Skip to content
Marketing
Content
Content production pipeline
Engineering content
Learning path
Case studies
Tools
Positioning
Email daily summary
AI Engineering summit in Melbourne
SMM STRATEGY & PLAN
LinkedIn (Company + Jack + Anton) | Instagram | Threads
Data: April 2025 – May 2026 | Prepared: May 2026
megaphone
Goal: Build a systematic lead generation machine: ICP-targeted content drives Playbook downloads, downloads enter a structured outreach sequence, sequence converts to booked discovery calls.

1. Situation Analysis: What the Data Shows

This analysis covers 96 unique MadAppGang LinkedIn company page posts and 84 original Jack Rudenko personal posts, from October 2025 to April 2026. Instagram data covers 100 organic posts from April 2025 21q12qaxzto May 2026, and 11 paid ad campaigns totalling $525.94 in spend across January–May 2026.

1.1 LinkedIn: Company Page + Jack’s Personal Page (October 25- April 26)

The Performance Gap Across LinkedIn Channels
Channel
Posts Analysed
Avg Engagement
Median
Peak
96
17
16
70 (Eugene anniversary)
84
33.1
22
494 (AI code quality)
16
31.1
28
84 (agents routing repost)
⁠
Jack's personal page outperforms the company page by 1.9x on average. That gap is structural, not a fluke — median is 22 vs 16. LinkedIn's algorithm consistently rewards personal voices over company pages. Every content decision should account for this.
Jack's Post Themes: The Real Performance Breakdown
Theme
Posts
Avg Eng
Median
Max
Key Insight
Personal confession
7
125.9
58
494
Best theme by 3.8x the overall mean — critically underused
Production receipt
6
72.2
60
145
Strong — underused relative to posting frequency
AI industry takes
14
21.7
22
49
Average — no production anchor limits upside
MAG tool announcements
19
21.5
22
60
Average — value depends on hook quality
Workflow / education
34
20.9
19
76
Below mean — 40% of volume, dilutes overall avg
Off-topic
4
10.2
9
19
Worst theme — remove from rotation entirely
⁠
megaphone
The single most important finding in Jack's dataset
Personal confession posts (n=7) average 125.9 engagement — 3.8x the overall mean of 33.1.
These 7 posts account for only 8% of Jack's posting volume but contain 3 of the top 4 performers in the entire dataset.
The format: 20+ years of specific production context + a counterintuitive argument + concrete horror stories with real numbers + a question only practitioners can answer.
Jack publishes this format roughly once per month. The data says it should run every 1-2 weeks.
Production receipt posts (n=6) avg 72.2 — the second-best theme. Also underused at 7% of volume.
Combined, these two themes represent 15% of Jack's posts but would deliver roughly 60% of total engagement if weighted correctly.
Jack's Posting Cadence: Volume vs Quality
Weekly Volume
Weeks
Avg Eng / Post
What it means
84.4
Best engagement — fewer, higher-quality posts win
~27
Solid — sustainable cadence range
26.9
Highest volume, lowest return per post
⁠
Week 8 is the clearest data point: 4 posts published, average engagement 172. Two of those four posts were the 494 and 145 performers. The week with the fewest posts produced the highest average by a factor of 3. The causal story is not 'fewer posts causes better engagement' — it is that weeks where Jack had less to say on autopilot forced him to publish only when he had something worth saying.
megaphone
Cadence recommendation: quality gate over volume target
Jack is already posting 6.5x per active week.
The correct recommendation: apply a quality gate. Every post should pass one test before publishing.
Test: does this post contain a personal confession, a production receipt, or a tool announcement with specific numbers?
Posts that don't pass the test — generic AI takes, off-topic content, vague workflow education — should be held or rewritten.
Target: 4-5 posts per week, with at least 1 personal confession or production receipt post per week.
If a week has nothing worth confessing and no production receipt to share, 3 posts is better than 7 thin ones.
Jack's Format Performance
Format
Posts
Avg Eng
Median
Recommendation
66
36.4
25
Primary format — use for all personal confession and production receipt posts
17
21.2
16
Secondary — use for quick takes and short stances, not production receipts
1
14
14
Jack is not the carousel channel — that belongs to the company page
⁠
The image advantage is consistent. A personal confession post with an image gets materially more reach than the same post as plain text. The image does not need to be elaborate — a screenshot of a code horror story, a terminal output, or a photo from the office anchors the post visually and signals effort. Jack should default to image format on any post that contains a personal story or production receipt.
Jack's CTAs: What Drives Comments vs What Doesn't
CTA Type
Example
Eng
Works?
'What's the weirdest code you've seen? Drop yours, I'll share mine.'
494
Yes — reciprocal exchange, lowers barrier to respond
'What's your current setup for managing multiple agents across branches?'
145
Yes — only someone who has done this can answer
'Anyone else sitting with this?'
115
Yes — low bar to respond, validates a shared feeling
'Does your current role reward execution or judgment more?'
76
Yes — simple, answerable, not rhetorical
'Fighting, evolving, or building?'
8
No — sounds like a listicle poll
'Want early access? Let me know in the comments.'
9
No — reads as a growth hack, not a real question
'What's the legitimate reason any team still distributes access keys in 2026?'
8
No — there is no good answer, so nobody replies
⁠
Jack's Reposts: Company Page Amplification
Repost content
Eng
Source
30+ AI agents, Claude ignored them (MAG company page post)
84
Eugene Simonov 10-year anniversary (company page)
70
I made AI models judge each other (Jack's own older post)
32
Context-switching productivity killer (company page)
28
⁠
Jack's reposts of company page content perform well — the agents post at 84 and Eugene at 70 are among the top repost performers. The amplification chain is working in one direction: Jack amplifying the company page. The reverse (company page amplifying Jack) is the missing half of the distribution mechanic.
Company Page: Performance Summary 1
Format
Posts
Avg Eng
Best Example
7
20.7
AI agents routing post — 26
46
17.8
Senior Dev bug fix meme series — up to 42
43
15.5
Eugene Simonov anniversary — 70 (outlier)
⁠
Company Page: Performance Summary 2
Theme
Posts
Avg Eng
4
34.8 — highest theme, 4 posts only
13
20.2
58
16.9 — 60% of volume, pulls average down
8
11.1 — below every other theme
​
​
⁠

