Brian Balfour on Growth Loops, Channel Fit and Building a Repeatable Growth System

Harry Stebbings asks Brian Balfour how he entered growth. Balfour traces his experience to the early Facebook platform and social gaming, including his company Viximo. Games combined viral and paid channels, quantitative analysis, user psychology and intense competition, making them a formative environment for the discipline.

As a founder, Balfour had to solve problems for which there was no specialist. He enjoyed finding opportunities others had missed and combining that search with his quantitative skills. He distinguishes how a product acquires users from how a company makes money: the two systems interact but are not identical.

Looking back, he says the menu of growth mechanisms is more constrained than it first appears. Through Reforge he has seen many companies and believes teams can innovate within a relatively small set of mechanisms, while genuinely new channels are rare. Knowing the menu helps teams choose realistic opportunities.

His second lesson is conviction and patience. Growth loops, sometimes called flywheels, can compound like interest: early progress looks small even when the system is improving. Founders need to look for signs that the inputs are strengthening instead of abandoning a promising strategy solely because the output has not yet accelerated.

Stebbings asks how to tell patience from persistence with something that will never work. Balfour says there is no simple threshold, but many teams look only at traffic or downloads. For a content strategy, he would also examine signals such as domain authority and the pace at which new pages are created and discovered.

Before selecting metrics or hires, Balfour wants a hypothesis about how the product grows. A financial model asks how a dollar produces more than a dollar; a growth model asks how one user brings in additional users. The steps and constraints between those points can be quite different from the financial model.

He uses Loom, where he was an early investor and adviser, as an example of organic, viral and content loops. A user records a video and sends it to another person, who may start using the product. Educating users about one-to-many company communications can change how many people each video reaches. His interest in Loom is relevant to the example.

Understanding the loop helps identify the current constraint and determine which people to hire and which metrics to track. When deciding whether a strategy works, Balfour asks whether the team can improve the relevant inputs enough for the system eventually to reinforce itself, rather than requiring continual manual effort.

Stebbings says founders often understand their machine but cannot identify the bottleneck. Balfour challenges the first assumption: if a leader cannot draw a clear picture of how users enter, move through and create more users, the team may not share a useful model. A collection of ten disconnected tactics is not the same as a growth system.

Once the loop is mapped qualitatively, the team can assign a metric to each step, estimate how changes affect the output and compare that sensitivity with the difficulty of improving each lever. The most sensitive variable is not automatically the best target if it is extremely difficult to move.

Paid acquisition can be tested with a short spending spike. Balfour suggests increasing volume rapidly from a small daily amount and observing where performance deteriorates: creative, sign-up conversion or a later step. Content loops are harder to pressure-test because they take longer to compound and have fewer immediately adjustable levers.

He says bottlenecks often follow product patterns. A horizontal tool such as Loom can spread widely once users understand it, but activation is difficult because people have many possible use cases. A vertical tool serving one persona and problem may have a clearer activation path, even if its potential spread differs.

Stebbings proposes hiring growth before product-market fit to bring enough people into the product for learning. Balfour says the answer depends on the product and what the proposed hire actually does. Some companies need dedicated acquisition early; in others the relevant growth work is already embedded in product and engineering.

As a hypothetical, an enterprise analytics product with setup costs and a substantial purchase price may need someone focused on marketing-driven demand from day one. A consumer social product or a product-led tool like Loom depends more heavily on its product and viral loops. In that case, a separate growth hire cannot rescue a loop that does not work.

Balfour clarifies his claim that product-market fit is not enough: he means it is not sufficient on its own for a venture-scale business, which he illustrates with a $100 million-plus revenue ambition over a reasonable period. He views fit on two axes, its strength and the size of the market. Strong fit in a small market or weak fit in a huge one can still support a business without meeting a venture-scale goal.

He adds product-channel fit. A founder cannot simply build in isolation and bolt on any distribution channel afterward. Major platforms set their own rules, and a product must work within them. For a content-led strategy, the company or its users need to create content that the channel can surface; product instrumentation must support that loop.

