SYSTEM RUNNING · SEOUL · 2026.04 PORTFOLIO · v1.0
Outbound Operator / Systems Builder

Shin.

Outbound operator turning B2B sales into a system.

4½ years on the front line — running F&B businesses, then leading global B2B sales for medical aesthetics. So I built a B2B GTM automation system in Claude Code, solo, in 2 weeks. 18 AI agents now run like a full SDR team — managed by a single operator.

Shin — Outbound Operator
Shin · Head of Global Sales, iCELmedi 2026
01
0½ yrs
F&B + Global B2B SalesExperience
02
0 wks
Solo build · 2026.04.06–04.20Build
03
0
AI agents · 5-stageAgents
04
0
Monthly capacity · emailsOutput
Scroll to read ↓
THE PROBLEM

B2B outbound has been stuck for 25 years

The average SDR personalizes 80–100 messages a month. Two or three turn into meetings.
Tools have changed every year, but productivity hasn't — because of three structural bottlenecks.

01

Lead discovery eats too much time

Narrowing pipeline
Weekly lead capacity leads / week · per sdr
ICP filter
10 Usable
20 Filtered · −67%
0102030

One SDR combing LinkedIn + Apollo by hand caps out at 20–30 leads/week. Run an ICP filter on top of that and the usable pool drops below 10/week.

02

You give up either personalization or volume

Stalled ROI
Personalization × Volume frontier
Manual Option A
Personal
9.2
Volume
1.8
Cost 40 min / lead
or Want both? Need 2–3× headcount
Template Option B
Personal
2.2
Volume
9.2
Cost reply 1–2%

Lean on templates and reply rate stalls at 1–2%. Personalize by hand and you burn 40 minutes per lead. Want both? Hire 2–3× more people.

03

Next week looks just like last week

No growth curve
Reply rate · 12 weeks slope ≈ 0
Weekly log · run by hand MANUAL
W01Weekly review 2–3h · cohort compare ✕ · angle test ✕
W02Weekly review 2–3h · cohort compare ✕ · angle test ✕
W03Weekly review 2–3h · cohort compare ✕ · angle test ✕
W04Weekly review 2–3h · cohort compare ✕ · angle test ✕
…( same routine, repeated )
W11Weekly review 2–3h · cohort compare ✕ · angle test ✕
W12Weekly review 2–3h · cohort compare ✕ · angle test ✕
reply rate · W12 1–2%Δ 0%
Weekly log · data-driven automation AUTOMATED
W01Baseline captured. 4 cohorts × angles A/B/C deployed
W02Angle A open rate +18% · auto-shift A weight ↑
W04ICP-B retired. reply 0.3% → resources reallocated
W063 pain-hook templates promoted · angle C dropped
W08Tier A expanded. 5 hot-lead signals auto-learned
W10Send-time tuned · Tue 10am cluster reply +40%
W12Reply 7% reached. 3 next-week hypotheses auto-proposed
reply rate · W12 7%Δ +9×

2–3 hours on the weekly spreadsheet review. Cohort and angle comparisons never happen, and next week starts the same. Learning never compounds.

Shin's Answer · Organizational Structure

Humans for judgment,
18 agents for execution.

To solve the three bottlenecks I hit working as a B2B sales operator, I built a GTM automation system structured as 1 operator + 18 AI agents across 5 teams. Machines handle research, personalization, sending, and analysis. The operator owns only one thing: does this email actually land in the field?

Core
Principle
Machines do the research and write the drafts.
Humans judge only whether the email lands in the field.
ORG · CHART 1 Operator · 18 Agents · 5 Teams
● HUMAN · L0
GTM Operator

Context-dependent · final calls · quality assurance. The human handles only what doesn't scale.

MANAGES 18 Agents
▾ Handoff to AI Layer
TEAM 01

Research & Scoring

TAM collection · ICP 120-pt scoring · Tier classification

A01TAM Collector
Pulls target accounts from public DB & web
A02ICP Scorer
6-axis × 120-pt automatic scoring
A03Tier Classifier
Score → Tier A/B/C, automatic
TEAM 02

Personalization

Public-source mining · pain hook · email drafts

A04Public Data Miner
Structures public data per company
A05Pain Hook Extractor
Extracts 3 concrete pain points
A06Draft Writer · A/B
Generates Angle A & B drafts
A07Draft Writer · C
Generates Angle C (control) drafts
TEAM 03

Delivery & Sequencing

3-angle send · deliverability · reply classification

A08Sequence Orchestrator
Runs 3-angle × 3-touch schedule
A09Deliverability Monitor
Tracks domain & IP reputation
A10Reply Classifier
Auto-classifies positive · OOO · decline · query
A11Bounce Handler
Auto bounce & suppress handling
TEAM 04

Engagement Intelligence

Behavior score · Tier rebalance · Hot Lead

A12Behavior Scorer
Open · click · reply → behavior score
A13Tier Re-adjuster
Behavior signals → auto Tier rebalance
A14Hot Lead Detector
Detects buy signal → instant alert
A15Intent Aggregator
Aggregates external intent signals
TEAM 05

Operations

Nurture DB · weekly reporting · AE handoff

A16Nurture DB Manager
Manages follow-up timing & content library
A17Weekly Reporter
Weekly funnel · per-angle performance report
A18AE Handoff Packager
Builds Hot Lead context package
Outcome · What This Structure Produces
01 · VOLUME
13×

Lead-discovery speed

200–400 emails/month manually becomes 2,400/month capacity. Quality held; designed volume 13× the manual ceiling.

