Consumer subscription businesses get analysed to death. Streaming, software, fitness, meal kits. The models are well understood, the cohort curves are public, and the multiples are priced accordingly.

Consumer video communication is not. It sits in a corner of the market most analysts skip, partly because the category is unglamorous and partly because almost nobody in it publishes a number. That is a shame, because the cost structure is unusually legible and the failure mode is unusually clean.

Here is the conclusion before the detail. The free advertising-supported model that carried consumer social video for roughly fifteen years was never viable. Not mismanaged, not out-competed, not beaten by a better product. Structurally unviable from the first day, because its cost curve and its revenue curve pointed in opposite directions and no amount of scale made them converge. It ran as long as it did because a small number of founders absorbed the deficit personally, which is a financing arrangement rather than a business model.

What replaced it charges money, and it works. The paid version fixed the direction of those two curves. Same technology, same users, largely the same product. The only thing that changed was who pays and when.

That makes this a useful case for anyone who prices platform businesses, because the variables are few and the causal chain is short. Three properties make it cleaner than most: marginal cost is real and visible rather than rounding to zero, the trust and safety cost is large and scales with humans rather than servers, and payment processing imposes hard operating constraints that show up directly in product design. You can read the economics off the product.

What follows is a full decomposition: the cost structure that killed the free model, the four pricing variants that replaced it, a worked contribution-margin example on a single session, how revenue is split with the supply side and how that compares to Twitch, YouTube, Cameo and Patreon, why payment processing rather than technology is the binding constraint, why customer acquisition in this category behaves unlike any normal consumer funnel, and where the structure suggests the category goes next.

Several figures below are estimates rather than reported company numbers. Each one is labelled as such. Reported figures are attributed. Treat the estimates as worked examples with stated assumptions, not as industry benchmarks.

Why this category is a useful case study

Three properties make consumer video a cleaner analytical example than most platform businesses.

First, the marginal cost is real and visible. Most software platforms have marginal costs close to zero, which is what makes them attractive and also what makes their economics slightly boring to analyse. Live video does not work that way. Every additional minute of usage consumes bandwidth that someone pays for. The cost of serving a user scales with how much they use the product, which puts this category closer to a utility than to a typical SaaS business.

Second, the trust and safety cost is unusually large and unusually explicit. In most consumer platforms, moderation is a line item that leadership would rather not discuss. Here it is a primary cost driver, it scales with volume, and it is paid in human hours.

Third, the category sits in a regulatory and payments position that forces every structural trade-off into the open. Payment processors impose requirements that shape the product. That means you can read the business model directly off the product design, which is rarely possible.

Those three properties together mean the unit economics cannot be hidden behind scale assumptions. Either a session makes money or it does not.

Part one: the model that failed

For roughly fifteen years the default structure for consumer social video was free access supported by display advertising. Users paid nothing. The platform sold ad inventory. Growth was the strategy and monetisation was assumed to follow.

To understand why this collapsed, separate the two sides of the income statement.

The cost side

Bandwidth and relay. A live video session between two people needs a path between them. Where both parties can connect directly, the platform pays only for signalling, which is trivial. Where they cannot, and a meaningful share cannot, the session must be relayed through a server the operator pays for, which means buying bandwidth for a live stream in both directions for the full duration of the call.

Here is the arithmetic, with assumptions stated. Assume a session running at 1 Mbps of combined audio and video per direction. That is 60 megabits per minute per direction, or 7.5 megabytes. The relay receives from each party and sends to the other, so two directions of egress, giving roughly 15 MB per minute, or about 0.9 GB per hour of relayed session. At 1.5 Mbps, a more realistic figure for good quality video, it is closer to 1.35 GB per relayed hour.

Now the share that relays. This depends entirely on the network mix of your users and cannot be assumed from someone else's data. A common planning assumption is 15 to 20 percent of sessions, and that is an assumption rather than a measurement. Operators who do not instrument this from the start frequently discover their real figure is far outside that range in one direction or the other.