1.2 Instagram: Organic Performance (April 25–May 26)

Format performance
Format
Posts
Avg Eng
Median
Max
Key finding
48
14.6
9
134
Dominant format — 4x image avg, all top-10 posts are videos
5
11.8
5
39
Strong when production-anchored — only 5 posts in 13 months
47
3.6
3
15
Consistently underperforms — wrong format for IG organic reach
⁠
The format gap is decisive. Videos average 14.6 — 4x the image average of 3.6. The account splits almost evenly between videos (48) and images (47). Every static image post should be evaluated for conversion to Reels. The 5 carousels average 11.8 and are criminally underused given their performance.
Theme performance
Theme
Posts
Avg Eng
Max
Verdict
40
16.5
134
Reach engine — keep running, but these users don't convert to Playbook leads
23
6.3
39
Right audience — underperforms because it runs as static images, not carousels/Reels
4
4
10
Underperforms on IG — milestone channel is LinkedIn
~20
2.2
6
Worst type — cut entirely
⁠
megaphone
The Instagram format problem
The '7 open source AI tools' carousel (39 eng) outperforms every static AI/technical image in the dataset. Format is the variable, not the topic.
Converting AI/technical content from static images to carousels and infographics closes most of the engagement gap.
Humor meme Reels are the reach engine but attract general developers with no budget authority. They should be capped, not eliminated.
Quote/opinion static images average 2.2 engagement. These slots are worth more as carousels or infographics.
People milestone posts average 4 on Instagram vs 34.8 on the LinkedIn company page. Stop cross-posting milestones to Instagram.
Volume trend: peak and decline
Period
Posts
Avg Eng
Signal
61 posts
9.3
Maximum volume — humor Reels machine at full speed
33 posts
10.3
Volume fell 46%, per-post eng slightly higher — quality improving
10 posts
10.5
Shifting to AI/technical — right direction, carousel format
⁠
The 46% volume drop from peak is not a crisis. Avg engagement per post rose (9.3 to 10.5) when volume fell — quality improving as cadence becomes more intentional. Use more AI-focused and technical content (infographics/carousels).