Channel-model fit asks whether distribution works with the pricing and economics of the business. Balfour doubts a viral loop will turn effectively around an upfront $10,000 purchase: people may share it less and recipients may convert at a low rate. Sales can work for a higher-priced product, but revenue must cover the cost of human selling.

Stebbings asks whether an enterprise company should therefore avoid YouTube or other broad channels. Balfour says awareness can help, but it will not necessarily create enterprise customers by itself. In his software examples, the core sales machine must work first; supplementary channels should accelerate it rather than distract the team from it.

He sketches a possible product-led loop for Riverside, the recording tool used for the interview: help a user publish a video easily to YouTube, perhaps with branding that viewers see. Those viewers could then discover the product. It is a hypothetical design for a growth system, not a claim about Riverside's results.

Stebbings cites HubSpot's Kip Bodnar on concentrating on one successful channel before developing another. Balfour agrees that a growing company should focus scarce capital, talent and attention on its working core. Diversification introduces complexity; at the same time, leaders need to anticipate saturation early enough to build the next engine before growth stalls.

Predicting saturation is difficult. Balfour thinks teams often mistake an intermediate ceiling for the end of a channel's potential. HubSpot's content loop, he says, encountered repeated ceilings that were overcome with new ideas, though finding those ideas required difficult work rather than simply applying the old playbook.

He credits HubSpot with forward planning, recalling that around 18 months before its IPO it worked backward from a five-year growth goal. The team concluded that one marketing product and one sales motion would not suffice and began looking for a second and then a third product. He also offers an earnings-record observation as his recollection at the 2023 interview, not an independently verified current record.

To develop a new channel or product, Balfour warns against starting too late, underestimating the time required or allocating too many people immediately. He favors a small number of parallel bets, not ten. The goal is to keep the core running while giving a new engine enough time to develop.

At HubSpot, he says, founders Dharmesh Shah and Brian Halligan treated new products like venture investments. A small team of roughly four or five people received a year of seed-like funding and reported to internal leaders as a board. After a year it pitched for a larger next stage; successful bets were eventually integrated into the core business.

Overstaffing, in his view, brings more voices and slows iteration. He favors a few generalists, some protection from the rest of the organization, and clear guardrails and checkpoints. At Reforge he says he made the same mistake, stopped a new product bet and restarted it with a handpicked team of three.

Stebbings describes an enterprise company whose short videos attract attention but do not visibly produce sign-ups or revenue. Balfour's answer is to adapt the activity to the core business or stop it. A popular channel is not automatically valuable if it takes scarce time away from finding the mechanism that actually produces customers.

He recalls an early HubSpot sales product with referral incentives and paid acquisition that produced growth. His approximate recollection is that it reached around 100,000 weekly active users, but many were very small businesses rather than the mid-market customers central to HubSpot's strategy. The figure and the description are his retrospective account.

The team kept the mid-market target fixed and changed the product, channel and pricing. It built more features for sales leaders and administrators, reduced paid acquisition, retained some viral loops, and added content and inside sales. It also adjusted prices and minimum seat requirements to fit the new motion.

Balfour says teams should test whether an attention-generating activity can be molded to the product and market they actually serve. If those bets do not connect, continuing merely because the numbers look impressive can divert attention from the core growth engine.

His first metrics mistake is putting measurement before strategy. Metrics should help answer whether a growth hypothesis is working, so teams need a view of their core loop before deciding what to track. His second mistake is going straight to numbers without understanding the user and the underlying problem qualitatively.

Retention illustrates the issue. Before selecting weekly or monthly active users, Balfour asks what problem the product solves, how often that problem naturally occurs and what behavior shows it was solved. A metric with the wrong frequency or an empty definition of “active” can make a healthy or unhealthy product look misleading.