Take 100,000 relayed hours in a month at 1.35 GB each and you have 135 TB of egress to pay for. Multiply by your provider's per-gigabyte rate. On metered cloud egress that is a significant bill. On bandwidth-inclusive bare metal it can be close to nothing beyond the hardware. The gap between those two procurement decisions is often larger than every other infrastructure cost combined, which is worth noting because it is a decision usually made early and casually.

Moderation. This is the cost that ends companies, and it is structurally different from infrastructure cost because it does not benefit from scale.

Live video cannot be pre-moderated. There is no review queue before publication because content is delivered at the instant it exists. So the platform reviews reports after the fact, runs automated classifiers on sampled frames during sessions, and employs people to make the judgement calls that classifiers cannot.

An illustration with assumed rates, not reported ones. Take a platform doing 100,000 sessions a month with a 0.5 percent report rate. That is 500 reports. At four minutes of average handling time, roughly 33 hours of review work a month, which is a part-time role. Now scale the platform to 10 million sessions at the same rates. That is 50,000 reports, 3,333 hours, and somewhere around 20 full-time reviewers before you count team leads, tooling, quality assurance, wellbeing support and coverage across time zones.

The important feature of that progression is that it is linear. Infrastructure cost per session falls with scale because you negotiate better rates and use capacity more efficiently. Moderation cost per session does not fall, because the input is human attention and human attention does not get cheaper. A platform that grows 100x grows its moderation cost roughly 100x.

Compliance and legal. Identity and age verification, record keeping, reporting obligations that differ by jurisdiction, and legal exposure. After several court decisions in the early 2020s established that platform design itself could be grounds for product liability claims rather than being shielded as a publishing function, the tail risk in this category stopped being a fine and started being the company.

The revenue side

Now the other half, which is where the model actually broke.

Advertising revenue depends on inventory quality, and inventory quality here is close to the worst available on the open internet. Brand safety systems used by major advertisers automatically flag unpredictable user-generated video. Premium advertisers will not buy it at any price, not because of any individual platform's behaviour but because the category classification is enough to disqualify it.

What remains is the low end of programmatic inventory, sold at CPMs that generally do not cover the bandwidth for the session that produced the impression, let alone the moderation.

There is a second problem that is less obvious and arguably worse. Advertising revenue is a function of impressions, which is a function of time on site. That means an ad-supported platform is structurally incentivised to maximise session duration regardless of what happens during the session. The incentive and the safety objective point in opposite directions, permanently.

The structural conclusion

Put the two sides together.

Costs rise roughly linearly with usage, with the largest component being human labour that does not get cheaper. Revenue is capped by a ceiling the operator does not control, set by advertiser risk policy rather than by audience size or engagement. And the incentive structure pushes against the safety spending that keeps the business alive.

That is not a business with an execution problem. Those unit economics were negative from the start and stayed negative at every scale. The free model in this category survived as long as it did because founders subsidised it personally, and it ended when the legal exposure came due.

For anyone pricing platform businesses this is the central lesson of the case: a model can be structurally unviable and still operate for a decade, because the losses can be absorbed quietly by a small number of people. Longevity is not evidence of viability.

Part two: the shift to paid, and the four variants

Once the free model was visibly dead, operators who intended to continue had one option, and the transition produced four distinct pricing structures. Each has different economics and different failure modes.

Per-minute billing. The user buys a balance and is charged per minute of session time. This is the most direct mapping of price to cost, since the dominant variable cost is also per minute. Revenue scales with usage in exactly the way bandwidth does, which makes gross margin stable and predictable.

The weakness is behavioural. A visible meter changes how people use a product. Sessions get shorter, users watch the counter rather than the conversation, and the experience degrades in a way that reduces repeat purchase. Operators know this and it is why almost nobody displays a live per-minute charge prominently.

Coin or credit economies. The user buys a bundle of tokens at one price and spends them inside the product at prices denominated in tokens. This is the dominant structure in the category and the reason is behavioural rather than financial.