02 · QUALITY
4×

Personalization quality

Per-company structured public-data DB + concrete pain hooks. Reply rate target 1–2% → 5–7% · measured in Phase 2.

03 · ECONOMICS
$30

Per-Meeting Cost

Industry CPM $500–2,000 vs this design's projection.
Tools-only basis · assumes 11 meetings/month.
Re-measured after Phase 2 accrues.

Funnel · Design Scenario

From 50,000 accounts to 11 meetings/month

Pipeline scenario designed against the medical aesthetics device segment.
Per-stage conversions are planned ceilings against industry benchmarks · calibration with Batch 1 actuals.

Final outcome · Target Qualified Meetings MEDDIC 5-axis qualified · target median
11/ month

Phase 1 actuals · Batch 1 deployed (NL 60 · SE 12 · HU 5 · 77 total) · funnel above is the design ceiling at steady state.

Lead distribution by Tier

120-pt ICP score · A / B / C split

Leads are graded and each grade gets a different messaging · cadence · review depth. A goes hand-held, B auto-personalizes and tests, C drops into nurture and waits. Numbers are the design scenario · actual distribution locks after Batch 1.

Tier A · 70+ pts
120/ month

100% manual review + priority reply. Shin runs MEDDIC qualification personally. The cohort most likely to respond.

Tier B · 50–69 pts
680/ month

Tests 3 messages A/B/C simultaneously. Auto-personalization (9-min pipeline). Whichever angle lands gets a heavier weight next week.

Tier C · 30–49 pts
1,600/ month

Moved to nurture DB · 6-month re-evaluation. When the account's context changes, auto-promotion to A/B (see next section).

System Architecture · Bowtie

A pipeline where the first send compounds into lifetime value

60–70% of B2B revenue comes not from the first contract but from repeat purchase and referrals. I dissected the Winning by Design bowtie into 8 stages so that each conversion bottleneck can be tracked independently. In iCELmedi Pro Direct Phase 1, the left 5 stages (Research → Close) are live; Onboard · Retain · Grow activate after the first deal closes.

Operational (Phase 1) Core · Close Designed · activates after first close Hover to see stage detail
01 Research STAGE 01 1 agent 02 Prospect STAGE 02 3 agents 03 Outreach STAGE 03 2 agents 04 Engage STAGE 04 2 agents 05 Close CORE 06 Onboard STAGE 06 2 agents 07 Retain STAGE 07 2 agents 08 Grow STAGE 08 4 agents
◁   ACQUISITION · PHASE 1 LIVE THE NARROWING DESIGNED · POST-CLOSE   ▷
STAGE 05 / 08

Close CORE

From reply to signature — the bowtie's center

●   CORE STAGE
GOALMove replied leads through meeting → proposal → signature on the shortest path, while structuring the captured pain · decision-maker · budget data in CRM as input for the next stages (Onboard · Retain · Grow).
Assigned agents · 4 AGENTS
Stages
8stages
Research → Grow, full-funnel dissection
Operational Now
5/ 8
Phase 1 — Research → Close
Agents
18total
12 live · 6 designed
LTV efficiency
5–25×
Expansion vs new acquisition
WHY 8 STAGES

The 3-stage view (Acquisition · Commit · Expansion) is an explanation; the 8-stage view is the anatomy of the bottleneck. Separating Engage from Close exposes that "why don't they reply" and "why don't they sign" are completely different problems — and only then can you fix the right one.

The right half (Onboard · Retain · Grow) is designed-only at iCELmedi today, but in SaaS · subscription · repeat-purchase environments it activates immediately. When B2B acquisition costs 5–7× more than retention and expansion, this is where the full-funnel operator separates from the outbound specialist.

ICP Scoring · 120 Points · 6 Axes

A good lead isn't a hunch —
it's a score.

120 points distributed across 6 weighted axes, each mapped to an automated data source. A new lead gets scored in 10–15 seconds via Apollo + Claude web research — the goal is to remove subjectivity from threshold decisions. Weights and thresholds will be calibrated against Batch 1 results · scores below are initial design values.