1.3 Instagram: Paid Ad Campaigns (Jan–May 2026)

Paid Ad Campaigns
Campaign content
Spend
Impressions
Profile visits
CPV
Post organic eng
Dev truths meme Reel #1
$99.94
18,573
2,444
$0.041
9 organic
Dev truths meme Reel #2
$99.84
16,792
1,788
$0.056
16 organic
Dev truths meme Reel #3
$95.89
14,920
927
$0.103
8 organic
Dev truths meme Reel #4
$28.30
5,579
706
$0.040
9 organic
'For more' content (meme adj.)
$25.44
4,802
792
$0.032
N/A
AI tools for backend devs (image)
$55.82
10,678
214
$0.261
0 organic
7 open source AI tools (carousel)
$40.37
5,402
155
$0.261
39 organic — best tech post
Best AI tools for frontend (image)
$21.65
2,179
54
$0.401
5 organic
30+ AI agents (carousel)
$24.73
3,392
39
$0.634
8 organic
5 years ago / AI co-founder (carousel)
$16.30
2,124
34
$0.479
5 organic
Productivity killer / context-switch (image)
$17.86
1,239
28
$0.638
5 organic
⁠
Paid Ad Allocation
Budget allocation
Spend
% of total
Visits
Avg CPV
$349.41
66%
6,657
$0.052
$176.73
34%
524
$0.337
$526.14
100%
7,181
$0.073
⁠
megaphone
Paid ads: easy problem, easy fix
Problem: 66% of spend went to meme campaigns that drove general developers with no budget authority to a page with no conversion destination. This content successfully supported audience growth and follower acquisition. Now that the priority shifts to lead generation, the spend allocation needs to shift too.
The one carousel tested (7 open source tools) got the best results of any technical campaign — 155 visits at $0.26 CPV and 39 organic comments. Carousels and infographics are the format to run in paid.
Root cause: paid spend applied to organic content as-is, without optimising format or objective for paid distribution.
Fix: Run only AI/technical content in Reels, infographics or carousel format. Target CTOs, VPs Engineering, technical co-founders. Stop spending on meme campaigns entirely. When Playbook page is live: switch to lead_generation objective.
Expected CPL benchmark: $15–30 per lead for a well-targeted B2B tech audience on Instagram lead gen.

2. What Works and What Doesn't

What Works


Jack: personal confession posts — 125.9 avg (new finding from 100-post dataset)
Seven posts. Avg 125.9. The format: 20+ years of specific context, a counterintuitive claim grounded in real production work, concrete failure modes with real numbers, and a question that only practitioners can answer. The top three examples: 'I stopped arguing about AI code quality' (494), 'AI was supposed to take my job, instead I now have 10 jobs' (123), 'I'm grieving something I didn't expect to grieve' (115). These posts succeed because they are honest in a feed that is not. They are also among the most time-efficient posts to write — they require no research, only recall.
Jack: production receipts with specific numbers — 72.2 avg
Six posts. Avg 72.2. 'AI wrote 1,000 unit tests in one day. Not a single one caught a real bug.' (53). 'My team built 30+ AI agents. Claude ignored every single one.' (84). '4 LLM models running in parallel, each with its own full Claude Code session.' (60). The formula holds in every case: specific number, counterintuitive result, production context, what we actually did about it. The 145-engagement 'two types of developers' post fits this pattern — it starts with a production observation and closes with an offer of internal documentation. Twelve comments asked for the docs. That is a working lead capture mechanic in a single post.
Company page: people milestones — 34.8 avg, highest company page theme
Four posts. Avg 34.8. The Eugene Simonov 10-year anniversary post at 70 engagement is the highest single company page post in the dataset. Keira Taylor new hire at 46. The formula is simple: name the person, state the specific milestone, attach a real number (years, projects shipped, engineers mentored), make it about them not the company. These posts earn reshares from the person's own network, which extends reach to audiences that have never seen the MAG page. This is a repeatable format with a near-zero failure rate. It is currently running at roughly 1 post per month — it should run at 2.
Company page: document carousels — 20.7 avg, best format
Seven carousels in the dataset. All above the company page average of 17. The format earns saves — which is the MOFU conversion signal that leads to Playbook downloads. Carousels are currently published roughly once every 2 weeks. They should run every week without exception.

