Make It Easier for Customers to Bring in More Customers

Task

Build product-led growth and referral loops.

Summary

Use invitations, sharing, referrals, product use, and natural expansion to help good customers create more demand.

Build Product-Led Growth and Referral Loops That Produce Real Expansion

Task ID: S5-05

Product-led growth works when using the product naturally introduces it to more people, creates more value for the existing customer, and gives the company a measurable path to paid expansion. This article explains how to choose the right loop, launch a credible test, measure activation and downstream value, and avoid mistaking invitations, shares, or free signups for durable growth.

The growth loop must be part of the product

A software company adds an “Invite a colleague” button, offers a referral credit, and sends reminder emails. Invitations rise. New accounts appear. The dashboard looks encouraging.

Three months later, most invited users have never completed the first useful task. Few accounts have added paid seats. Support is handling reward disputes and duplicate accounts. The company has created more activity, but not more valuable customers.

That is the problem this work should solve.

Product-led growth, or PLG, means that the product performs a meaningful part of customer acquisition, conversion, retention, or expansion. A growth loop is the repeatable path through which one user’s successful use helps produce another user, more use, or more revenue. A referral program is only one possible loop, and often not the strongest one.

The operating principle is straightforward:

Do not ask customers to distribute the product until distributing it helps them do useful work.

A good loop has value on both sides. The existing user has a practical reason to share, invite, publish, collaborate, or increase usage. The recipient can understand and experience value without completing a long sales or onboarding process. The product can then observe whether that interaction leads to activation, retention, or paid expansion.

This work belongs after the company has a product that customers can buy, adopt, and use with reasonable consistency. A formal dependency may not have been specified, but several practical prerequisites remain:

• The team can define the customer’s first meaningful result.
• Onboarding works well enough that new users do not routinely require rescue.
• Product events can be connected to users, accounts, plans, invitations, and payments.
• Pricing has a logical relationship to seats, usage, features, or customer value.
• The company can support additional users without causing a sharp decline in service quality or margin.

Without these conditions, a growth loop usually magnifies a weak experience. It may generate more signups, but it also sends more people into the same onboarding failures, unclear pricing, or support bottlenecks.

What the evidence says about growth and referrals

The research does not support the idea that adding rewards or sharing buttons automatically creates rapid growth.

An analysis of 16 million recommendations among four million people found that recommendation cascades were generally short and that recommendations were not very effective at causing purchases on average. Results varied substantially by product, community, and price category. The practical lesson is not that referrals fail; it is that broad averages hide where they work. A company should identify the particular users, moments, and offers for which sharing is useful rather than assume every customer can become a distributor.

A randomized field experiment involving the social connections of 9,687 users found that product design could cause measurable peer adoption. Passive broadcasts produced more total adoption because people used them more often, while personalized messages were more effective per message and were associated with stronger engagement and continued use. This creates an important design choice: a low-effort share may create more reach, while a deliberate invitation may produce fewer but better recipients. Both should be measured through activation and retention, not message volume alone.

Research on approximately 10,000 bank customers found that referred customers had higher retention and, after adjusting for observable differences, were worth at least 16% more on average than comparable non-referred customers. The advantage varied by customer segment, and some of the contribution-margin difference declined over time. This is useful evidence for selective referral programs, not a universal promise that every referred customer is more valuable.

The incentive itself can also change how a recommendation is perceived. Experiments comparing cash and non-cash referral rewards found that monetary incentives can create a social cost: the recipient may question whether the recommendation is sincere. Cash performed worse in some conditions, especially when the brand was not already strong. Rewarding both the sender and recipient, or using a reward closely connected to the product, reduced some of that problem.

These findings suggest that a company should distinguish several kinds of loop rather than place every growth idea under “referrals.”

A referral is primarily an acquisition mechanism. An invite can be both an acquisition and an expansion mechanism. A usage threshold can produce expansion without introducing a new customer. Shared content can create awareness but may not create a true loop unless recipients can enter the product, receive value, and repeat the process.

A useful growth system may contain several of these paths, but the first test should isolate one clear mechanism.

Lessons from products that made sharing useful

Dropbox provides a clear example of the difference between a product action and a promotional request. Its business model has long included direct referrals, content sharing, and collaboration with people who do not yet have accounts. Its current referral program gives additional storage to both participants after the new account is created and validated. The reward is part of the product’s utility rather than an unrelated prize.