Decoupling the purchase from the spend removes the meter effect. The user makes one purchase decision, then makes usage decisions in a currency that does not feel like money. It is the same mechanic that makes casino chips and mobile game currencies work, and it is well documented in behavioural economics.

It also gives the operator pricing flexibility that direct currency pricing does not. Bundle sizes can be tiered to encourage larger purchases. Different actions can carry different token prices, adjusted without changing any displayed money price. Promotional balances can be issued at zero marginal cost. And unspent balances sit on the books as deferred revenue, which improves working capital.

The weakness is regulatory and accounting complexity. Stored value attracts scrutiny in several jurisdictions, refund handling is harder, and unspent balances are a liability rather than earned revenue until consumed.

Subscription. A flat recurring fee for access or for an allowance of usage. Revenue is predictable, which financiers like, and customer lifetime value is easy to model.

The weakness is the mismatch between flat revenue and variable cost. A subscriber who uses the product ten times more than average costs ten times more to serve and pays the same. Any subscription model in a business with real marginal cost needs either a usage cap, a fair use policy, or a distribution of usage heavily weighted toward light users. Most operators in this category use subscription for feature access and a coin system for usage, which resolves the mismatch.

Freemium. Not a pricing model in itself but an acquisition structure layered on top of one. A free tier exists to produce the first session, because the conversion problem in this category is getting someone to their first real experience of the product.

The cost discipline that freemium requires is stricter here than in software. In a typical freemium SaaS product, an unconverted free user costs approximately nothing. Here, a free user consumes bandwidth and generates moderation load, so the free tier is a real expense that must be capped and measured against the conversion rate it produces.

In practice most premium video chat platforms run a hybrid: a capped free tier for acquisition, a coin economy for core monetisation, and an optional subscription for status and feature access.

Part three: unit economics, and where the margin sits

This is the part worth working through carefully, because the answer is not where people expect.

Take a single paid session and decompose it. All figures below are illustrative assumptions chosen to show the structure, not reported industry numbers.

Assume a user spends coins equivalent to $10 on a session. Work down the income statement.

Payment processing. Standard consumer card processing runs roughly 2.9 percent plus a fixed fee. Merchants classified as higher risk, which most of this category is, pay considerably more. Reported ranges for high-risk processing generally sit somewhere between 3.5 and 8 percent plus per-transaction fees, often with a rolling reserve of 5 to 10 percent of volume held for around 180 days. Take 6 percent as a mid-range assumption. That is $0.60, and the reserve is a cash flow cost rather than an expense, but it is real.

Infrastructure. Using the relay math from earlier: if the session is shorter than a relayed hour at 1.35 GB, the bandwidth cost is cents. Call it $0.10 as a generous assumption including signalling and platform overhead. This is the smallest line on the page, which surprises people who assume video is expensive to serve.

Revenue share with the host side. The largest line by a wide margin. Reported splits across the category vary considerably and many are not published. A 30 to 50 percent share to the earning side is a reasonable estimate range. Take 40 percent, which is $4.00.

Moderation and trust and safety, allocated per session. Using the illustration from part one, if a platform runs 20 reviewers against 10 million sessions a month, allocated fully loaded cost per session lands in the low single digits of cents. Call it $0.05. This looks trivially small per session and is not trivial in aggregate, which is the standard trap with allocated costs.

Chargebacks and fraud. Assume 1 percent of gross revenue lost to disputes plus fees, which is $0.10. Note that this number is not merely a cost. It is also a compliance threshold, which is covered below.

Add those: $0.60 plus $0.10 plus $4.00 plus $0.05 plus $0.10 is $4.85. Contribution margin is $5.15 on $10, or about 51 percent before any fixed costs.

Two observations follow, and they are the useful output of the exercise.