What Doesn't Work

Jack: workflow/education posts at scale — 20.9 avg across 34 posts
Thirty-four posts, 40% of Jack's output, averaging 20.9 engagement — below the account mean of 33.1. These are not bad posts. They are fine posts. The problem is volume: 34 workflow/education posts at 20.9 avg drag the overall account average down and occupy publishing slots that could be used for personal confessions or production receipts at 4-5x the engagement. The fix is not to stop writing these — it is to reduce their frequency from roughly 3/week to 1/week and replace the slots with the formats that perform better.
Company page: Golang content — 11.1 avg, below every other theme
Eight posts. All below average. Golang developers are not the audience for the Playbook, not the buyer persona for discovery calls, and not the engineers making AI adoption decisions. Stop publishing Golang-specific content on the company page. The time is better spent on one additional AI-technical carousel per month.

Instagram: quote/opinion static image posts — avg 2.2 engagement
Approximately 20 posts are quote graphics, culture statements, or opinion images ('We don't hire resource units', 'Clean Code killed more projects than bad code'). Average engagement: 2.2. These posts produce no saves, generate no comments worth reading, and attract no audience segment that converts to Playbook leads. They are not uses anymore. The publishing slot is more valuable as a Reels, infographics or carousel post.

Instagram paid: 66% of budget on meme content with wrong objective
$349 of $526 total spend went to humor meme Reels campaigns optimised for profile visits. These campaigns achieved the cheapest CPV ($0.04–0.10) but drove general developers with no budget authority to a company page with no conversion destination. The best-performing technical campaign in terms of CPV ($0.26) used the open-source tools carousel.

3. Channel Strategy

3.1 Jack Rudenko's Personal LinkedIn — Primary Revenue Channel
Dimension
Current State
Target State
6.5 posts/active week (13 active weeks)
4–5 posts/week with quality gate applied
~1/month (7 posts in 15 weeks)
1/week minimum — highest-performing theme
~1/2 weeks (6 posts in 15 weeks)
1/week minimum — second-highest theme
~3/week (34 posts total)
1/week maximum — quality gate required
79% (66 of 84 posts)
90%+ — lead with image on all posts
1 post total (14 eng)
Zero — Jack is not the carousel channel
None yet
1 BOFU post per 1-2 weeks once Playbook live
⁠
megaphone
Jack's quality gate — one test before every post
Ask: does this post contain (a) a personal confession with 20+ years of specific context, (b) a production receipt with a real number and a real failure mode, or (c) a tool announcement with specific performance data?
If yes: publish. Add an image. Close with a specific question that only practitioners can answer.
If no: rewrite until it passes, or hold it.
This eliminates the 40% of posts (workflow/education) producing 20.9 avg while occupying slots that could produce 125.9 avg.
3.2 MAG Company Page — Distribution Amplifier
Dimension
Current State
Target State
2-3x/week
leave
60% humor, 20% AI tech, 13% Golang, 7% people
30% TOFU, 50% MOFU, 20% BOFU + milestone
~1 per 2 weeks
Every week — non-negotiable
Every week. Not stable
Every week — non-negotiable
~1/month (reactive)
2/month (planned 90 days ahead)
8 posts in period
Zero — discontinue entirely
17.0 current
25+ by Q3 2026
⁠

3.3 Anton's Personal LinkedIn — CEO Channel

megaphone
Currently untapped — the most underused distribution asset in the strategy
Zero posts in the analysis period. CEO LinkedIn pages consistently outperform company pages by 5-8x on the same platform.
Anton's voice should differ from Jack's: less technical implementation detail, more strategic and decision-level framing — for the buyers, not the builders.
Target audience: C-suite buyers, investors, APAC tech leaders, scale-up founders — the people who commission the work.
Start at 1 post per week. The bar for early performance is not high when starting from zero.
Anton reposts every Jack personal confession and production receipt post within 24 hours.
Content distribution
Content type
Freq
Example topic
Funnel layer
Client outcome (anonymised)
Fintech client reduced agent cost attribution errors from blind to traced in 6 weeks
Strategic stance on AI adoption
Why the engineers who survive the next 18 months are the ones who stopped defending their stack
Company direction / milestone
Why MAG is going deep on orchestration patterns instead of broad on AI feature delivery
Repost Jack's/company best posts
Every personal confession and production receipt post from Jack within 24h
⁠
3.4 Instagram
Dimension
Current State
Target State
Peak: 20/month (Nov-Jan). Now: 10/month
12–16/month — quality over volume
48% Reels, 47% images, 5% carousels
15% Reels, 50% carousels, 35% infographic
Humor meme Reels (avg 16.5 eng)
AI/technical content (carousels, infographics, reels) (target: 12+ eng)
Quote/opinion static images (avg 2.2 eng)
Cut entirely — zero publishing slots for static opinion graphics
4 posts, avg 4.0 eng — wrong channel for this content
Stop cross-posting milestones. LinkedIn only.
Follow/save the post
Follow/save the post, but add download the guide → Playbook landing page
⁠