Dropbox’s filings also show why reach cannot be treated as proof of conversion. For the year ended December 31, 2025, the company reported more than 700 million registered users but 18.08 million paying users. It generated more than 90% of its revenue through self-service channels and described word-of-mouth referrals, direct product referrals, content sharing, prompts, trials, email, and lifecycle marketing as parts of its model. The large difference between registered and paying users is a useful warning: even a well-known product with strong distribution must keep working on conversion and paid value.

Zoom illustrates a participation loop. A host invites other people because a meeting requires attendees. Those participants can experience the product before deciding whether to become hosts or paying customers themselves. Zoom’s 2026 annual filing still describes free users and meeting attendees as part of its customer-acquisition model, alongside online sales, direct sales, resellers, and partners. The product loop supports the other channels; it does not eliminate them.

Atlassian illustrates expansion within a company. Its 2025 annual filing describes self-service adoption followed by expansion to more teams, additional applications, and higher-value editions. Its direct sales team focuses mainly on deeper relationships with larger existing customers rather than replacing product adoption. This is an important boundary: PLG and sales do not have to compete. Product use can identify where human help is economically justified.

The examples teach three different lessons:

Dropbox makes sharing and the referral reward useful inside the product. Zoom exposes recipients to the product while they are completing a real task. Atlassian lets initial team use create evidence for broader account expansion.

None provides a universal formula. Dropbox’s storage reward would make little sense for software in which storage is unimportant. Zoom’s participant loop depends on a naturally multi-person activity. Atlassian’s model assumes that teams can begin using the product before a company-wide purchasing decision.

The design must follow the customer’s work rather than the fame of the example.

How to design and launch the first test

Start by tracing one successful customer’s path from first value to expansion. Do not begin with referral copy, rewards, badges, or email sequences.

Ask four questions:

What useful result has already happened? A customer should normally reach a meaningful result before being asked to recommend the product. A referral request immediately after account creation measures willingness to click, not informed advocacy.

Who naturally needs to see or participate in that result? The answer may be a colleague who approves work, a client who receives a report, an employee who contributes data, or a peer with the same problem.

What value will the recipient receive immediately? “Create an account to learn more” is weak. “Review this report,” “approve this request,” “join this workspace,” or “use this template” is concrete.

What commercial event could follow? The result might be a new account, another active seat, greater consumption, an additional product, or a higher plan. The path should be explicit before the test begins.

A basic loop looks like this:

flowchart LR
    A[Existing user reaches useful result] --> B[Product presents a relevant share or invite]
    B --> C[Recipient receives immediate value]
    C --> D[Recipient completes an activation event]
    D --> E[Team, account, or usage expands]
    E --> F[Paid expansion or new paying account]
    F --> G[New active user can repeat the action]
    G --> B

The diagram shows a complete loop: first value leads to a useful invitation; the recipient reaches value; the account or customer base expands; another active user can repeat the process. A sequence that ends with an email sent or a link opened is not yet a complete loop.

Choose the smallest credible intervention

The first version should test the core behaviour with limited engineering work. Examples include:

• A contextual invitation after the user completes a collaborative task.
• A shareable result page with a clear recipient action.
• A notice when an account approaches a meaningful usage or seat threshold.
• A two-sided product credit offered only after sustained successful use.
• A customer referral code that the customer can share through their own chosen channel.

The intervention should appear at the moment it is useful. Permanent banners and repeated pop-ups may increase exposure while reducing trust. A request after a successful result is easier to interpret and less likely to interrupt unfinished work.

Define eligibility before exposure

Do not show the same request to every user. Eligibility might require that the user has completed the main activation event, used the product on several days, produced a shareable result, achieved a minimum satisfaction signal, or belongs to a segment with strong retention.

Selective eligibility serves two purposes. It avoids asking poorly served users to recommend the product, and it creates a cleaner experimental population.

Instrument the full path

The event model should connect:

1. The originating user and account.
2. Eligibility for the loop.
3. Exposure to the invitation, sharing option, or upgrade prompt.
4. The user’s action.
5. Delivery to or access by the recipient.
6. Recipient registration or account association.
7. Recipient activation.
8. Subsequent retention, payment, seat expansion, or usage expansion.
9. Reward qualification, issuance, reversal, and possible abuse.

Identity resolution is often the difficult part. A recipient may use another email address, already belong to the company, open a shared link without registering, or later join through a salesperson. Decide how attribution will work before launch rather than assigning credit manually afterward.

