Investor Brief · May 2026
Prepared for Dr. Bhupesh Manoharan

You asked if Aditya was random. He wasn't.

A 12-stage agent pipeline, written into a git-synced brain, captured one VC's voice in a 2-hour call and grew his monthly LinkedIn impressions from 200K to 4M in five months. This document is the why before the ask.

200K
Aditya · Day 0
4M
Aditya · Month 5
5
Months
20×
Verified growth
Dr. Bhupesh Manoharan
Dr. Bhupesh Manoharan
Dean · Associate Professor of Marketing · Masters' Union
Built for one investor
Scroll
Was Aditya random? The system. Not the operator. A 12-stage pipeline. A brain that doesn't forget. Was Aditya random? The system. Not the operator. A 12-stage pipeline. A brain that doesn't forget.
§ 02 · The question on the table

The right falsifiable question.

You asked it directly in our last meeting. If Aditya grew 20× in five months, was that the Snowball system, or was that Aditya? It is the only question worth asking at this stage. We owe you an answer that holds under cross-examination.

"
Was the growth from the system, or was it random? — Dr. Bhupesh Manoharan, Apr 2026
20×

If it were the operator, Aditya could fire us tomorrow and keep the curve. He has not. He won't.

The brain is in our repo. The engine runs on our pipeline. The voice was captured by our system. The next ten pages show why.

§ 03 · Why the agency model broke

$2,000 to a human. $200 of it writes.

A founder pays a traditional LinkedIn agency $2,000 a month. The agency keeps $1,800. A ghostwriter in another timezone gets $200 and is asked to think like a VC partner. When the ghostwriter leaves, the voice leaves with them. This is the whole industry. Almost nobody says it out loud.

Legacy Agency
$2,000 to a human. $200 writes.
VC signs for $2,000 / month
Account manager takes the lead
Outsources to a ghostwriter ($200)
Ghostwriter can't think like a VC partner
Ghostwriter churns. Context evaporates.
Client fires agency. Cycle repeats.
90% of the money is rent on a broken system. The only asset that compounds is the client's frustration.
vs
Snowball
$3,500 builds a compounding system.
VC signs for $3,500 / month
Manager runs a 2-hour voice capture
Voice, tone, ICP written to client's memory
Agent generates 4 posts per week on-voice
Every post and metric deepens the memory
If the manager leaves, the system stays.
Every dollar compounds into an asset the client cannot find anywhere else. Their own voice at industrial scale.
§ 04 · The brain

One repository. Seven operators. Zero memory loss.

Snowball runs on a single git-synced filesystem. Every decision, voice transcript, content draft, and analytics reading lives in one place. Read by humans, written by AI agents, immune to staff turnover. This is the infrastructure layer most agencies don't have.

01 · Business
Business Context
business/
  • Pipeline of every deal
  • Strategy + offer documents
  • Finance and runway tracking
  • Team responsibilities
02 · Product
Product Skills
product/skills/
  • Hook engine, humanizer, fact-check
  • 12-stage pipeline definition
  • Versioned, plaintext, agent-readable
  • One change ships to every client
03 · Memory
Client Memory
clients/{slug}/
  • Voice + tone + ICP locked at onboarding
  • DNA kit, persona brief, mission
  • The agent's ground truth for that voice
  • Never re-derived after capture
04 · Output
Output History
clients/{slug}/
  • Every draft, ever
  • Every published post, every metric
  • What hook landed, what flopped
  • The agent reads this before every run
§ 05 · The engine

Twelve stages. One agent. Every post.

Built on the Anthropic Agent SDK. Every post that ships passes through the same pipeline. Each stage is a discrete, audited, replaceable skill. The same engine that produced Aditya's 4M.

Capture
STAGE 00 — 01
Trigger + Schedule Lock
Pull latest brain. Lock slot. Manage retry queue.
STAGE 02
Pillar Selection
Compute content debt per pillar. Pick the highest.
STAGE 03
Topic + Dedup
Cluster-key match against last 20 posts. Block repeats.
Generate
STAGE 04
Content Draft
Load voice + tone + ICP. Apply ten persuasion laws.
STAGE 05
Humanizer v3
10-step anti-AI pass. Banned words, em-dash purge, rhythm.
STAGE 06
Hook Engine
10 hooks across 5 Ruben Hassid techniques. Pick winner.
STAGE 07
Quality Gate
10-check hard loop. Fail returns to Stage 5.
Ship
STAGE 08
Infographic
Gemini 3 image gen. Adversarial verifier. 3-attempt cap.
STAGE 09
Save + Ledger
Write with v4 frontmatter. Finalise dedup entry.
STAGE 10
Perfect Preview
Shareable LinkedIn preview link for client approval.
STAGE 11 — 12
Publish + Learn
Unipile posts. Analytics loop back into memory.
§ 06 · The double compounding

Why every day gets easier.

Two loops compound at once. Together, they explain why marginal cost per post drops as the system runs longer, and why we can't be matched by a competitor who starts later.