3.5 Threads

Threads launched on April 15, 2026. The dataset covers 7 posts across 28 days. The early numbers are better than expected for a 4-week-old account.
28 Days of Real Data (Apr 15 – May 13, 2026)
Post content
Likes
Shares
Format
Assessment
7 new open source AI tools (tools list)
30
25
Repurposed IG carousel
Best performer — tools lists work on Threads
Best AI tools for frontend developers
23
3
Repurposed IG carousel
Strong — same format, same result
Best AI tools for backend developers
21
9
Repurposed IG carousel
Consistent — tools list format confirmed
Free AI stack to run your workflow in 2026
9
2
Repurposed IG carousel
Decent — still a tools list variant
13 Hidden Claude Code tips (blog link)
2
0
Link post
Low — link posts underperform on Threads
15 Hidden Claude Code tips (blog link)
2
0
Link post
Low — same pattern
Ultimate cheat sheet / token limits
2
0
Tips list
Low — generic tips without production anchor
⁠
megaphone
Threads data: 2 key findings from 28 days
Finding 1 — Tools list format works: The 3 tools list posts avg 24.7 likes. Everything else averages 3.8. The format gap is 6.5x in 28 days of data.
Finding 2 — Link posts don't work: Both blog link posts scored 2 likes each. Threads suppresses external links in distribution — this is a known platform behaviour, not a content quality issue.
The baseline is real and usable: 14.9 avg likes across 7 posts in 28 days on a new account is a credible starting point. Tools lists at 24.7 avg and production takes at (TBD — not yet posted) give two format benchmarks to track.
Revised Threads strategy based on actual performance
Dimension
Current state (Apr 15 – May 13)
Target state
Post frequency
8 posts / 28 days (roughly 2/week)
5/week — short format, low production cost
Content type
100% repurposed Instagram carousels
50% standalone production observations, 30% tools lists, 20% hooks from Jack's LinkedIn posts
Format used
Multi-line tool lists, link posts
Short paragraphs. 2–4 lines max. One observation, one number, one question. No external links in body.
Best performing type
Tools lists (avg 24.7 likes)
Keep running tools lists weekly — confirmed format
Missing content type
Zero production observations, zero hooks from Jack's posts
Add 3/week — first line of Jack's LinkedIn post + standalone production takes
Link posts
2 posts, 2 likes each — suppress distribution
Remove from rotation. Blog traffic belongs on LinkedIn bio and IG bio, not Threads body.
⁠

4. Content Structure

4.1 Monthly Content Mix Targets
Column 1
Column 2
Column 3
35%
TOFU (Entertainment)
Humor, culture
55%
MOFU (Expert)
Cheet sheats, frameworks, articles
15%
BOFU (Sales)
Case studies, Playbook CTAs
⁠
4.2 Jack's Personal Post Mix
Column 1
Column 2
Column 3
Column 4
25%
Personal confession
Target: 1–2/week vs current ~1/month
25%
Production receipt
Target: 1–2/week vs current ~1/2wk
30%
Workflow/education
Down from 40% — quality gate applied
20%
Tool announcements
With specific performance data only
⁠
4.3 Content Pillars and Format Map
Pillar
Format
Freq
Channel
Funnel
CTA
AI workflows (Claude Code, MCP, tmux-mcp)
Jack + Company
Save / Playbook
Multi-agent orchestration (production receipts)
Jack + Company + Anton + IG
Playbook download
Personal confessions (identity + AI in production)
Jack
Comment / follow
AWS/cloud infrastructure (ECS, Meroku, cost visibility)
Company + Anton + IG
Save / discovery call
Engineering leadership (AI adoption, senior buy-in)
Jack + Anton
Save / Playbook
Open-source updates (Passflow, Claudish, Dingo)
Company + Jack + Anton + IG
Follow / GitHub
Honest AI takes (real vs hype — anchored to production)
Jack
Follow / repost
People milestones (named, specific, with numbers)
Company + individual
Follow
Humor / meme content
Company + IG
Follow
⁠