① Region Fit 24 / 25 ② Size Fit 20 / 20 ③ Category Gap 18 / 25 ④ Buy Intent 15 / 20 ⑤ DM Access 10 / 15 ⑥ Competitor Pressure 5 / 15
Ideal ICP · 120 / 120 Sample score · Madrid distributor (illustrative)
01
Region Fit
HQ in target region · primary sales coverage match. Drives distance · logistics · regulatory cost.
Apollo Public website
25
POINTS
02
Size Fit
Headcount & revenue inside the ICP band. Too big → slow approvals; too small → deal size insufficient.
Apollo LinkedIn
20
POINTS
03
Category Gap
Real portfolio gap our product can fill. Cross-checked against the competitor lineup.
Public website Industry news
25
POINTS
04
Buy Intent Signals
Last 6 months — new product launch · certifications · job postings · partnership announcements.
Claude web research Industry news LinkedIn
20
POINTS
05
DM Accessibility
Decision-maker LinkedIn activity · public email available · seniority clarity.
LinkedIn Apollo NeverBounce
15
POINTS
06
Competitor Pressure
Direct competitor moves in the market. Urgency rises on competitor wins, churn, or pricing shifts.
Industry news Claude research Public reports
15
POINTS
Worked Example · Illustrative scenario
Example · 5–20 person aesthetic distributor · Madrid

Dead-center ICP · only weakness is competitor-pressure axis · scoring-logic test

92 / 120
TIER A · top score
01
Region Fit
Madrid — core target city in EU
24 / 25
02
Size Fit
12 people — dead center of ICP band (5–20)
20 / 20
03
Category Gap
No PDRN/peptide line — portfolio gap confirmed
18 / 25
04
Buy Intent Signals
Recent K-beauty line addition + 2 new hires
15 / 20
05
DM Accessibility
CEO active on LinkedIn · public email confirmed
10 / 15
06
Competitor Pressure
Competitor B launched Korean PDRN line this month — first-mover risk
5 / 15
TIER A
90+ pts
100% manual review then personalized send. Top priority of the monthly cohort.
TIER B
70 – 89 pts
Automated test sequence. Promotes to A on response.
TIER C
50 – 69 pts
Moved to Nurture DB · re-scored on signals after 6 months.
DROP
< 50 pts
Excluded from this cohort. Only re-evaluation candidates kept past Q3.

Thresholds 90 / 70 / 50 are initial design values · adjusted quarterly using Batch 1 response rates and reply-cohort accumulation.

ENGAGEMENT-BASED SCORING · PHASE 2 DESIGN

After the reply, the score moves with behavior.

The 120-pt ICP is a static pre-send score. Real deals are decided by what happens after the send. That's why Engagement-Based Score (EBS) sits as a separate layer — 4 weeks of webhook logs accumulate during Batch 1, then Phase 2 goes live.

DEFINITION
EBS = a dynamic score assigned in real time to each lead's actual behavior
Each signal — open · click · reply · sample request — carries a fixed weight (−100 to +200). Instantly webhook + A5 classifier compute and accumulate them live.
FORMULA
Composite = ICP + EBS
ICP is the pre-send static score from account data (0–120). EBS is the post-send dynamic score from behavior (−100 to +200). The sum drives automatic Tier rebalancing.
SELF-TUNING
Weights are retuned quarterly by the Growth Loop
If "sample request +40" converts to meetings unusually well, next quarter it's bumped to +50. The signal → score mapping itself learns from data.
WORKED EXAMPLE · Design scenario
Barcelona furniture retailer Co. M (illustrative) ICP 95 · TIER C
DAY 0 → DAY 9 · 4-touch sequence · PHASE 2
DAY 0
Pre-send
static snapshot
  • ICP scored95
ICP95
EBS0
COMPOSITE95
TIER C
DAY 2
Touch 1 · Angle A sent
webhook: opened
  • Email opened+2
ICP95
EBS+2
COMPOSITE97
TIER C
DAY 5
Touch 2 · Angle B sent
webhook: link clicked
  • 3+ opens+5
  • Link clicked+10
ICP95
EBS+17
COMPOSITE112
TIER C
DAY 9
Touch 3 reply · sample request
classifier: positive + sample
  • Reply positive+30
  • SKU mention+25
  • Sample request+40
ICP95
EBS+112
COMPOSITE207
TIER A → HOT
DAY 9 +30m
HOT LEAD triggered
Email + HubSpot Task · meeting brief
  • sample triggerALERT
  • Shin steps inMANUAL
ICP95
EBS+112
COMPOSITE207
HOT LEAD
COMPOSITE SCORE BAND
< 70 · DORMANT
70–119 · TIER C
120–149 · TIER B
150–179 · TIER A
180+ · HOT
0
70
120
150
180+

Signal catalog

This quarter we track 12 positive + 7 negative signals. Each weight is retuned quarterly by the Growth Loop against actuals.

POSITIVE · score up12 signals
Email opened (1×)
+2
Email opened (3+×)
+5
Link click landing / calendar
+10
Reply (any content)
+15
Reply · positive class A5 classifier
+30
Specific product / SKU mention
+25
Catalog / asset request
+35
Sample request → AE alert
+40
Pricing / terms question → AE alert
+45
Decision-maker intro
+50
Meeting proposal → AE alert
+60
Re-meeting / reference shared
+70
NEGATIVE · score down7 signals
Unopened after 14 days
−5
No response after 3 touches → Nurture
−10
30+ days out of office
−10
"wrong person" reply → retarget
−20
"not interested" reply → Nurture
−30
Unsubscribe drop
−100
Hard bounce drop
−100

Composite → automatic Tier rebalance

No SDR intervention — Tier reassigns the moment thresholds are crossed. HubSpot stage syncs at the same instant.

