The Portfolio Problem

There is an unclaimed category leadership position in the software market serving subscription businesses. It has been open for several years. Twelve credible vendors are within reach of it, several hold half of what it requires, and none has claimed it — because each is competing to answer the question its category inherited rather than the one its customers now ask.

Whether that reads as an opportunity or a threat depends on where you sit. For anyone holding equity in one of these assets it is arithmetic, because the same revenue prices two ways. As a capable vendor competing feature-by-feature inside an established category, it earns the category multiple. As the company that defined the category everyone else is now measured against, it earns something else entirely — and the difference is not a function of growth.

The question the market has started asking is short.

Which subscribers are we actually building a business on, and which ones are inflating our metrics while quietly degrading the economics underneath?

Every subscription business runs on numbers that are accurate and insufficient at the same time. Subscriber count. Net revenue retention. Churn rate. ARPU. These are not wrong — they measure precisely what they were designed to measure. But the leaders in this market have begun asking something the metrics were never built to answer, and the tools have not moved. What is true at Disney and Netflix today is where the rest of the market is heading, because the question is not a fashion. It is what happens to any subscription business once growth stops covering for composition.

That gap is the position. What follows is how the market arrived at it, which vendors are closest, and what it would take to claim it. It is worth saying at the outset that this test can come back negative: some categories have no opening, and a company losing on execution inside a healthy category should be fixed rather than repositioned. This one has an opening.

It is not a measurement problem. It is a question problem. And it has a history.

Part one — the executives said it first

The shift is not a theory about where the market is going. It is documented in the public statements of people running the largest subscription businesses in the world, in language that maps directly onto a question their own tooling cannot resolve.

Bob Iger, describing Disney's streaming strategy, said the company had been in a global arms race for subscribers and had become too aggressive on promotion — and that what Disney wanted was subscribers loyal enough to be insensitive to price increases. Read that carefully, because it is a confession about instrumentation as much as strategy. Disney had spent years running a promotion machine at enormous scale, and when its chief executive finally asked which subscribers were the durable ones, the answer was a wish rather than a number. The analytics told the company subscribers were growing. Nothing told it which subscribers it was growing.

David Zaslav put it more compactly when he was running Warner Bros. Discovery — since split into Warner Bros. and Discovery Global — arguing that driving down churn is what makes growth substantial. On its own that is the conventional position — retention beats acquisition, a thing every subscription business now says. What matters is the word doing the work underneath it. There is an entire product category built to reduce churn, and it is well funded and technically capable. But churn-reduction tools answer who is about to leave. They do not answer who is worth keeping — or, more expensively, who is worth acquiring more of. A save programme that retains a subscriber whose economics never justified acquiring them has hit its target and cost the business money. So has an acquisition budget that buys another quarter of the same cohort.

EchoStar's Gary Schanman was the most operationally specific of the group, describing a focus on engagement, loyalty and subscriber quality alongside churn improvement and ARPU growth, and framing the stated goal as onboarding the highest-quality subscribers. Highest quality. Per-subscriber profitability. These are not marketing phrases. They are operating objectives — a management team distinguishing between subscribers by their economics rather than counting them, and reporting it to investors in those terms.

And the most telling statement came from inside the industry rather than from a company that had been burned by it. The CEO of a subscription billing platform declared retention the new growth driver, said the pursuit of subscriber lifetime value was essential, and stated plainly that it is not about volume but quality — that spray and pray no longer works. When an infrastructure vendor tells its own market that the old question was wrong, the category is already in motion.

Four executives, four companies, one question. None of them described a system that answers it — each described an intention instead.

Part two — four generations, four blind spots

The question did not appear suddenly. It is the residue of four generations of the subscription economy asking progressively better questions. Each generation's tools answered the question they were built for — and answering it made the next question askable for the first time. The blind spots were not negligence. They were invisible until the problem in front of them was solved.

Generation one: reach (roughly 2000–2010)

The first decade of digital publishing was organised around one question: how do we move print audiences online without destroying advertising revenue? The answer was reach. Pageviews, uniques, impressions. The analytics platforms of the era measured aggregate traffic. Subscriber quality, where it was considered at all, was a print concept — household income and geography, not behaviour.

The blind spot: almost no connection between individual reader behaviour and subscription economics. Analytics were aggregate. Subscription experiments were rare, and where they existed they were all-or-nothing walls with no framework for asking which subscribers were worth more.

Generation two: the paywall rule (roughly 2010–2015)

The New York Times' metered paywall in 2011 was the watershed. It proved you could meter access and sustain an audience at the same time, and it became the template everywhere. The American Press Institute reviewed 98 US newspapers with circulation above 50,000 and found the number using a digital subscription model went from six in 2010 to seventy-seven by 2015.