4.4 Weekly Cadence Template
Day
Channel
Format
Content type
Funnel
Personal confession OR production receipt — quality-gated
First line of Jack's post or standalone production observation
Weekly tools/stack list — confirmed best format (avg 24.7 likes)
Repurpose company page infographic
Repurpose company page infographic
Strategic stance or client outcome
Workflow/education post — only if it passes quality gate
1-line observation with specific number
Technical framework or checklist (save CTA)
Repurpose company page carousel — keep slides, shorten captions, add link-in-bio CTA
Repurpose company page carousel — keep slides, shorten captions, add link-in-bio CTA
Production receipt or tool announcement with performance data
Specific practitioner question derived from the week's posts
⁠

​5. KPIs: Before and After Playbook Launch


Phase 1 KPIs — Pre-Playbook
KPI
Channel
Period
Real Baseline
30-day Target
90-day Target
Measurement
Avg engagement
17.0
22
28
Total eng / posts published
Infographic posts
~2
4 (weekly)
4 (weekly, sustained)
Content calendar review
Carousel posts
~2
4 (weekly)
4 (weekly, sustained)
Content calendar review
People milestone posts
~1
2
2
Content calendar review
Posts carousel MOFU clicks rate
25%
30%+
45%+
LinkedIn analytics
Avg engagement
9.3
12
15+
Total eng / posts published
Infographic posts
~2
4 (weekly)
4 (weekly, sustained)
Repurpose LinkedIn infographic same day
Carousel posts
0.4
4 (weekly)
4 (weekly, sustained)
Repurpose LinkedIn carousel same day
Reels
~4
2
2
Content calendar review
Post MOFU save rate
1%
1,5%+
2.5%+
Instagram analytics
Avg engagement
11.6
15
20+
Total eng / posts published
Infographic posts
3
4 (weekly)
4 (weekly, sustained)
Repurpose Instagram infographic same day
Carousel posts
~3
4 (weekly)
4 (weekly, sustained)
Repurpose Instagram carousel same day
Production observation posts
0 (none published yet)
1-2/week
2/week
Content calendar review
Post carousel MOFU share rate
5.5
8.2
15+
Threads analytics
Posts
0
2
3
Count from Anton's page
⁠

Phase 2 KPIs — Playbook Launch
KPI
Source
30-day Target
90-day Target
Notes
Posts carousel MOFU save rate (LinkedIn)
LinkedIn analytics
30%+
45%+
Baseline: 25%. Mirrors Phase 1 carousel MOFU clicks rate target
Post MOFU save rate (Instagram)
Instagram analytics
1.5%+
2.5%+
Baseline: 1%. Mirrors Phase 1 Instagram save rate target
Post carousel MOFU share rate (Threads)
Threads analytics
8.2
15+
Baseline: 5.5. Mirrors Phase 1 Threads share rate target
Playbook landing page visits (from LinkedIn)
UTM tracking
N/A until live
500/month
utm_source=linkedin on all CTAs. Activate when Playbook page is live
Playbook form fills (leads)
Landing page
N/A until live
50/month
~10% visit-to-fill conversion rate
Instagram paid CPL (post Playbook launch)
Meta Ads Manager
N/A until live
Under $25/lead
Lead gen objective, carousel/infographic format, CTOs and VPs Engineering targeting
Valley sequence reply rate (Message 1)
Valley analytics
1-2%+
5%+
First outreach message. Track per vertical separately
Valley sequence reply rate (Message 3+)
Valley analytics
5%+
8%+
Architecture insight message. Indicator of genuine purchase intent.
Discovery calls booked from social
CRM / calendar
N/A until live
4/month

Primary revenue KPI. 30-day target activates post-Playbook launch
⁠
megaphone
KPI hierarchy — what matters most
Saves indicate MOFU intent. Form fills indicate BOFU intent. Discovery calls indicate purchase intent.
Phase 1 targets (saves, shares, engagement) are the leading indicators. Phase 2 targets (form fills, calls) are the lagging indicators.
Track the full funnel from impression to call. When conversion drops at any stage, that stage is where to fix — not the top of the funnel.