COMPOSITETIERTREATMENT
≥ 180HOT LEADShin steps in immediately · AE alert · meeting brief auto-generated
150 – 179TIER A100% manual review · priority reply handling
120 – 149TIER B3-angle A/B/C auto-sequence · reassigned by performance
70 – 119TIER CNurture DB · 6-month re-evaluation cycle
< 70DORMANTLong-term archive · revisit after 12+ months
ACQUISITION · DETAIL

One lead in 9 minutes · 7-stage pipeline

From finding a lead to sending a tailored email: 6 minutes automated + 3 minutes Shin reviews.
Machines do the research and drafting; Shin only judges whether the email lands in the field.

1
Lead discovery
Apollo search for accounts & contacts against the ICP
2
ICP scoring
6-axis 120-pt scoring · A/B/C grade
3
Claude signals
Web crawl + LLM analysis · detects experience · certs · launches · hires
4
Background
Site · LinkedIn · news · blogs structured into a DB
5
Pain extraction
Pulls one concrete pain sentence from background
6
Solution match
Picks SKU/offer aligned to that pain automatically
7
Tailored email
PAS 100 words · 3 subject lines tested
One real lead · anonymized

Spain Co. A · Aesthetic distributor · Madrid · 5–20 people

Mid-size clinic channel · no PDRN/peptide category

01 Discovery
48 Apollo candidates → 1 passed first ICP filter
02 Scoring
92 / 120 → Tier A (see previous section)
03 Signals
K-beauty experience · CPNP ready · recent Korean skincare launch · headcount +2
04 Background
Competitor B launched a new Korean PDRN line this month · A has not responded yet
05 Pain
PDRN portfolio gap + lack of direct Korean-manufacturer relationships
06 Solution
PDRN X-series + MOQ 0 + EU CE/CPNP done + regional exclusivity available
07 Email
Subject lines (3 A/B/C generated):
A. "Co. A's PDRN category · direct Korean-manufacturer offer (5 min)"
B. "Madrid aesthetic · PDRN supply options to share"
C. "Co. A's PDRN portfolio direction · 5 min question"
6 min automated · Apollo + Claude + Gemini
3 min Shin review · subject pick + fact check
9 min total / lead
Email Campaign · 9-Cell Test Matrix · Design example

3-Angle × 3-Touch
9 hypotheses tested in parallel over 2 weeks

A Tier B cohort is split into 3 groups, each pitched at a different angle, and each group gets 3 touches over 14 days. Nine variants run in parallel per cohort — the design intent is to surface, in two weeks of data, which pain narrative actually lands on this segment. Subject lines and cohort structure below are design examples · actual reply rates measured post-Batch 1.

ANGLE ↓ TOUCH →
DAY 0
Pain · name the gap
PAS 100 words · low-commitment CTA
DAY 4
Case · prove with numbers
Same-segment case + 15-min call ask
DAY 10
Break-up · Yes/No
≤ 50 words · binary choice
3-TOUCH REPLY %
ANGLE A
Outcome-led
Money on the table comes first
"Co. A's PDRN category · direct Korean-manufacturer offer (5 min)"
OPEN 47%REPLY 4.1%
WINNER
"Re: ... same segment · 1-page case on 35% margin lift"
OPEN 38%REPLY 9.2%
"If the PDRN gap is still open, next quarter's timing might not be us"
OPEN 34%REPLY 13.8%
27.1%
3-touch cumulative
▲ WINNER · 1.6–1.8×
ANGLE B
Efficiency-led
Lead time / inventory pain comes first
"Spain distributor avg lead time 45 days → 18 days direct"
OPEN 41%REPLY 2.7%
"Re: ... Co. B cut MOQ 50% and improved turnover 2.1×"
OPEN 29%REPLY 4.8%
"If you just want the 1-page summary then nothing more, reply 'N'"
OPEN 31%REPLY 7.4%
14.9%
3-touch cumulative
vs A · −45%
ANGLE C
Risk-led
Competitor moves / first-mover risk first
"Competitor B launched a PDRN line this month · Spanish distribution snapshot"
OPEN 44%REPLY 3.5%
"Re: ... 2 first-movers in · position dilutes if not decided by September"
OPEN 33%REPLY 5.6%
"Competitor-moves brief: 'Y' if useful, otherwise we stop sending"
OPEN 30%REPLY 8.1%
17.2%
3-touch cumulative
vs A · −37%
What Each Angle Tests · Hypothesis & Hook
ANGLE A · OUTCOME
Does money move first?

Use revenue / margin numbers as bait. The strongest axis for aesthetic distributors — inventory turnover and portfolio gaps hit P&L immediately.

