Two teams have the same budget. One refills a funnel forever; the other makes every new customer produce the next one. A year later, the first has to keep spending to stand still, the second's growth keeps turning after the push stops. The difference isn't effort. It's that one built a funnel and the other built a loop.

The quick version

  • Funnels leak. They run one direction and have no return path, so growth stops the moment you stop pouring budget in at the top.
  • Loops compound. Each cycle's output (new users, content, revenue) is reinvested as the next cycle's input, so growth feeds itself.
  • The move: map your actual loop end-to-end, find the one rate-limiting step that caps the whole cycle, and deliberately reinvest the output back into the input.
  • The honest caveat: loops can spin down as fast as they spin up, every loop eventually saturates, and not every business has a strong one.

The idea in depth

Start with what a funnel actually is: a measuring tool, not a growth model. It tells you where people drop off between steps so you can fix the leakiest one. Useful, but it has no memory and no return path. Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen made the point in Reforge's much-cited practitioner essay "Growth Loops are the New Funnels" (2018): funnels "operate in one direction." You put more in at the top to get more out at the bottom, with nothing said about reinvesting what comes out. Their verdict is blunt, the funnel is "too micro of a view to answer 'How does your product grow?'" One caveat to keep in view: this is a practitioner framework, not peer-reviewed research.

The consequence is structural. A funnel depends on outside fuel by design, to grow more you need more money, more people, more channels, forever. Which is why the better question isn't how to optimise the top of the funnel. It's this: when this cycle ends, what did it leave behind to start the next one?

The flywheel is the idea underneath

The loop isn't new thinking dressed up. In Good to Great (2001), Jim Collins gave it its enduring image, the flywheel. Great results never arrive "in one fell swoop," he wrote; the process "resembles relentlessly pushing a giant, heavy flywheel, turn upon turn, building momentum until a point of breakthrough, and beyond." There is "no single defining action, no grand program, no one killer innovation, no solitary lucky break, no miracle moment." Collins later devoted a whole short book, Turning the Flywheel (2019), to helping organisations map their own.

The most famous flywheel of all came out of a meeting with Collins. Around 2001, in the dot-com wreckage, Jeff Bezos brought him in to advise Amazon, and afterwards the team sketched their virtuous cycle on a napkin. Brad Stone tells the story in The Everything Store (2013): lower prices brought more customer visits; more customers attracted more third-party sellers; more sellers meant broader selection; greater scale spread fixed costs thinner, which allowed still-lower prices. Feed any part of that cycle, the thinking went, and the whole thing accelerates. Notice what Amazon was not doing, buying growth at the top of a funnel. It was building a machine where each round of growth made the next round cheaper.

"Loops are closed systems where the inputs, through some process, generate more of an output that can be reinvested in the input.", Reforge, "Growth Loops are the New Funnels"

What actually makes a loop

Strip the jargon and a loop has three moving parts that bite their own tail: an input (a new user, a dollar, a piece of content), a process (something the user does that creates value), and an output that feeds straight back into the input. Run it once and you get a little growth. What makes it compound is that cycle one's output becomes cycle two's fuel. Andrew Chen's The Cold Start Problem (2021) adds the accelerant: when each new participant makes the product more valuable to everyone already in it, you have a network effect. But networks don't switch on at scale. They begin as a single self-sustaining "atomic network", Facebook at one campus, Uber in one city, that has to reach critical mass first. The implication for a network loop is awkward but freeing: don't try to spin it globally on day one. Get one small network turning under its own power, then go again.

flowchart LR
  subgraph FUNNEL["The funnel, refill forever"]
    direction TB
    A1(["Paid acquisition (top-up)"]) --> A2(["Visitors"]) --> A3(["Sign-ups"]) --> A4(["Activated users"]) --> A5(["Customers"])
    A5 -. "no return path" .-> A1
  end
  subgraph LOOP["The loop, output feeds input"]
    direction LR
    B1(["New users"]) --> B2(["Create value / content"]) --> B3(["Distribution"]) --> B1
  end
A funnel runs one way and needs topping up; a loop closes, so each cycle's output becomes the next cycle's input. Leaders Loop

Where loops break down

The flywheel metaphor seduces people into thinking it only ever speeds up. It doesn't. The same compounding runs in reverse: loops spin down as well as up. If a cycle produces fewer inputs than it consumed, momentum bleeds away, often faster than it built. Saturation is the other ceiling. Every loop eventually meets the edge of its market or channel, and one that grew effortlessly can stall hard when it gets there. And the blunt one: plenty of businesses don't have a strong loop at all. A lot of good companies grow through sales and brand, with no meaningful self-feeding cycle, and forcing the metaphor onto them just produces a flywheel that was never going to turn. So the honest first question isn't "what's our loop?" It's "do we have one, and is the arrow actually closing?"