Loop 01 · Horizontal
Product upgrades ship to every client at once.
When we learn something new — a better hook technique, a fresh humanizer rule, a new persuasion law — it becomes a skill file. Push to the repo. Every client's agent picks it up on the next run. No migrations. No retraining. No tickets.
new skill → git push → all clients upgraded
Loop 02 · Vertical
Each client's memory deepens autonomously.
Every published post and analytics reading feeds back into the client's memory. Which hook converted, which ICP followed, which topic flopped. The agent reads this before writing the next post. No human manager has to remember a thing.
post live → analytics in → next post smarter
§ 07 · Proof, running today

The same engine ran Aditya Arora from 200K to 4M in five months.

CEO of Faad Capital. Investor in 130+ Indian startups. One 2-hour voice capture call. Zero hours from him after.

Live · Snowball managed
Aditya Arora
Aditya Arora
Investor in 130+ Startups · CEO, Faad Capital · Bestselling Author
164,161 followers
4M impressions / month
+9.4% engagement
Post impressions · last 5 months
Month 0 → Month 5
Impressions
200K → 4M
20× in 5 months
Followers
148K → 164K
+16K, organic from long-form
Founder DMs
12 → 91 / mo
Cap-table on LinkedIn
0
In five months · zero hours from him after kickoff
§ 08 · Unit economics

Why this scales.

Three numbers explain why Snowball is not a service company with a margin problem. It is a software company with a service skin.

ACV · Monthly Plan
$3,500/mo
per client · cancel anytime
Quarterly $3,000/mo. Semi-annual $2,500/mo. Every tier beats the legacy $2,000 agency on both price and output.
Marginal Cost · Per Post
< $2
API spend · end-to-end
Anthropic SDK tokens, Gemini image, Unipile call. The agent replaces the ghostwriter without the ghostwriter's risk.
Human Leverage
1 : 20+
manager to clients
One Tanvee-tier manager and one agent can service 20+ clients. The brain holds context. The human holds taste.
§ 09 · Why this is the moat

Infrastructure. Not culture.

Most "AI agency" pitches assume the wedge is the prompt. Ours assumes the wedge is the substrate. This is what's hard to copy.

01
New joiner onboarding
Clone the repo. Open Claude Code. Ask any question. The full operational history is loaded as context within minutes.
02
Context never walks out
When an operator leaves, nothing walks with them. The client's voice, strategy, and history stay in the brain.
03
Agent-ready substrate
Because context lives in plaintext markdown, autonomous Claude agents can read and write to the brain without us in the loop.
04
Compounds per client
Every post, every analytics review, every voice note gets committed. The brain deepens every single day in production.
§ 10 · The three people running it

Small team. Senior on every post.

Ansh Mamgain
Ansh Mamgain
Founder · Technical lead

Building products since 13. Started his first company at 15. Admitted to MIT at 17 on a 100% scholarship. Masters' Union now. Can build any tech end-to-end, and built every piece of the Snowball engine himself. Writes Aditya Arora's content personally.

VK
Vedish Kukshal
Co-founder · Distribution & sales

Built businesses since boarding school in 11th grade. Self-taught from Hormozi, Sam Ovens, Tony Robbins, Russell Brunson. Masters' Union now. Closes every Snowball deal. The reason this proposal is in front of you.

TM
Tanvee Maheshwari
Co-founder · Content mastermind

Always wanted to be an author. Interned at Schbang. The content mastermind behind Snowball. Previously ghostwrote for founders. Has taken clients past 10 million LinkedIn impressions.

§ 11 · The ask

₹10 crore valuation. Two options.

We're raising a tight angel round to harden the engine and lock in the founding-rate client base before pricing power tightens. You and Pavit get a founder-rate stake in a distribution-intelligence company.

10Cr
Pre-money valuation · May 2026
Anchored on the AceCard ₹90Cr valuation you already validated. Snowball is the cash-flow vehicle. AceCard is the equity multiplier.
Option A · Light entry
₹25L for 2.5%
Ticket
₹25,00,000
Equity
2.5%
Vehicle
SAFE
Implied stake at 10×
₹2.5Cr
Option B · Preferred
₹50L for 5%
Ticket
₹50,00,000
Equity
5%
Vehicle
SAFE
Implied stake at 10×
₹5Cr
01
Pipeline hardening
Stage 5 humanizer, Stage 8 infographic verifier, Stage 11 publish reliability — locked to production grade.
02
Two founding managers
Tanvee-tier hires. Each can service 20+ clients. Sales pipeline already supports the headcount.
03
18-month LLM runway
API budget locked through the model-cost compression cycle. Margin only improves from here.
04
Operating reserve
Six months runway as floor. Allows aggressive pricing on founding partners without survival risk.
§ 12 · Next step

Let's close, Bhupesh.

If the system answers your question, we'd like to lock the term sheet this week. Reply, ping, or pick a slot. We'll have Pavit's counsel on the call within twenty-four hours.

Book the close call
vedish@thesnowball.cc · info@thesnowball.cc · thesnowball.agency