CORE HOOK
35% margin lift in the same segment
ANGLE B · EFFICIENCY
Is operations the pain?

Approach via lead time · MOQ · turnover. For targets already running the category — the narrative needed isn't "why change" but "switching is easier".

CORE HOOK
Lead time 45 → 18 days, MOQ halved
ANGLE C · RISK
What if a competitor moves first?

Use competitor wins / pricing shifts as trigger. Stronger in narrower markets — effective in categories where missing first-mover means losing a 2-year window.

CORE HOOK
Competitor B launched a PDRN line this month
Cohort Results · Tier B · 3 Cohorts × monthly
01 · CELLS TESTED
9
Parallel variants
3 angles × 3 touches · 14 days per cohort
02 · WINNING ANGLE
TBD
Selected by Batch 1 results
First-cohort reply distribution picks the winning angle · next month's resources concentrated there
03 · DECISION CYCLE
2wk
Per-cohort learning cycle
Angle ranking after every 2-week cohort · monthly resource reallocation
04 · STRUCTURAL LIFT
3×
Designed lift vs single-touch
Industry multi-touch sequence benchmarks · replaced with actuals post-Batch 1
NURTURING · COMPOUNDING ASSET · DESIGN ASSUMPTION

If you don't drop them, time turns into an asset.

Leads tagged Tier C are kept in the Nurture DB, not deleted. Monthly automated signal monitoring catches account and market changes, and once thresholds are crossed they're auto-promoted back to Tier B/A into the active pipeline. Numbers below are pipeline design assumptions · adjusted in Phase 2 once actuals accrue.

NURTURE POOL · M12 (design)
19,200leads
1,600/month × 12 months cumulative (theoretical)
Re-promotion rate (target)
5%
Quarterly re-evaluation · Phase 2 KPI
REVIVED LEADS · M12 (target)
960leads
19,200 × 5% · 12-month re-engagement target
vs COLD RESPONSE
3-5×
Industry research basis · replaced with actuals in Phase 2
FLYWHEEL

Nurture isn't a one-time stop — it's a loop that keeps spinning.

Entry → watch → re-evaluate → promote, then a new cohort enters again. The DB doesn't get thrown away — it gets sharper every month.

PIPELINE CYCLE
Leads cycle between the two pipelines

Leads that fall out of active drop into Nurture and stay watched · once signals accumulate, they return to active.

ACTIVE PIPELINE Tier A · B
Outbound running · EBS 120+ · email sends
① DEMOTION · falls down

EBS drops below 120 · 90-day no-reply · account status changes.

Not deleted — moved into the Nurture pool
② PROMOTION · climbs back

Signal detected → EBS ≥ 60 · captured signals fuel email personalization.

3–5× reply vs cold
NURTURE POOL Tier C
EBS 70–119 · watched · re-evaluation cohort · 19,200 cumulative at M12
01
Watch

Claude agent scans signals monthly. LinkedIn · news · website · certifications DB.

· monthly
02
Re-evaluate

EBS recomputed against detected signals. Tracks score change and threshold crossings.

· quarterly
03
Promote

Returns to active when EBS ≥ 60. Captured signals fuel email personalization.

· when conditions met
12 months in, the Nurture pool builds up like this
1,600 leads enter every month — none discarded, all watched. The thicker the pool, the sharper the promotion accuracy.
1.6k
M1
3.2k
M2
4.8k
M3
6.4k
M4
8k
M5
9.6k
M6
11.2k
M7
12.8k
M8
14.4k
M9
16k
M10
17.6k
M11
19.2k
M12
INPUT1,600 leads / month
M12 POOL19,200 leads
12M cumulative promotions~835 returned
CASE · NURTURE → REVIVAL
Milan spa chain Co. L ICP 50 · TIER C
M0 → M6 · 6-month promotion case
M0 · entry
Saved to Nurture DB
ICP 50 · 3 staff
  • Initial snapshot50
ICP50
EBS0
COMPOSITE50
TIER C
M2 · signal
Headcount growth detected
LinkedIn scan · +3 hires
  • Size +3 (3→6)+15
ICP65
EBS0
COMPOSITE65
TIER C
M4 · signal
Competitor launch confirmed
News · Milan lineup
  • Competitor pressure+9
ICP74
EBS0
COMPOSITE74
TIER C
M5 · re-eval
Quarterly re-scoring
nurture agent · re-rank
  • CPNP cert detected+18
  • Headcount confirmed+8
ICP100
EBS0
COMPOSITE100
TIER C
M6 · promote
Returns to Tier B · re-engaged
3-touch auto sequence
  • Subject: competitor launchsend
  • Instant open + click+12
ICP100
EBS+12
COMPOSITE112
TIER C → B

Promotion triggers · which signals fire

5 signal categories tracked automatically through the nurture window. Cross any category's threshold → re-scoring → promotion candidate queue.

Size change

+3 staff or +30% revenue. Monitors LinkedIn headcount · website team page.

