an apology, which costs that call and nothing else.
- Guessed pain routes here, with the pain line hedged (see Pain-hypothesis line). A guess asserted as fact hands over the top-five brush-off "I don't have the problem you're describing" (Cognism 2025).
2. **Direct "reason for my call"** - the cheapest, and the only one that composes.
- Value: about double baseline in Gong's 2018 study, the smallest measured lift on this menu. Gong describes it as pairable with any other opener, so it never fully loses a round.
- Effort: minutes. One sentence, tested by John Barrows' rule: if the rep cannot finish "the reason for my call is...", do not make the call.
- Promote it to first when any of these hold: a hard deadline, a rep in their first weeks, a re-attempt (name the prior attempt honestly, never pretend it is the first call), or an owner, SMB, or consumer line.
- On a consented B2C list, the consent event _is_ the reason for the call ("you asked for a quote on our site Tuesday").
3. **Context-first / peer-proof** - tied for best measured value, an order of magnitude more effort.
- Effort: pre-call research per account, plus the composure to defend a peer claim when the prospect challenges it live.
- Fails hardest: an unsupportable name-drop or surveillance-flavored trivia loses the call outright, the personalization-backfire pattern documented in the reference.
- State the verified signal before the ask.
- Promote it above permission-based only when all three hold: director or above, a verified strategic signal already in hand, and a rep who has run the play before.
4. **Pattern interrupt** - last, and deliberately unranked on value. No dataset isolates it, so a number here would be false precision dressed as evidence.
- Effort: high. Live improvisation, no script to fall back on.
- Practitioners report the payoff decays as a novelty goes mainstream. Reported, not measured.
- Use sparingly.
**Deleted, not demoted.** Small-talk openers ("How have you been?", "How's your day going?") are off this menu.
- The only forms either Gong dataset measured are addressed to a stranger, which fails this skill's own self-check gate on fabricated familiarity.
- An option the constraints rule out has to be removed: parked at the bottom, it comes back as a variant.
- Its evidence stays in the reference as a contradiction to report, never as an option to offer.
**What this order starves.** Context-first/peer-proof ties for the highest measured value and loses every efficiency round on research cost, the standard failure of ratio-ranking. Promote it deliberately on a short, durable list of high-value director+ accounts, where per-account research is affordable.
Gong's seniority conditionality applies here: company-specific personalization pays most at director+, measured on email reply rates. Carrying it to calls is therefore an inference rather than a finding.
**Re-rank before drafting.** This order is a default, not a law: it shifts with context and with who executes it. Re-rank against what you already know about the user, then name which interview answer moved which pattern when presenting the choice.
- A veteran rep moves context-first up.
- A first-week rep moves it off the menu.
- A warm list or a re-attempt promotes direct reason-for-call.
- A persona who rarely picks up (B2C especially, see B2B vs B2C) raises the effort cost of every connect, which promotes the cheap patterns further.
## Pain-hypothesis line
- Write one sentence a prospect would recognize as their own words. Test: would they say this out loud themselves? If not, rewrite from their vocabulary, not the product's.
- Zero jargon, zero feature language. The line names the problem, never the solution.
- Concrete beats abstract: "your SDRs are booking off a list that's a year stale" beats "data quality challenges".
- Verified signal: lead with the observation and let it carry the relevance - it needs no hedging.
- Guessed pain: hedge it as something heard from peers ("Most ops leads I talk to at firms your size tell me X - I'm guessing that's on your plate too, but you'd know better"). Hedging makes disagreement cheap for the prospect instead of confrontational, and turns the assertion into a question.
- Never state a guess as a fact about their company, and never imply research the rep did not do.
## B2B vs B2C
Identical in both - say so, do not re-derive:
- The four-part anatomy.
- The 5-30 second budget.
- The one-ask rule.
- The pause discipline.
- The humanizer pass.
- The variant-testing method.