A worked example

Make it concrete. Picture a small B2B software company whose customers are bookkeepers. Instead of buying clicks, the team ships a feature that turns each customer's ordinary work into something public and search-friendly: a shareable, indexable summary page the bookkeeper sends on to their own client. Trace the turn.

flowchart LR
  N(["New user signs up"]) --> U(["Uses product, produces a shareable page"])
  U --> I(["Page indexed by Google / shared with their client"])
  I --> D(["Searcher or recipient discovers it"])
  D --> N
A content loop: each user's normal work creates a search-indexed asset that recruits the next user. The last step arrows back to the first, that return path is the whole point. Leaders Loop

The point isn't "do content marketing." It's that the loop's output is a by-product of the core action, it costs nothing extra to make, and it keeps working after you stop pushing. Compare the two: write a hundred blog posts by hand and you've built a funnel you have to keep refilling; ship one feature that makes ten thousand customers each leave a durable trail and you've built a loop that turns itself. Here's the line worth taking into your next planning meeting. A funnel asks "how do we get more in?" A loop asks "how does each customer make the next one?"

Once the loop closes, two numbers govern it. Cycle time, how long one full turn takes, and amplification, how many new inputs each turn produces. Above one, the loop spins up; below one, it winds down. So find the slowest, leakiest turn and fix that constraint before anything else. A loop is only ever as fast as its weakest step.

flowchart LR
  S1(["New users"]) --> S2(["Activation"])
  S2 --> S3(["Create / share value"])
  S3 -->|"slowest, leakiest turn, fix here first"| S4(["Distribution"])
  S4 --> S1
  style S3 stroke:#9333ea,stroke-width:3px
Finding the rate-limiting step: improve the one constraint that caps the cycle, not every step at once. Leaders Loop

Frequently asked questions

Isn't this just a fancy funnel?

A funnel and a loop can describe the same steps, sign-up, activation, payment. The difference is the arrow at the end. A funnel's last step exits the system; a loop's last step returns to the first. That single closing arrow changes the economics: a funnel's growth is bounded by what you spend each period, while a loop's growth is a function of what it already produced. One adds; the other compounds. If you can't draw a credible arrow from your output back to your input, you have a funnel, which is fine, as long as you know it and price the top-up accordingly.

How is a loop different from a flywheel?

They're the same idea at different altitudes. "Flywheel" (Collins) is the strategic, company-level metaphor for momentum that compounds turn upon turn. "Growth loop" (the Reforge practitioners) is the operational, mechanism-level version: a specific, drawable input → action → output → input cycle you can measure and improve. Use the flywheel to align the leadership team on direction; use the loop to actually engineer the turns.

How do I find my rate-limiting step?

Draw the loop end-to-end, then look for the step with the worst conversion or the longest delay, the one turn that, if it improved, would speed up everything downstream of it. It's usually not the step you spend the most time on. Resist the urge to optimise all four steps a little; a loop only moves as fast as its slowest turn, so concentrating effort on the single constraint beats spreading it thin.

Do I still need a funnel?

Yes, as a measuring instrument, not a growth strategy. The funnel is still the best way to see where a given cohort leaks between steps, which is exactly the diagnostic you need to find the loop's rate-limiting turn. Keep the funnel on the wall as your gauge; just stop treating "pour more in at the top" as the plan.

What if my loop has amplification below one?

Then it's technically a funnel with extra steps, it consumes more than it returns, so it can't compound. That's a real, common finding, not a failure. Either fix the leakiest step to push amplification above one, or accept it's a funnel and budget the top-up honestly. The danger is believing you have a self-feeding loop when the arrow back to the input is broken.

Where to go next

And two from our own Toolkit: the companion AARRR growth-lever framework, the diagnostic that tells you which step of acquisition, activation, retention, revenue or referral is your real constraint, and our profile of product leader Rob Alford, on improving the system rather than pushing the existing one harder.

Related in the Toolkit