+15 ~ +25
Competitor moves

Same-region competitor wins or churn. Channel share shift → market pressure signal.

+9 ~ +18
Certification / regulatory

CE · CPNP · FDA and other required certs newly acquired. Clear signal of EU cosmetics market readiness.

+18 ~ +30
Product change

New category launch · SKU expansion or pruning. Catalog reshuffle triggers re-evaluation.

+12 ~ +20
Hiring trends

New sales · purchasing · marketing role postings. Org expansion = more buying decisions.

+8 ~ +15
Human × AI · Division of Labor

Humans for judgment, AI for repetition.
17 tasks split into 3 swim lanes.

This system isn't AI that runs without humans. It's designed so that humans focus only on the highest-leverage decisions. AI owns research · scoring · drafting · sending · classification. Humans spend their time on strategy · relationships · negotiation.

Swim-lanes · who does what
AI ONLY
10tasks
18 Claude Agents
Repetitive · high-volume · rules-based. No human intervention — fully automated.
01
Apollo lead search
Auto search by ICP queries · hundreds–thousands per day
02
ICP 120-pt scoring
Apollo + Claude web research, 6-axis auto · 10 sec / lead
03
Signal detection
Web crawl + LLM analysis · 4 signals (experience · cert · launch · hiring)
04
Background research · DB build
Public site · LinkedIn · news crawl + structured storage
05
Pain extraction · SKU match
From the Research DB · 1-sentence pain + auto-selected SKU per account
06
Send scheduling
Auto distribution across 4 domains · time slots · 3-touch sequences
07
Reply classification
positive / nurture / out · GPT tagging → automatic HubSpot stage
08
Nurture re-evaluation
Monthly auto re-scoring · promoted leads enter the new cohort
09
Weekly performance report
G1 agent · cohort · angle · CPM · auto-published Friday
10
AE meeting brief generation
Auto summary of context + pain + competitor · delivered via Email/HubSpot Task
AI DRAFT · HUMAN APPROVE
3tasks
AI × Shin
AI drafts → human checks facts · tunes tone · gives final sign-off.
11
Tailored emails (3 angles)
AI drafts 6 min + Shin reviews 3 min · subject · facts · tone
12
Meeting-prep review
AI summary → Shin reviews 10 min · context calibration
13
Quote & contract draft
AI draft → Shin final review · assesses room to negotiate
HUMAN ONLY
4tasks
Shin
Strategy · relationships · real-time judgment. AI can't substitute.
14
ICP weight · threshold setting
One-time setup + quarterly review · the strategic starting point
15
Positive reply response
Shin replies manually within 24h · the core of relationship-building
16
MEDDIC qualification · meetings
Direct verification in call/meeting · real-time objection & negotiation handling
17
Strategy direction · hypothesis setting
After the weekly report · decides next week's hypothesis
AE Handoff · auto-generated package

When a positive lead is handed to AE, the system auto-generates a 4-page brief.

The AE walks in 30 minutes before the meeting fully briefed. Deal close accelerates from 2–3 meetings → 1–2 meetings on average.

01 · MEETING BRIEF

1-page account context

  • ICP score 6-axis breakdown
  • Company overview · size · main products
  • Last 6 months of events (Claude web research)
  • Competitor moves · market position
  • Estimated current supplier (public info)
  • Contact's LinkedIn profile summary
02 · DEAL CONTEXT

Conversation history so far

  • 3 emails sent (subject + body summary)
  • Reply text + A5 classification result
  • Pain · urgency · objection mentioned
  • Probable decision-maker · champion
  • Proposed solution (A2 recommendation)
  • Expected deal size · close timeline
03 · PREDICTED OBJECTIONS

Expected objections · response scripts

  • The 5 most common objections in this segment
  • Sandler "negative reverse" script per objection
  • Evidence pack for competitor pricing comparisons
  • 2–3 reference customers (same region · size)
04 · AE ADVANTAGE

Why the meeting goes better

  • Zero 30-min prep — already ready
  • Pain already understood — no company intro needed
  • Walks in knowing competitor moves — instant positioning
  • Objections predicted — scripts in hand
  • Deal close accelerates — 2–3 → 1–2 meetings
Manual vs System · what the automation buys you
METRIC
Manual (1 person)
Automated system
Weekly lead discovery
20–30 leads ceiling
Hundredsdesigned capacity 2,400/month
Personalized email writing
40+ min / lead
9 min / lead− 77% TIME
Reply rate
1–2%
Designed target · 3–7%PHASE 2 ACTUALS PENDING
Meetings booked / month
2–4
Designed target · 11PHASE 2 ACTUALS PENDING
Nurture management
Effectively impossible (spreadsheet)
Auto re-eval · promotion
Weekly review
2–3h manual
G1 auto report · 15-min review
AE handoff
30 min manual prep
Auto-generated · 0 min
PRODUCT STACK

9 tools · 18 agents · one engine

Industry-standard SaaS, combined without reinvention. 18 Claude Code agents act as the orchestrator.
Shin's approach: a thin layer that intelligently wires existing tool APIs together.