A third mistake, especially in SaaS, is focusing on revenue before usage. In his formulation, usage creates revenue; revenue does not itself create usage. Teams also need to distinguish the paying customer account from the people using it. At HubSpot, an administrator and an end user required different approaches even when both belonged to one customer.

On AI, Balfour expects surface-level tactics and tools to change substantially while the underlying growth questions remain. Analyses that once required specialist manual work, such as identifying an activation or “aha” moment, may become easier to automate. This is his 2023 expectation, not a measured result for every tool or team.

He doubts AI will fully replace the qualitative work of understanding a user's problem, its natural frequency and the psychology behind behavior. He summarizes growth as finding an arbitrage opportunity, using it to spark a compounding loop, optimizing that system and repeating before saturation. In his view, the tactics may change while this framework persists.

A new AI distribution channel could create opportunities as well as disrupt existing traffic. Balfour says changes that threaten Google-based acquisition may also create unfamiliar openings, much as the early Facebook platform did. His point is a prediction about possible market dynamics, not a guarantee that any particular tactic will work.

Asked about mistakes, Balfour says teams learn by making things, observing the result and revisiting their thinking. At HubSpot he became attached to an early product's viral growth despite its mismatch with the core target market. In hindsight, he says, a veto that redirected the product helped the team use HubSpot's existing mid-market knowledge and channels.

At Reforge, moving from transactions to subscriptions initially produced rapid growth, which he says obscured whether the product supported sustainable recurring use. The team was late to add features that encouraged habitual usage and had to work through a subsequent stall. He offers it as an example of impressive output metrics concealing weaker underlying behavior.

Balfour recalls an early Facebook virtual-gift app that automatically sent gifts to randomly selected friends of a new user. It grew so fast that it overwhelmed servers, but was spammy and sometimes sent a romantic-looking gift to an ex-partner. He calls it a mistake: the short-lived arbitrage did not create a sustainable system.

In the quick-fire round, he says word of mouth and users telling other users remain durable mechanisms. Direct messaging varies by context: he sees SMS as potentially intrusive, while noting that WhatsApp has supported growth in some markets. A tactic can fade because a platform changes its rules or users grow tired of abuse.

He says “growth” is not one interchangeable skill. Different systems call for different talent, metrics and methods, much as engineering contains many specialties. He names Casey Winters for strategy, Darius Contractor for user psychology and Guillaume Cabane for experimentation as practitioners he admires.

Asked about incumbents, Balfour expects AI both to alter and enhance existing growth tools. He considers Facebook strong because of its talent, capital, user base and data, while initially identifying Twitter as vulnerable. These are comparative views expressed at the time, not established subsequent outcomes.

Stebbings disagrees about Twitter and predicts Threads will struggle with retention. They compare rough user figures and offer one-year daily-active-user forecasts; Balfour estimates Twitter might remain flat while Stebbings expects modest growth. Their figures and forecasts belong to the 2023 discussion and should not be read as verified results.

Balfour frames Twitter's challenge as increasing the size of the market while retaining strong product-market fit with existing communities. Changes intended to attract a broader audience could weaken the original fit. He links the pressure to expand to the economics of the acquisition, presenting this as his interpretation.

He says a failure by Threads to overcome Twitter's network would illustrate the strength of network effects despite a rival's talent, distribution, data and capital. The open question, in his view, is whether Threads can attract the right creators quickly enough to make its network sustainable. This is a contemporaneous hypothesis, not a conclusion about its later performance.

Asked which growth strategies impress him, Balfour notes that Reforge's membership gives him a partial, potentially biased view of many companies. He names Canva, Faire and DoorDash among examples, while distinguishing a rapidly growing product from a team that had to build a difficult growth engine. He attributes OpenAI's early growth more to technology and product than to a particular growth strategy; these are his assessments.

He also praises the talent density he sees at Shopify and the recruiting work behind it. Stebbings thanks Balfour for the discussion, closing an interview that moved from growth loops and channel fit to company-building mistakes and AI's possible effects.