6. Lead Generation System — Valley

This system uses Valley as the primary outreach tool. All filter names, sequence steps, and export functions reference Valley.
The 11 steps below follow the exact workflow sequence: ICP definition → company search → validation → contact filtering → list cleaning → sequence → message writing.
FLAG: This workflow was adapted from guidance originally written for Amplemarket. Verify that Valley’s Current Company filter, CSV export, and sequence configuration match the steps described before running the first campaign.

6.1 Foundation Layer: Infrastructure Before Outreach + Workflow Sequence

megaphone
No foundation = no results. Fixing one piece of a broken system fixes nothing.
Good deliverability + bad copy = more 'not interested' replies.
Bad deliverability + good targeting = no replies at all.
Good deliverability + good copy + wrong target = more replies, zero leads.
All three must work before the system produces results. Build the foundation first.

11-step workflow sequence

1st step: Build a Separate ICP Per Vertical

MAG works across multiple industries. One generic ICP produces generic outreach — and generic outreach produces low reply rates.
Take MAG’s company background, what is being sold (the Playbook + orchestration engagements), the problem it solves, and examples of existing clients. Feed that into Claude and ask it to produce a structured ICP for each vertical separately.
Verticals to start with: fintech, edtech, logistics, healthtech. Add others based on where existing client proof is strongest.
Each ICP must define: target regions · company size range · relevant job titles (CTO, VP Engineering, technical co-founders) · whether the company has already invested in AI · whether it is growing over the last 6 months · revenue range or funding stage.
Data source for revenue and funding: Crunchbase — surfaces investment rounds and company stage information that helps qualify accounts before outreach begins.

2nd step: Set Up Company-Level Search in Valley

Work one vertical at a time. Start with the vertical where MAG has the clearest client proof.
Filters to configure: company size · industry (include relevant verticals from the ICP list) · region (APAC) · keywords · industry exclusions (crypto, gambling, anything adjacent to excluded verticals).
Do not mix verticals in a single search. A fintech search and a logistics search produce different ICP-matched accounts — running them together produces a mixed list that is harder to write personalised messages for.

3rd step: Use Keywords to Sharpen the Search

Ask Claude to generate include and exclude keyword lists for each ICP.
Prompt template: “Generate include/exclude keywords for finding companies that match this ICP: [paste ICP here]”
Critical: keywords that are too narrow will collapse the result set — sometimes to 20 companies or fewer. Test the search without keywords first. Add them incrementally and monitor how the result count changes. Target: 800–1,200 accounts in the initial company filter output.

4th step: Search Companies First — Not Contacts

Do not jump straight to finding individuals. Run the company filter first.
A well-configured company filter returns 800–1,200 potential accounts. Export that list as a CSV.
The export should include: company name · website · LinkedIn URL · company size · industry.
This becomes the raw account list. Every subsequent step runs against this list only.
Validating companies before finding contacts means the contact search is already scoped to accounts likely to convert. This removes the largest source of wasted outreach.

5th step: Validate Account Relevance with Claude

Create a dedicated Claude conversation. Give it the ICP criteria upfront: target profile, required characteristics, disqualifying signals.
Then feed it company websites in batches.
Batch size: 5–10 websites per message. More than that and the analysis degrades — Claude starts producing surface-level assessments and occasionally hallucinates fit signals.
Ask Claude to score each company against the ICP and flag yes/no with a one-line reason.
At 10 companies per batch, validating 1,000 raw accounts takes roughly 100 messages — under an hour with a systematic workflow.

6th step: Build the Validated Account List

From a raw list of 1,000 companies, expect Claude to validate roughly 150–250 as genuinely relevant.
Save those to a separate sheet. This is the list that matters. Every subsequent step runs against this validated list only.
Naming convention: use a consistent label per vertical and date. Example: Validated Fintech APAC — June 2026. This makes it easy to track which list each reply came from.

7th step: Find the Right Contacts at Validated Companies

Return to Valley and configure the contact filter. Set it to pull only from the companies on the validated account list.
Titles to target: CTO · VP Engineering · VP Product · Head of Engineering · Technical Co-Founder · COO (at smaller companies where the COO owns engineering decisions).
Add the validated company list to the Current Company filter field. Valley will return contacts only from those accounts.
This step separates targeted outreach from spray-and-pray.