① Lead discovery & verification

Apollo

275M B2B contact DB · matching account lists discovered through ICP queries

NeverBounce

Email validity check · minimizes bounces, protects domain reputation

② AI · decision brain

Claude API

Opus/Sonnet/Haiku auto-routed by task · 80% LLM cost reduction

Gemini API

Auxiliary model for research and structuring · multimodal input handling

HubSpot

Two-way CRM sync · outbound_sequence · message_angle custom fields

③ Sending & meetings

Instantly

4 warmed domains · 2,400 emails/month capacity · A/B/C 3-angle parallel testing

Google Workspace

Gmail integration · sent-folder sync · automatic reply-thread tracking

Calendly

Auto meeting booking · two-way calendar sync · link delivered the moment a positive reply lands

④ Infrastructure

Google Cloud Platform

8 Cloud Functions + 5 Schedulers · runs 24/7 without interruption · auto-scales

BUILD STATUS · WHAT WAS SHIPPED

2-week solo build · Batch 1 deployed

Market KPIs only read meaningfully after Phase 2 accrues, so
the most honest evidence I can show right now is the system that's been built.
Below: codebase & operating-tool actuals as of 2026-04.

BOTTOM LINE

Solo-built in 2 weeks: 18 AI agents · 41 CLI commands · 6,500 lines of code · Batch 1 with 77 leads deployed · sequence running. KPI actuals will be shared after Phase 2 accrues.

6,500lines
Production Code
Python · GCP Functions · prompt management
41cmds
CLI commands
Sequence · ICP · warm-up · reporting
18agents
AI agents
Acquisition / Commit / Expansion
77leads
Batch 1 deployed
NL 60 · SE 12 · HU 5 · sequence running
WHAT'S RUNNING NOW
  • 5 Cron Schedulers · daily auto ICP scoring · warm-up · sequence triggers
  • 8 Cloud Functions · Apollo / NeverBounce / HubSpot / Instantly webhook pipelines
  • Batch 1 sequence · 77 NL/SE/HU leads live · D+0/D+3/D+10/D+17 auto touches
  • HubSpot Pipeline · positive replies auto-progressed · Calendly link auto-delivered

Note: open / reply / meeting figures will be shared formally as a Notion report once the Phase 2 sequence completes. Evidence of the implemented system can be shared via Instantly / HubSpot / GitHub dashboards on request.

COST TRANSPARENCY

What it actually costs · nothing hidden

Sales automation is often pitched with aggressive numbers.
Here, monthly subscription & usage costs are fully disclosed.
Split into two buckets: fixed subscription + usage-based. Monthly send volume of 2,400 and tier distribution below are design assumptions · Shin's hourly rate is a simulation basis.

① FIXED SUBSCRIPTIONS
Tool USD/mo
Apollo (Basic · 10K credits included)$99
HubSpot (Starter)$50
Instantly (Hyper-growth, 4 domains)$97
Google Workspace (4 accounts)$50
Calendly (Professional)$12
GCP base (always-on minimum)$30
Fixed subscription total$338
② USAGE-BASED
Item USD/mo
Claude API · Haiku-led, Sonnet only on hard cases$35
Gemini API · multimodal assist$10
NeverBounce (2,400 verify)$12
Usage total$57
Model-mix strategy · 90% of ICP scoring, signal detection, and email drafts run on Haiku. Only complex context and objection-response drafts escalate to Sonnet. Average API cost ~$0.04 per lead.
MONTHLY TOTAL
$338 (fixed) + $57 (usage)
= $395 / month
Per email: $395 / 2,400 = $0.16
Per unique lead (3-touch): $395 / 800 = $0.49

How monthly send volume is computed

Per-domain daily
30emails
reputation safe band
×
Domains
4
warmed
×
Business days / mo
20
weekends excluded
= MONTHLY CAPACITY
2,400 emails +@
SCALE UP A · 2× volume
8 domains → 4,800 emails/mo
$595/mo (+$200) · $0.12 per email (↓ 25%)
SCALE UP B · 3× volume
12 domains → 7,200 emails/mo
$795/mo (+$400) · $0.11 per email (↓ 31%)
Why scale gets cheaper · (1) Fixed subscriptions mostly stay flat (only Apollo/HubSpot/Workspace partial plan upgrades). (2) Higher API volume → more Haiku, more cached research reuse. (3) Average tokens-per-lead drops. The email generation itself compounds.

Cost Per Lead by Tier

Based on Shin's time at $60/hr · Tier A includes manual review and MEDDIC time, hence the highest cost.

Tier Leads / mo Shin time Time cost CPL (tools) CPL (all-in)
TIER A · 70+ 120 20hr $1,200 $0.49 $10.49
TIER B · 50–69 680 34hr $2,040 $0.49 $3.49
TIER C · 30–49 1,600 0hr $0 $0.25 $0.25

Cost Per Meeting (design basis)

Based on monthly investment structure and target meeting plan · actuals will be re-shared after Phase 2 accrues.