The question this generation asked: how do we design a paywall that doesn't kill traffic but generates subscription revenue? The tools that emerged — entitlement platforms, A/B testing frameworks, conversion funnel analytics — were organised around exactly that. They optimised the rule.

The blind spot: the unit of analysis was the paywall configuration, not the subscriber. Operators could see which configurations produced more subscribers. Far fewer could see whether those subscribers were economically durable or fragile, because the tooling of the era was not organised to ask. Retention was tactical — win-back emails, promotional pricing — rather than a structured problem.

Generation three: engagement and churn (roughly 2015–2020)

By the mid-2010s subscriptions had moved from experiment to core revenue. The question shifted from how many subscribers we have to how we grow subscriptions and reduce churn by understanding engagement. Dynamic paywalls, propensity models, funnel and cohort analytics, engagement scores. Real progress: lifetime value entered the vocabulary, churn was measured by segment, and habit formation became a design goal.

The blind spot: two of them, and they compounded. Data stayed fragmented — analytics in one system, billing in another, marketing automation in a third — so connecting behaviour to offer to lifecycle to economics required a data warehouse project and an engineering team to maintain it. And the tools, even sophisticated ones, optimised metrics rather than identifying which specific cohorts were building or eroding value. Engagement and retention were established as strongly associated. Subscription businesses still treated their subscribers as a monolith when deciding where to spend.

Generation four: quality reaches the boardroom at the largest players (roughly 2020 to now)

The last few years made subscriber quality an executive concern at the largest subscription businesses. The question shifted again, and this time it is the right one: which subscribers are we actually building a business on? The statements in part one document the shift in named voices — and what those boards are asking now is a reliable leading indicator for the rest of the market, because the question is not a matter of sophistication. It arrives whenever growth stops covering for composition.

This is the first generation whose question has no tooling. Every previous blind spot was closed by the generation that followed it. This one is still open, and no vendor I can find has made subscriber portfolio quality its primary product narrative. The market still speaks generation three's language — understand your audience, grow recurring revenue, reduce churn, activate first-party data. The insight exists in pockets, at sophisticated operators with strong data science teams and the patience to build it themselves. It does not exist as a repeatable capability available to the subscription economy at large.

That is what an open position looks like: a question the market is asking out loud, and no product organised around answering it.

Part three — why the tools cannot answer it

A digitally native subscription business generates five distinct kinds of data about its subscribers. Each is valuable alone. None of them answers the question.

Pre-conversion behaviour — what someone read before they subscribed, how often they returned, which topics held them, how many times they hit the wall before converting. This lives in analytics.

Offer and acquisition context — full price or promotion, free trial or paid trial, which wall variant, which channel. This lives in the subscription and paywall layer.

Post-subscription engagement — whether they logged in or went dark, whether consumption widened or narrowed, whether they opened retention messages, whether they are showing the patterns that preceded cancellation in similar subscribers. This lives in journey orchestration.

Subscription lifecycle — renewals, upgrades, downgrades, pauses, save offers accepted, voluntary churn versus payment failure. This lives in billing.

Forward-looking propensity — likelihood to cancel, upgrade, or return, modelled on the four above. This lives in a data science environment and gets refreshed when someone has time.

Plenty of subscription businesses hold all five somewhere. That is not the problem. The problem is that in most stacks they are five records about the same person, held under different keys, in systems that were never built to agree on who that person is. Answering the portfolio question requires all five connected to a single identity in one model — because the question is inherently a join.

Consider what it actually takes to answer: of the subscribers acquired in a given quarter, through a given channel, on a promotional offer, who read primarily one content category in their first thirty days — what was their twelve-month retention, their average renewal price, their upgrade rate, and their current churn propensity?

That is not a hard question. It is an unanswerable one in a stack where each layer holds a different quarter of the record and none of them shares a key.

Part four — the four moments where the metrics look fine and the decision collapses

The strategic version of this problem is abstract. The operational version is not. There are four recurring moments where a leadership team has good numbers and still cannot make the decision in front of them.

The acquisition budget. Marketing proposes increasing acquisition spend by thirty percent. The CFO asks which channels produce subscribers who renew at full price versus those who need discounting to save. The data shows subscribers acquired by channel and conversion rates. It does not show which channels produce cohorts with pricing power. The decision gets made on instinct or deferred.