What differs:
- **Who answers.** B2B: the titled persona from the interview, behind a switchboard or direct dial. B2C: whoever owns a personal line - Hiya's State of the Call 2026 (a survey of 12,000+ consumers across six countries) found only 14% immediately answer calls from unrecognized numbers, so expect a lower answer floor and more re-attempts.
- **Pain framing.** B2B: a business outcome the persona is measured on. B2C: money, time, or hassle in plain household words - no business vocabulary at all.
- **The ask.** B2B: usually a calendared meeting or a referral. B2C: usually the conversation itself - a quote, an appointment, a sign-up - so the transition often continues the same call rather than booking a future one.
- **Regulatory scrubbing.** Consumer calling sits under consent-based regimes. B2B is mostly carved out, with exceptions. Before any B2C list is dialed: check the jurisdiction's do-not-call register and consent basis, and keep an internal do-not-call list. Jurisdiction specifics (US, EU, UK, France - including France's consumer opt-in regime effective 11 August 2026) are in [references/consumer-calling-constraints.md](references/consumer-calling-constraints.md). Factual note, not legal advice: have counsel confirm before a campaign.
## Delivery
- Pace: unhurried. Rushing signals fear; the identification especially gets said slowly.
- Tone: match the rep's natural register from the interview - a script in someone else's voice reads as a script.
- Stand up, or sit upright: breath support carries energy down a phone line. Practice advice, not a measured effect.
- After the ask: pause, and do not fill the silence. The ask ends the opener; whoever speaks next is the prospect.
- **Rhythm.** Gong's call data shows successful cold calls are not interrogations.
- No statistical difference in the number of questions asked.
- Successful calls contain _longer_ rep monologues (37s vs 25s longest burst).
- Prospects' longest turns are _shorter_ on successful calls (3.5s vs 8s).
- Successful calls show 77% more speaker switches per minute.
Read together, this means: deliver the opener as one confident monologue, then stop cleanly. Keep the exchange brisk after that, with short turns and fast back-and-forth.
Do not read this as "talk over the prospect." Do not teach "listen more, talk less" as if the data supported it. On these measures, it points the other way.
## Variants and testing
- Produce 2-3 variants per run, never one. One variable isolated per variant.
- Sequence the test: run variants in alternating blocks on comparable list segments, same daypart, same rep.
- Judging threshold: do not compare variants before roughly 100 conversations each - a working convention, since no vendor publishes a cold-call sample-size benchmark. At Cognism's 2025 connect-to-conversation rate (65.6%), that is roughly 150 connects per variant. That is the unit price of every option on the pattern menu: each variant kept costs another ~150 connects before it says anything.
- Treat small gaps as noise: the 0.06-point gap that ties the top two patterns in Choosing the pattern is the scale of what a test this size cannot separate. A difference that small is a coin flip, not a winner.
## Measurement
Track per variant, in the dialer or call recording tool:
- **Connect-to-conversation rate** - connects that become a real dialogue. Benchmark: 65.6% (Cognism 2025, computed from 41,936 connects producing 27,513 conversations). This is the opener's own KPI: the opener is what turns a pickup into a conversation.
- **Meetings booked per conversation.** Benchmark: 2.7% industry average (Cognism 2026); Cognism's internal team reports 11.3%, so the average is a floor, not a ceiling.
- **Early hang-up rate** - share of connects ending inside ~30 seconds. No vendor publishes this as a benchmarked KPI; compute it from call-duration distributions (see the source ranking below).
- **Conversation length** - trend it against the team's own baseline only. Cross-vendor averages disagree (Cognism 82-93s; Chorus 132s; Gong 5:50 successful vs 3:14 unsuccessful) because each measures different populations with different definitions.
Two places those numbers can come from, and the axes disagree:
- value: recorded-call duration data > dialer dispositions. Chorus found 40% of cold calls are mis-dispositioned or not dispositioned at all, so dispositions measure rep habits as much as prospect behavior.
- compliance cost: recorded-call duration data > dialer dispositions. Recording triggers a disclosure obligation and a retention decision, and consent rules vary by jurisdiction (all-party consent in several US states, notification duties in the EU and UK). Dispositions trigger neither. Cost it as the review it forces - legal sign-off once per jurisdiction dialed - and as reversibility: a recording cannot be un-made.