CPM · TOOLS ONLY · DESIGN ASSUMPTION
~$36
$395 (monthly tools) ÷ 11 meetings (design target) ≈ $36 / meeting

Numbers are based on the design-stage funnel · actual CPM is recalibrated after the Phase 2 sequence accrues · varies with the decision cycle.

BREAK-EVEN · DESIGN BASIS

Monthly tools cost $395. Assuming a $5K average deal size, closing just one deal per month hits break-even.

ROI numbers swing significantly with positive→meeting→close conversion rates, so treat these as design assumptions only.

ABOUT SHIN

An operator built in the field, for the field

I didn't learn sales from a book. I sold roasted beans to cafe owners face-to-face, ran a wine bar watching consumer behavior up close, and today I lead global B2B sales for medical aesthetics — feeling the pain firsthand. This system is the distillation of that experience.

4½ years of F&B operating taught me the real feel of cold outreach in the field; global B2B sales taught me how to negotiate directly with mid-size distributors and individual owners; and operating gave me the D2C-brand pain only operators can see — these three are the design foundation. The mid-market D2C accounts I'm targeting are exactly the kind of customer I ran for 4½ years. Enterprise has internal dev resources; mid-market owners decide and operate themselves — I've run that decision structure from both sides.

2021 – 2022

Roastery · B2B bean supply

Cold visits to indie cafes · supply operations
  • 30 cafes on average, 40 at peak as active accounts
  • Monthly supply KRW 23–34M (annual KRW 280–400M)
  • Longest customer retention 18 months
  • Sample tasting · seasonal price negotiation · inventory turnover

What I learned: ran the full B2B sales cycle by hand — cold visit → sample → repeat purchase. Watching cafe owners pick beans, I built a 3-axis pitch myself: taste · price · inventory turnover.

2023 – 2024

Wine bar · solo D2C operation

Hapjeong-dong · solo operator · space business
  • Monthly revenue KRW 10–15M (annual KRW 130M)
  • 40–50 wine SKUs managed · price · inventory · seasonality
  • Target: wine beginners + young demo · revisit rate 10–20%
  • Menu planning · ambience · service flow — designed solo

What I learned: in a low-revisit market, I personally made the call to reposition the wine bar as a space business. Felt the post-first-purchase retention pain that mid-market D2C brands face — from the operator's seat.

2025.10 – present

iCELmedi · Head of Global Sales

Global medical aesthetics · K-beauty distribution
  • EU customs partnership · built 10-country, 50 validated buyers
  • Pro Direct Batch 1 launch: NL 60 · SE 12 · HU 5
  • B2B GTM automation system, 2-week solo build (2026.04)
  • Week 1 sequence running · Batch 1 77 leads deployed (KPI actuals after Phase 2 accrues)

Now: running the system daily in the field, accumulating learning. Started from real pain — "manually finding leads and writing emails every day makes no sense" — so it's not on paper, it's field-tested.

FROM SHIN

There's only one reason I think I'm a fit for SDR · Outbound · Inside Sales roles. For 4½ years I ran the cold-visit cycle as a D2C operator, and now I'm running the same cycle from the other side as a global B2B sales lead.

I know why mid-market D2C and small distributors stay up at night, and I have the gut feel for which messages get a reply. Lay this system on top of that feel, and transplanting it to your context takes 1–2 weeks. No rebuild needed — just swap the ICP filters, hooks, and sequence content.

SYSTEM TRANSPLANT

Transplanted into a B2B/D2C outbound system · 1–2 weeks

Built in iCELmedi's B2B global medical-aesthetics context, this system can be transplanted —
no rebuild, just 4 swaps — into other industries · ICPs · GTM contexts (cosmetics export · Korean SaaS · mid-market D2C, etc.).
Wing 02 Commit and Wing 03 Expansion stay reusable at the framework level.

STEP 1

Swap ICP filter

EU aesthetic distributor 5–20 people → redefined as ICP signals & filters tuned to your target market

STEP 2

Swap message hook

margin / regulatory / competitive → replaced with your ICP's core pain & decision triggers

STEP 3

Swap sequence content

iCELmedi product pitch → replaced with your product/service value + differentiation references

STEP 4

Swap research DB sources

aesthetic industry news → replaced with your industry's signal sources · channel data · buy-intent markers

PROPOSED NEXT STEP

Keep the architecture, swap the 4 layers together against your internal data.

Through 4½ years of running my own cafe · roastery · wine bar plus global B2B medical aesthetics sales, I've personally lived through the decision context of both D2C operators and B2B buyers. Concrete numerical targets should be set together against your GTM plan.

CONTACT

Would love to jump on a call.

Happy to dig into role fit in detail. Week 1 Learnings (Notion) and full résumé PDF available on request.

EMAIL
Gmail
t68632899@gmail.com
PORTFOLIO
Portfolio deck
PDF · 22 slides
PROFILE
LinkedIn
mk-shin
Shin · Outbound Operator · 2026