The price increase. ARPU is growing. NRR is above 100%. The board approves an increase. Within ninety days churn spikes in one cohort — subscribers acquired on a promotion two years earlier who never renewed at full price and whose engagement had been sliding for six months. The increase revealed a fragility the aggregate had concealed. Iger described the shape of this publicly: Disney wanted subscribers insensitive to price increases, which is a way of saying it had them and could not identify them in advance. A price increase resolves that ambiguity, and does it in the most expensive order — you learn who was fragile by losing them.

The content investment. A category is driving session volume and newsletter opens. Editorial wants to invest. The CFO asks whether the subscribers who engage most with it are the ones who renew, upgrade and tolerate price increases — or whether they are high-engagement, low-value readers who consume heavily and cancel at the first renewal friction. Engagement is measurable. Its connection to lifecycle economics is not, without the join. The investment proceeds on proxies.

The model question. The board asks whether to introduce a premium tier, move to metered, or hold. Each option attracts and retains different cohorts. The analysis shows overall NRR and average ARPU. It does not show what happens to cohort economics under each scenario for the segments that are currently strongest and most fragile.

The common thread matters more than any single case: the metrics are not wrong. They accurately measure what they were designed to measure. Boards have simply started asking questions the metrics were never designed to answer, and the gap between those two things is where the decisions collapse.

Part five — the receipts

None of this is a prediction. Composition repricing is an observed event that has happened in public, repeatedly, in the filings of companies that were growing at the time.

The pattern is consistent enough to be boring. Revenue rises. The growth story holds. And in the same filing — frequently in the same table — the per-customer figures are moving the other way: orders per customer falling, revenue per customer falling, cohorts thinning behind an expanding top line. Both truths are printed together, available to anyone reading past the headline.

Then the price resets. Not gradually, as the deterioration accumulates, but abruptly, at the moment the composition becomes legible to the market. Blue Apron is the cleanest instance. Revenue rose across five consecutive quarters while orders per customer and revenue per customer both fell — the growth and the decay printed in the same filings. The IPO range was cut from $15–17 to $10–11 before trading opened. By August the stock traded near $5, roughly half the price it listed at and about two-thirds below the range it originally filed.

The temptation is to read this as concealment. It is the opposite, and the opposite is worse.

Blue Apron disclosed the composition itself. Orders per customer and revenue per customer were in its own filing, in the same table as the revenue line. Every number required to reach the conclusion was published by the company that would be damaged by it. What was absent was not the data. It was the read — composition was never the thing that set the price internally, because the growth story was.

So the first time anyone treated composition as price-setting was during an IPO process. That is the worst moment it can happen. The number is being fixed, the window is open, the market has its own weather, and there is nothing left to change. A company that had understood its own composition two years earlier could have acted on it — shifted acquisition, repriced offers, fixed the cohort mix. A company that learns it in diligence can only watch.

That is the risk this article is actually about. Not that a subscription business has bad numbers. That someone else will do the composition read on it for the first time at the moment it is most expensive to be told.

Part six — where every vendor actually sits

The position is open, and the reason nobody occupies it is specific to each vendor. Named, so the claim can be checked.

Digital analytics — Google, Adobe, Amplitude. Sold to the analytics team, optimising pageviews, sessions and conversion. Amplitude is the most interesting of the three: it holds a deep behavioural graph and has begun reaching toward predicted value. What none of them is organised around is the subscription lifecycle. Revenue and subscription events can be piped in through integrations, but the native model is the session and the event, not the contract. Behaviour first, economics bolted on.

Subscription billing — Zuora, Chargebee, Recurly. Sold largely to finance and revenue operations, organised around MRR, ARR, churn rate and NRR. They hold the lifecycle and the financial buyer. What they do not natively own is the behaviour before the payment — the reading, the returning, the offers encountered — which is what would explain either. Zuora is, in my reading, the most credible challenger in this group: it popularised and trademarked the term subscription economy, owns the CFO relationship, and in its 2025 Subscription Economy Index introduced a Product Portfolio Balance Score measuring the balance between the size of a product catalogue and how frequently those products are actually sold and used. That is portfolio thinking, arriving from exactly the right relationship — and it is scoring the catalogue, not the subscriber base. Recurly's own CEO has said publicly that retention is the new growth driver and that it is not about volume but quality. The vocabulary has arrived. The instrument has not.

Customer success — Gainsight, ChurnZero. Sold to customer success teams and organised around health scores, lifecycle stages and at-risk accounts. Genuinely capable at what they do. They begin, though, after the customer is acquired — so the acquisition economics that determined whether the customer was worth having sit outside the frame.