Use an experiment that can support a decision

Randomized controlled experiments are valuable because they can distinguish the effect of a product change from behaviour that would have occurred anyway. Microsoft’s published work on online experimentation emphasizes that many plausible ideas produce neutral or unexpected results, which is why prediction alone is unreliable.

For a straightforward upgrade prompt, randomization by user may be sufficient. For invitations and collaborative features, one treated user can affect another person’s outcome. This is known as interference. Randomizing by account, workspace, organization, or connected group can reduce contamination between treatment and control populations. Research on cluster-randomized experiments shows why ordinary individual-level tests can be biased when treatment effects spill across connected users.

Before launch, record the hypothesis, eligible population, treatment, control, primary measure, guardrails, expected duration, owner, and decision rule. Extending or changing these after viewing early results makes the evidence less trustworthy.

Measure expansion, not motion

The primary measure for this task is expansion loop activation. That term should be defined operationally for the company rather than used as an industry benchmark.

A practical definition is:

Expansion loop activation rate = eligible active accounts that complete a qualifying expansion action ÷ all eligible active accounts exposed to the opportunity

The qualifying action should be stronger than opening a modal or clicking a button. Depending on the loop, it might mean:

• An invited recipient completes the product’s activation event.
• An existing account adds an active participant.
• A shared result produces a qualified new account.
• An account crosses a sustained paid-usage threshold.
• A customer adopts an additional paid product or feature.
• A referred customer reaches first value and remains active through a defined observation period.

Several supporting measures are needed to explain the result.

Action rate measures how many eligible users send, share, invite, or begin the upgrade path.

Recipient activation rate measures activated recipients divided by delivered invitations or qualified shared visits.

Loop cycle time measures the median time from the originator’s activation to the recipient’s activation or account expansion. A loop that takes six months behaves differently from one that completes in two days.

Qualified viral coefficient can be estimated as:

Average qualified invitations per eligible active user × recipient activation rate

The word qualified matters. Invitations to invalid addresses, duplicate accounts, existing users, bots, or people who never reach value should not count. The result is useful for comparing versions of the same loop, but it should not replace retention or revenue measures.

Expansion conversion measures the proportion of eligible accounts that generate an additional paid seat, higher plan, additional product, or sustained usage charge.

Incremental lift compares the treatment group with an appropriate control. Without a control, the team may credit the loop for invitations or upgrades that would have happened anyway.

Quality and economics should include recipient retention, expansion revenue, gross margin, incentive cost, support effort, fraud loss, and any increase in churn among existing customers. A loop can grow revenue while destroying margin, attracting unsuitable customers, or making the experience worse for established users.

The working target, “test launched,” is a sensible milestone for a backlog item, but it is not proof that the loop works. A credible launched test means that:

• The change is operating in the intended production environment.
• A defined eligible population is receiving treatment or control.
• End-to-end events are being recorded and checked.
• The primary measure and guardrails were set in advance.
• An owner and decision date exist.
• The company can pause the test if users, compliance, system performance, or economics are harmed.

The next target should depend on actual baseline volume, sales cycle, product price, activation frequency, and statistical power. A product with thousands of daily collaborative actions may reach a reliable result quickly. Enterprise software with ten new accounts a month may need account-level analysis, longer observation, qualitative evidence, or several coordinated tests. A single universal target would conceal these differences.

Build a backlog that can be tested

The required evidence should be a prioritized backlog of referral, sharing, invite, usage-based upgrade, or other growth-loop candidates. A list of feature names is insufficient. Each item should state why the behaviour may occur, how it benefits both sides, and what evidence would justify further investment.

The following structure is ready to use:

A live backlog should also record evidence confidence, expected engineering effort, data readiness, legal review, segment size, and the reason for priority. Scoring can help organize discussion, but it should not replace judgment. A simple, high-volume sharing loop with clear instrumentation may deserve priority over a theoretically valuable referral program that requires identity changes, billing work, and manual fraud review.

The backlog should contain alternatives. The team may discover that a referral reward is weaker than a product share, or that an additional-seat loop is stronger than new-account acquisition. Maintaining several mechanisms prevents the company from forcing every growth problem into an incentive program.

After each test, preserve the result even when it fails. Record the population, exposure, primary outcome, guardrails, limitations, and decision. A rejected idea with credible evidence is more valuable than an indefinitely “promising” item with no test.