The first is that the margin is dominated by a single decision. The revenue share with the host side is roughly eight times larger than every other variable cost combined. Infrastructure, which is what people assume drives the economics of a video business, is about 2 percent of revenue. Anyone modelling this category and focusing on server costs is optimising a rounding error while the actual lever sits untouched.

The second is that contribution margin of around 50 percent is healthy but not software-like, and it has to cover a fixed cost base that includes engineering, compliance, the fixed portion of trust and safety, and customer acquisition. Whether the business works depends almost entirely on that last item, which is why part six matters more than this section does.

Part four: revenue splits in context

Since the split with the earning side is the dominant variable, it is worth benchmarking against comparable platforms. These are publicly stated rates from other creator platforms, useful as reference points.

Twitch's standard subscription split is 50/50 between platform and streamer, with more favourable terms available to some partners. YouTube's Partner Program pays creators 55 percent of associated advertising revenue. Cameo's published structure takes 25 percent, leaving 75 percent to the talent. Patreon's platform fee sits in the high single digits to low teens depending on the plan selected, which is much lower because Patreon provides payment rails and a storefront rather than an audience.

That range, from under 15 percent to around 50 percent, maps closely onto how much demand the platform generates versus how much the creator brings.

The logic is consistent. A platform that provides only payment infrastructure and hosting, where the creator arrives with their own audience, can charge a low fee. A platform that provides the audience, matching and discovery, where a host with no external following earns because the platform sent someone, captures much more, because it is supplying the scarce input.

Premium video chat platforms sit at the high end of that range for exactly that reason. Demand generation is the platform's contribution, and demand generation is expensive in this category for reasons covered below. A host on one of these platforms does not typically bring their own audience.

There is a strategic tension worth naming. Raising the platform's share increases margin per session but reduces the supply side's earnings, which affects retention on the side of the market that is harder to replace. Supply-side retention is usually the binding constraint in two-sided marketplaces, because a user who leaves can be replaced by marketing spend while a host who leaves takes accumulated reputation and repeat customers with them. Most operators who have optimised aggressively on split have discovered this the expensive way.

Part five: payment processing, the hardest problem in the category

Ask anyone who has operated in this category what the hardest problem is and the answer will not be technology. It is getting paid.

High-risk classification and what it costs

Payment processors classify merchants by risk, and the classification drives everything downstream. A high-risk classification means higher processing rates, rolling reserves that hold a percentage of revenue for months, longer underwriting, more documentation, and a standing possibility of account termination with limited notice.

The classification is applied by category, not by individual conduct. An operator can run a clean business with low disputes and still carry the classification, because the classification attaches to what the business does rather than how well it does it.

Chargeback thresholds as a hard constraint

This is the part most analyses miss, and it is the single most important operational constraint in the category.

Card networks run monitoring programs with defined thresholds. Under Visa's Acquirer Monitoring Program, the merchant ratio moves to 1.5 percent from April 1, 2026, reduced from 2.2 percent. For merchants in the United States, the excessive tier requires both a VAMP ratio of 1.50 percent and at least 1,500 combined fraud and dispute items in a calendar month. Merchants in the excessive tier incur a fee of $8 per dispute. Mastercard's Excessive Chargeback Merchant program triggers at a ratio of 1.5 percent combined with at least 100 chargebacks in a month, with a higher tier at 3 percent and 300 chargebacks. Breaching these thresholds brings per-dispute fees, monthly assessments, and the possibility of being listed on MATCH, which effectively bars a merchant from obtaining processing elsewhere for five years.

Read that as a business constraint rather than a compliance detail. It means a premium video chat platform is operating with a hard ceiling on dispute rate, and exceeding it does not produce a fine, it produces the end of the ability to accept payments at all. Practitioners generally target well under 0.65 percent to stay clear of every program simultaneously.

This single constraint explains an enormous amount of product design across the category. Identity verification at signup exists partly for safety and substantially to reduce disputes. Clear billing descriptors exist because unrecognised line items on a statement are a leading cause of disputes. Spending caps, confirmation steps before large purchases, accessible transaction history and responsive refund policies all trace back to the same number. When a product in this category adds friction to a purchase flow, the reason is usually the dispute ratio rather than anything to do with user experience.