The closest thing to a counterexample — Mather Economics. Mather positions publicly around subscriber yield management, renewal pricing and subscription benchmarking, which is nearer to this question than any platform on this list. The distinction I would draw is that Mather delivers this as expert services against a client's data, not as a product a subscription business operates itself, quarter after quarter, without an outside analyst. That distinction is real, but it is a distinction — not an absence.

Customer data platforms — Twilio Segment, BlueConic, Permutive, and until recently mParticle. The strongest behavioural half in the market, and the group whose distance to this position is most often overstated. Two of these are worth flagging on ownership before anything else: BlueConic has been majority-held by Vista Equity Partners since 2022, and mParticle was acquired outright by Rokt in January 2025 for $300 million — which means the most capable independent behavioural environment in the group now sits inside an ecommerce marketing business. That is itself an instance of the pattern in this article. The data half keeps getting bought by companies pointing it at a different buyer. Segment's Predictions ships four out-of-the-box models: likelihood to purchase, likelihood to churn, predicted lifetime value, and custom predictive goals — running natively on a unified, identity-resolved behavioural profile, at scale, inside a platform Twilio acquired in November 2020 in a deal valued at approximately $3.2 billion. That is the hardest component of the environment, shipped and generally available.

Read the documentation closely, though, and the gap is exact. Predicted lifetime value forecasts a customer's spend over the next 120 days, and is only computed for customers who have performed the purchase event two or more times. Likelihood to churn is defined by whether a customer will perform an action you specify inside a window you specify. Churn, to a CDP, is the absence of an event you defined — not a cancelled contract with its own economics, price history and save record. And the entire product is sold on the promise that marketers can build predictive audiences without tapping technical teams.

Excellent behavioural half. What is absent is native subscription semantics — no contract, no renewal, no price history, no save record — and the product is marketed to a buyer above the marketing team only by exception. Segment can certainly ingest subscription events. What it does not ship is a subscription-native economic model of a subscriber base.

The ones already holding both halves. A small number of companies have assembled the linked environment — behaviour before conversion joined to subscription economics after it, on a single identity — and describe themselves as something else entirely. Piano is the clearest example. Majority-held by the growth equity firm Updata Partners, it built its metered-paywall and subscription-access business through Press+ and Tinypass, acquired its behavioural and customer-data layer with Cxense in 2019, and added dedicated analytics through AT Internet, announced in 2021 and completed in 2022. Stitched onto one identity, that is the environment this category requires. It is positioned as an integrated customer intelligence and engagement platform, competing feature-by-feature in three categories it did not define.

And the entrants. MonetizationOS, founded by the team behind Zephr, launched in February 2026 as an edge-native infrastructure layer for governing and monetising human and machine traffic, partnered with Mather Economics, and opened with a free tier of a million operations a month. Building the plumbing, from the infrastructure side, without yet naming the business challenge that makes a CFO lean forward.

Read that list carefully and the shape of the open position resolves — but it depends on two separate requirements, and collapsing them is what makes the market hard to read.

The first requirement is the architecture, and it has two halves: a behavioural environment, and native subscription understanding, joined on one identity. Almost every vendor above holds one half and not the other. A small number hold meaningful parts of both.

The second requirement is the claim: naming the question the architecture answers, and aiming it at the executive who owns the decision rather than the practitioner who runs the tool.

Holding both halves of the architecture is not the same as occupying the position. It is the harder prerequisite, and it is worth far less than it looks while the company describes itself in the vocabulary of the generation it was founded in. That is the situation the market is actually in: the architecture exists somewhere, and the claim has not been made anywhere.

That also sets the clock. No single vendor is uniformly closest: the CDPs hold the strongest behavioural half, the billing platforms hold the lifecycle and the financial buyer, and only a small number hold meaningful parts of both. The software gap is subscription-native understanding plus executive-level positioning. The first may be an instrumentation decision. The second is a go-to-market decision. Neither necessarily requires a multi-year build. But closing the software gap does not recreate the accumulated history already sitting inside an established environment. That is a different clock.

What the category produces

Everything above describes a question and the cost of not answering it. This is what answering it looks like.

The instrument is not another dashboard. A dashboard can show more metrics without changing the unit of analysis. A portfolio question requires something different: connect subscriber history to price and cost, compare the groups management can actually act on, compare successive cohorts at the same age, and return every quarter to show whether the economics are improving or deteriorating.

The output is a Subscriber Portfolio Statement. It keeps observed results separate from forecasts and assumptions. It shows what current subscribers are expected to contribute, where that value comes from, whether newer cohorts are holding up, what deserves management attention, and what decisions will be tested. From the second Statement onward, it also walks the value of the portfolio from one quarter to the next and separates changes in the business from changes in the model.

