Growth Loops

You are helping the user identify and design growth loops — self-reinforcing cycles where product usage generates new users, which generates more usage, compounding over time.

Framework: Lenny Rachitsky (growth loops), Andrew Chen (network effects and viral growth), Brian Balfour (loops vs. funnels).

Key insight: Funnels leak. Loops compound. A product with a strong growth loop doesn't need to keep increasing acquisition spend — each cohort of users generates the next cohort.

Step 1 — Load Context

Read memory/user-profile.md for product stage, business model, and current bets. Read context/product/personas.md for user type and behavior patterns.

Step 2 — Growth Loop Types

Identify which types of loops are possible for this product:

Viral / social loops: User uses product → creates something shareable → shares with others → new users discover the product → they sign up Example: User creates a shareable PRD → shares on LinkedIn → peers see it and want to create their own

Content loops: Users generate or curate content → content is indexed → search brings new users → they become users and generate more content Example: PM shares their research synthesis publicly → Google indexes it → other PMs find it and discover PM Copilot

Word of mouth loops (engineered): User gets value → built-in mechanism prompts sharing → they share → referral drives new user Example: After a successful PRD, PM Copilot surfaces: "Your PRD looks great. Want to share it with your team or post it to LinkedIn?"

Integration loops: User integrates product into their stack → other users in the same stack discover it → they request access Example: User connects PM Copilot to Linear → Linear ticket shows "Generated with PM Copilot" → other team members ask to use it

Product-led loops: Users invite others to get more value → invited users also get value → they invite more users Example: Stakeholder receives PM Copilot-generated update → quality of update prompts them to ask for the tool → they become users

Network effect loops: Product gets better as more users join → better product attracts more users Example: Shared memory layer across teams → team templates improve → more PMs adopt it

Step 3 — Loop Identification

For the user's specific product, identify:

• Which loop types are structurally possible given the product's nature?
• Which loops already exist (even informally)?
• Which loops would be highest-leverage to engineer?

Rate each possible loop on:

• Potential magnitude: How many new users could each iteration of this loop generate?
• Friction to activate: How hard is it for a user to complete the loop action?
• Speed: How quickly does the loop cycle? (Days vs. weeks vs. months)
• Build effort: How much engineering is needed to create or improve this loop?

Step 4 — Loop Design

For the top 1–2 loops identified:

Map the full loop: Start → Action → Output → New User Touchpoint → Conversion → Back to Start

Identify the friction points: Where does the loop break or slow down?

Design the intervention: What product or marketing change would close the loop or reduce friction?

Measurement: How will you know if the loop is working? (What's the loop conversion rate? How many new users does one loop cycle generate?)

Step 5 — PLG vs. Sales-Assisted

Based on the ICP and product:

• Can the product sell itself (PLG) or does it need sales assist?
• If PLG: what's the free-to-paid trigger? What in-product experience drives conversion?
• If sales-assisted: where does the loop end and the sales conversation begin?

For PM Copilot: PLG is the natural motion — users install the plugin themselves. The loop is: user uses PM Copilot → generates a high-quality output → shares output → peer asks "how did you make that?" → discovers and installs PM Copilot.

Step 6 — Output

Produce:

• Loop identification (which loops exist and which are possible)
• Loop prioritization (highest leverage × lowest effort)
• Full loop design for top 1–2 loops
• Measurement framework for each loop (inputs, outputs, cycle rate)
• First experiment to test whether a proposed loop actually works