The verification cost stack

Meeting processor requirements means building and paying for a compliance function: identity verification for users and more rigorous verification for anyone on the earning side, age assurance, sanctions and watchlist screening, transaction monitoring, record retention, and the staff to operate all of it.

Per-verification costs for document checking services are typically quoted in the low single digits of dollars. That is not a large number until you multiply it by every signup including the ones that never convert, at which point verification becomes a meaningful component of customer acquisition cost rather than a compliance line.

The app store layer

Platforms with mobile apps face a second intermediary. Apple's standard commission is 30 percent, reduced to 15 percent for Small Business Program members, who qualify at up to $1 million in prior-year proceeds, and for subscriptions after the first year.

That situation is currently unsettled in the United States. Following the Epic v. Apple contempt ruling and the Ninth Circuit lifting Apple's stay in April 2026, with the Supreme Court declining to pause it in May 2026, US apps can currently link out to external payment for no Apple fee. The Supreme Court agreed in June 2026 to hear the case, and Apple has proposed charging up to 15 percent on external link purchases, so the current zero rate should be treated as a window rather than a settled rate.

The magnitude matters. A 30 percent platform commission is larger than the entire contribution margin calculated in part three. Any operator in this category whose primary funnel runs through a mobile app has a fundamentally different business from one whose users pay on the web, and this is why so many of these products are web-first by deliberate choice rather than by neglect.

Part six: customer acquisition, and why search beats paid media here

For most consumer subscription businesses, customer acquisition is a solved arithmetic problem. Estimate lifetime value, spend up to some fraction of it to acquire a customer, measure payback period, scale the channels that work. The constraint is capital.

In this category the constraint is access, and that changes the entire strategy.

Paid channels are restricted by policy

Google operates a Dating and Companionship advertising policy, effective March 4, 2025, that requires advertisers to complete a certification process before running ads in the category. Restrictions vary by ad type, user age, local law and user search behaviour. Certain subcategories are limited or prohibited outright, and the policy lists a set of countries where dating and companionship ads are not eligible to serve at all, including Algeria, Bahrain, Egypt, Iraq, Jordan, Kuwait, Lebanon, Libya, Morocco, Nepal, Oman, Pakistan, Palestine, Qatar, Saudi Arabia, Sri Lanka, Tunisia and Yemen.

Other large ad networks apply comparable restrictions with their own approval processes and their own exclusions.

The practical consequence is that the standard consumer growth playbook is either unavailable or available only in a restricted and unpredictable form. An operator cannot simply raise a round and buy growth, because the inventory is gated by policy rather than by price. Approval can also be withdrawn, which makes any acquisition strategy built primarily on paid media structurally fragile regardless of how well it performs.

Why organic search behaves differently

Search visibility is not gated the same way. Organic results are governed by relevance and authority rather than by advertising policy, and a page that ranks continues to deliver traffic without per-click cost.

That produces a different financial profile. Paid acquisition is an operating expense with linear returns: spend stops, traffic stops. Organic acquisition behaves more like a capital investment, with high upfront cost, a long delay before return, and a durable asset at the end that produces traffic at near-zero marginal cost.

For a business that cannot reliably buy traffic, the second profile is not merely preferable. It is frequently the only option that scales.

This is why operators in restricted categories generally build content and search operations well before they would be a priority for an unrestricted consumer product, and why the content produced tends toward genuinely informational material. Thin promotional pages do not rank in competitive categories, and in a restricted category there is no paid fallback when they fail.

The acquisition arithmetic

Work the numbers with stated assumptions, which are illustrative rather than reported.

Suppose average revenue per paying user is $60 over their lifetime on the platform, and contribution margin is 50 percent as calculated earlier. Contribution per paying user is $30. Assume a 3 percent conversion from registered free user to paying user, which means a registered user is worth $0.90 in expected contribution.