8th step: Use the Current Company Filter

Paste the names or LinkedIn URLs of validated companies directly into Valley’s Current Company field.
This narrows the contact search to exactly the accounts already qualified in Steps 5–6.
Why this matters: without this filter, Valley returns contacts from the general industry pool, not from the specific accounts that passed ICP validation. The two lists are not the same.

9th step: Clean the Contact List

Before loading into a Valley sequence, clean the output.
Remove duplicates. Remove contacts who are clearly not decision-makers. Cap to 2–3 contacts per company — avoid saturating the same account with multiple simultaneous threads.
Why the 2–3 cap matters: reaching 5 people at the same company with the same message in the same week signals mass outreach. It also creates internal noise that makes it harder for any contact to champion a conversation.
Naming convention: save the cleaned list as a named segment. Example: Validated Fintech Leads — June 2026.

10th step: Load into a Valley Outreach Sequence

Add the cleaned contact list to a Valley sequence. No CRM required at this stage — Valley handles the sequencing natively.
Configure: number of touches · follow-up delays · message templates · LinkedIn steps.
The 5-message sequence structure from Section 6.2 maps directly into Valley’s sequence configuration: Message 1 as the initial touchpoint, Messages 2–4 as follow-up steps at Day 3 / Day 7 / Day 14, and Message 5 as the soft close at Day 21.

11th step: Write the Messages

Option A — Manual: slower, but produces higher reply rates. Recommended for Tier 1 accounts (CTOs at funded scale-ups with confirmed AI investment).
Option B — Valley AI: faster. Scales across large lists. Requires careful review before sending — generic AI messages look generic even when the targeting is correct.
The rule that matters: even a small amount of personalisation — referencing what the company is building, a recent LinkedIn post, or a signal from their job postings — moves reply rates meaningfully.
For accounts where Jack has already engaged on LinkedIn (comments, saves, post reactions), write those messages manually. The warm signal is already there — a generic opener loses it.

6.2 ICP Definition: Primary and Secondary Targets

Relevancy-first outreach is the only outreach that works in 2026. Without a matched pain-to-value-proposition, even perfect deliverability produces 'not interested' at scale. The ICP for the Playbook is specific and the buyer signals are observable.
MAG Playbook ICP: primary and secondary targets
ICP tier
Role
Company profile
Pain signal
Intent signal
Tier 1 (highest intent)
CTO / VP Engineering
50–500 person tech company, APAC or US, 1+ year of AI investment
Has shipped 1 agent, now struggling to get 5+ to work together. No observability. No coordination layer.
Posts about AI agents, follows AI engineering content, comments on orchestration problems
Tier 2 (high intent)
Technical co-founder
Series A–C scale-up, engineering-led product
Evaluating 3+ orchestration frameworks, none fully solve the problem. Senior engineers not trusting agents.
Job postings for AI engineers, AI infrastructure spending signals, GitHub activity on agent repos
Tier 3 (warm)
Senior engineer / tech lead
Same profile — building agent systems day to day
Context persistence failing. Memory not persisting across sessions. Cost attribution missing.
Engages with Jack's posts on LinkedIn, saves technical carousels, asks questions in comments
⁠

6.3 The Outreach Sequence: LinkedIn DM

The sequence runs inside Valley. Five messages: initial + four follow-ups. Each message has one job. The sequence is not a drip — it is a structured conversation that moves a prospect through a decision. Message length decreases as the sequence progresses.

LinkedIn DM Sequence


Initial message — Specific reason + specific pain + Playbook delivery
Opening: name the specific reason you are reaching out. Reference the signal — the post they engaged with, the job posting you found, the company blog that mentions Claude Code.
Structure: why you reached out (1 sentence, specific) → the problem (name the exact pain — not ‘AI challenges’, but ‘no visibility into which agent spent $400 in tokens last week’) → short value prop (1 sentence) → Playbook link.
CTA: soft. ‘Here’s the Playbook — Section 3 covers the coordination problem directly.’
Length: under 100 words. No paragraphs. No preamble.

Follow-up 1 (Day 3) — One specific client outcome
Opening: one-line reminder. ‘Sent you the Playbook a few days back.’
Body: one specific client outcome with real numbers. ‘A fintech client had no traces, no cost attribution, no idea which agent spent $400 in tokens. We built the observability layer in 3 weeks. Attribution errors dropped to zero.’
Want to print your doc?
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.