Failure modes and the decision to scale

The most common mistake is to treat the visible action as the result. Shares, invitations, referral codes, upgrade-page visits, and free accounts are intermediate events. The business result occurs later, when the right customer receives value, stays, expands, and produces workable economics.

Other failure modes require equal attention.

The product asks before earning trust. A referral request shown immediately after signup may produce accidental clicks or incentive-seeking behaviour. It cannot demonstrate that the user has enough experience to make an informed recommendation.

The sender benefits but the recipient does not. A reward may motivate the existing user while imposing a sales message and onboarding burden on another person. Two-sided utility is generally stronger: useful content, access to a shared task, an allowance that supports first value, or a reward for both parties.

The incentive is disconnected from product value. Cash, gift cards, or unrelated prizes can attract participants who care more about the reward than the product. They can also make a personal recommendation feel less sincere. Product credits, additional capacity, or benefits tied to successful use may fit better, although their effects still need testing.

The account grows but does not become healthier. Additional seats may be purchased but remain inactive. Higher usage may come from inefficient workflows rather than added value. Expansion should be examined alongside retained activity, support effort, customer outcomes, and renewal.

The loop depends on spam. Automated “refer a friend” email creates legal and reputational risk. In the United States, commercial email is subject to requirements concerning sender information, deceptive subject lines, advertising identification, postal addresses, and opt-outs. UK privacy guidance is stricter about businesses instigating refer-a-friend messages and warns that valid recipient consent may be difficult to establish. Design and review the mechanism for each relevant jurisdiction rather than assuming that a user-entered address transfers responsibility to the user.

A safer design may let customers copy a personal code or link and choose how to share it, rather than asking the company to send marketing directly to another person. The correct implementation depends on the message, jurisdiction, customer relationship, and product, so legal review remains necessary.

The reward is hidden. In the United States, the Federal Trade Commission’s endorsement guidance requires clear disclosure when a material connection could affect the credibility of an endorsement. A referral reward, free product, or other benefit may create such a connection.

The team scales an observational correlation. Users who invite colleagues may already be the most engaged customers. Their higher retention does not prove that the invitation feature caused retention. A controlled test, staged rollout, or carefully matched comparison is needed before assigning causal credit.

The product is not naturally shareable. Some products are individual, confidential, heavily regulated, rarely used, or expensive to implement. A customer-reference process, expert consultation, partner referral, or sales-assisted expansion may be more suitable than an automated invite loop. Product-led does not mean product-only.

The work is complete enough to support the next decision when the company has:

A clearly defined loop tied to real customer work; an instrumented and prioritized backlog; at least one production test with an explicit hypothesis and control or comparison; a definition of expansion loop activation; downstream measures for retention, revenue, cost, and quality; and a documented decision to stop, revise, continue, or scale.

The central decision is not whether the product has a referral feature. It is whether a repeatable customer action creates enough value for another person to join, activate, remain, and contribute to profitable expansion.

Sources

Primary sources

• Dropbox, 2025 Form 10-K, including its descriptions of content sharing, direct referrals, self-service revenue, registered users, and paying users.
• Dropbox, official referral-program documentation.
• Zoom Communications, 2026 Form 10-K, including its free-user, meeting-attendee, referral, and multi-channel acquisition model.
• Atlassian, fiscal 2025 Form 10-K, including its self-service, team-expansion, product-expansion, and direct-sales model.
• U.S. Federal Trade Commission, CAN-SPAM Act and rule guidance.
• U.S. Federal Trade Commission, revised endorsement guidance.
• UK Information Commissioner’s Office, electronic-marketing guidance for refer-a-friend programs.

Open research

• Aral and Walker, “Creating Social Contagion Through Viral Product Design: A Randomized Trial of Peer Influence in Networks.”
• Leskovec, Adamic, and Huberman, “The Dynamics of Viral Marketing.”
• Schmitt, Skiera, and Van den Bulte, “Referral Programs and Customer Value.”
• Jin and Huang, “When Giving Money Does Not Work: The Differential Effects of Monetary Versus In-Kind Rewards in Referral Reward Programs.”
• Kohavi and colleagues, “Online Experimentation at Microsoft.”
• Brennan, Mirrokni, and Pouget-Abadie, “Cluster Randomized Designs for One-Sided Bipartite Experiments.”