- Default: recorded duration data, with the disclosure cleared once per jurisdiction before the test starts. Fall back to dispositions only where recording is not cleared, and mark the early hang-up number directional.
Iteration target this skill chases: connect-to-conversation at or above the 65.6% Cognism benchmark, and meetings per conversation at or above the 2.7% average. If a variant sits below either after the judging threshold, change one variable and rerun. Keep iterating until a variant clears both.
## Self-check gate
Every variant passes all items before handoff. Iterate - redraft, re-run the humanizer - until none fail.
1. Read aloud at natural pace in 30 seconds or less (~75-word working cap).
2. Exactly one ask, matching the outcome the user chose in the interview.
3. Pain sentence passes the would-they-say-it-out-loud test; no jargon; no feature language.
4. Guessed pain is hedged; verified pain names its real signal.
5. No fabricated familiarity, no implied prior relationship, no research claims the rep cannot back.
6. Identification is honest: full name plus company.
7. Ends on the ask followed by a planned pause - no trailing filler.
8. Reads human aloud after the humanizer pass.
9. B2C only: list scrubbed for the jurisdiction and consent basis confirmed.
## Common failure modes
| Failure | Fix |
| --------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------- |
| Sounds like a script | Humanizer pass + read-aloud; contractions; the rep's own vocabulary from the interview |
| Over-long opener | Cut to one pain sentence; 30-second timer is the law |
| Fake familiarity ("How have you been?" to a stranger) | Cite a real signal or drop the claim; small-talk openers are off the pattern menu, not a judgment call |
| Stacked questions | One ask; park the rest for discovery |
| Asking for time the rep does not need ("Can I get 30 minutes?") | Ask for the conversation already in progress |
| Feature dumping | The opener names a pain, never describes the product beyond the one-liner |
| Apologizing for calling ("sorry to bother you") | Own the interruption instead - the permission-based pattern names it without apology |
| Inviting the exit ("Did I catch you at a bad time?") | Both Gong datasets rank better options (0.9% in 2018; 2.15% in 2024 - the two disagree on the baseline, not on the ranking) |
| Treating the opener as discovery | One hypothesis, one ask; question batteries belong after engagement |
## Expected output
```
COLD CALL OPENER - <persona>, <product>, <date>
Chosen pattern : <pattern> + the axis that decided it against the other candidate angles
Variant A/B/C : verbatim opener + the one variable it isolates
Delivery notes : pace, pause points, tone - per variant
If brush-off : hand to mbfinotti/sales-skills@sales-objection-handling
If engaged : hand to mbfinotti/sales-skills@sales-discovery-questions
Test plan : conversations per variant, KPIs, thresholds, judge date
Compliance : (B2C) jurisdiction + scrub/consent confirmation
```
Full worked examples - two B2B, one B2C, one annotated negative - are in [references/example-openers.md](references/example-openers.md), never inline here.
## Reference
- See [references/opener-patterns-and-evidence.md](references/opener-patterns-and-evidence.md) after choosing a pattern - the evidence trail, attributions, and dataset contradictions behind the rates quoted in Choosing the pattern.
- See [references/example-openers.md](references/example-openers.md) before drafting - worked B2B and B2C openers plus a negative example with the reasons it fails.
- See [references/consumer-calling-constraints.md](references/consumer-calling-constraints.md) before dialing any B2C list - per-jurisdiction scrubbing notes.
- See `mbfinotti/sales-skills@sales-objection-handling` for what follows a brush-off.
- See `mbfinotti/sales-skills@sales-discovery-questions` for what follows engagement.
- See `mbfinotti/sales-skills@sales-call-review` for scoring recorded calls against a rubric afterward.
- See `mbfinotti/sales-skills@sales-hiring` for the hiring loop's mock-call work sample, which draws on this skill's opener craft when writing realistic scenario material for an SDR/high-velocity candidate exercise.