Now the free tier cost. Every registered user consumes verification cost, some bandwidth, and some moderation attention whether or not they ever pay. If that comes to $0.40 per registration, expected contribution per registration falls to $0.50.

That is the ceiling on what can be spent to acquire a registration and still break even, before any fixed cost. At those assumptions, paid media at typical consumer cost per registration is marginal at best even where policy permits it. Organic acquisition at an effective cost approaching zero per incremental visitor is the difference between a viable business and an unviable one.

Change the conversion rate to 6 percent and the picture transforms completely. This is why operators in this category spend disproportionate effort on the free-to-paid conversion step rather than on top-of-funnel volume. The sensitivity of the whole model to that one rate is higher than to anything else under management control.

Part seven: where the category is heading

Four structural trends follow from the economics above rather than from anything speculative.

Verification becomes universal. It reduces disputes, satisfies processors, lowers moderation load and improves the experience on both sides. Every incentive points the same direction, and the cost falls as verification services commoditise. Platforms that resist it will be pushed by their acquirers rather than by regulators.

The operators already combining verification with a coin economy show what the converged version looks like. Privé Moi is one documented example, running a coin balance alongside identity verification on both sides of a session, with an unpaid entry tier positioned as a free video chat platform worth trying before any coins are purchased. Structurally that is the freemium funnel from part two applied to a verified supply side: the free tier exists to produce the first session, the coin purchase is the conversion event, and the verification layer is what keeps the dispute ratio inside the thresholds described in part five. Each of those three elements is doing a different job, and the video chat business model only works when all three are present.

The supply side professionalises. Where the earning side is a genuine talent pool rather than incidental traffic, the platform's structure converges on agency mechanics: vetting, scheduling, rate setting, payment infrastructure, performance data and conduct standards. Platforms that treat hosts as suppliers to be managed and retained will outperform those treating them as inventory, because supply-side retention is the binding constraint.

Payment diversification. Given that processing is the category's hardest problem, operators are building redundancy: multiple acquirers, alternative payment methods by region, and deliberate reduction of dependence on any single provider. This is defensive engineering against an existential single point of failure.

Consolidation on compliance cost. Verification, monitoring and trust and safety carry large fixed components. Fixed costs favour scale. As requirements tighten, the minimum viable size of an operator rises, and small operators either grow, sell or exit. This is the standard consequence of rising compliance burden in any category and there is no reason to expect an exception here.

Key takeaways

For anyone modelling this category, the transferable conclusions are these.

Structural viability is not the same as survival. The free model in this category operated for well over a decade while being fundamentally unviable, sustained by founder subsidy. Duration is not evidence that a model works. Examine the direction of the cost and revenue curves independently before drawing conclusions from longevity.

Identify the dominant cost before optimising. In this case infrastructure is roughly 2 percent of revenue and the supply-side revenue share is roughly 40 percent. Analysts consistently focus on the former because it is technical and interesting. The margin is decided almost entirely by the latter.

Costs that scale with humans do not benefit from scale. Moderation cost per session stays flat as volume grows, while infrastructure cost per session falls. Any model that assumes total cost per unit declines with scale needs to separate those two categories before it means anything.

Constraints on access shape strategy more than constraints on capital. Where advertising policy restricts paid acquisition, organic search shifts from a growth channel to critical infrastructure, and the correct financial treatment of content investment moves from operating expense toward capital investment.

Payment infrastructure can be the binding constraint on a technology business. A dispute ratio threshold of 1.5 percent is not a compliance footnote. It is a hard operating ceiling that shapes product design, pricing, verification and refund policy across the entire company.

The general principle underneath all of these is that the interesting question about a platform is rarely what it does. It is which single number, if it moved by ten percent, would change the outcome most. In this category that number is the revenue split, followed closely by the free-to-paid conversion rate, and not at any point the cost of servers.