That last property is the point. Reporting tells you what happened. Portfolio management creates a decision, measures what followed, and returns next quarter with the result.

Below are two moments from an illustrative publisher: its first Statement, and the same business after a year of use. All figures are invented.

Part seven — the category, named

Markets buy unnamed things all the time — custom projects, bundles, capabilities without a label. But a category does not form around an unnamed thing, and this argument is about a category. So here is the name.

Subscriber portfolio intelligence. The discipline of understanding how different subscriber groups create or erode economic value over time, and using that understanding to make better acquisition, retention, pricing and product decisions.

The word that matters is portfolio. A portfolio manager does not report the average performance of their holdings and call it a strategy. They know which positions are strengthening, which are weakening, what changed, and where the next dollar should go. Subscription businesses hold thousands of subscriber relationships and usually manage the averages.

It is worth being precise about what already carries this vocabulary and what does not. One major billing platform now publishes a Product Portfolio Balance Score, measuring the balance between the size of a product catalogue and how often those products are actually sold and used. That is portfolio thinking, arriving from a credible direction, and it is aimed at the catalogue. The subscriber base is a different portfolio. Plenty of vendors score individual customers — predicted lifetime value, health scores and churn propensity are all widely shipped. What is still missing as a primary product is the management view that connects subscriber economics, cohort change and decisions over time.

Part eight — the second question

Naming the real question is the first move. There is a second, and in an agentic market it decides what the answer is worth.

Once a company can see which subscriber groups are creating or eroding value, it has to decide what to build. And the choice is more consequential than it looks: is what you build a workflow, or an environment?

A workflow is a sequence of steps. Ingest the data, calculate the economics, compare cohorts, produce the view, recommend the action. The steps are largely legible from outside, and enough of them can be assembled by an agent stitching together other systems that the whole becomes contestable. Workflows are being compressed across every category right now, and the compression shows up in multiples before it shows up in product.

An environment is different. It is a place where something hard to replace accumulates — in this case, the linked record of which channels, offers and early behaviours actually predicted durable subscribers in one specific audience. A competitor can buy the company that holds it, but they cannot reconstruct it from outside on the timeline that matters, and a new entrant starts cold. It compounds every renewal cycle it runs. An agent doing portfolio work has to operate inside it, because the history that makes the judgement possible exists nowhere else.

The two moves are not independent. Build the environment around the inherited question and you get real switching costs wrapped around a ceiling — defensible, and capped. Name the right question but build only a workflow and the insight is correct while the company stays structurally replaceable. Both together is, I would argue, the only position that compounds without a visible ceiling.

This is the part the market has not begun to discuss. The category conversation, when it starts, will be about who answers the subscriber question. The valuation conversation will be about whether the answer is a workflow or an environment.

Part nine — so who does it

The obvious question is which of them claims it. The honest answer is that it will not be settled by who is best positioned, because several of them are close enough that positioning is no longer the constraint.

Zuora has the buyer, the brand, the language and the billing data, and would need a behavioural layer it does not natively own. It is also, since February 2025, privately held by Silver Lake and GIC in a $1.7 billion take-private — which removes the quarterly reporting constraint that makes a repositioning of this kind hard for a public company to attempt. Twilio Segment has the behavioural environment and the models already shipping. What it would need is native subscription semantics rather than configurable event definitions, and a buyer above the marketing team. mParticle held the same position and has effectively left the field: inside Rokt, its roadmap now answers to ecommerce marketing rather than to subscription economics. Amplitude has the graph and the least distance to travel on data, and the furthest on buyer. Piano holds more of the required architecture than anyone else on this list — both halves, in meaningful measure — and describes itself in the vocabulary of the generation it was founded in. What it would take to claim the position is simple to describe and hard to do, because it means telling a market you have been one thing that you are now another.

For most of them the missing software half is a product decision and a go-to-market decision rather than a foundational rebuild — which is not the same as saying it is easy. Identity resolution, contract semantics and a new sales motion are real work. That part may be closer to a roadmap than to a decade. The accumulated history is different: even after the architecture exists, a new entrant still starts cold.

That is where recombination enters. An incumbent missing one half has two paths. It can build the missing capability and wait for the history to accumulate, or it can acquire an environment in which that history already exists. The first closes the product gap. The second can close the time gap. Which is rational depends on how quickly the category is moving and how much of the advantage sits in accumulated, connected history rather than software capability. That is the M&A implication of the map; it does not imply any one pairing.

It will be whichever company has the risk tolerance to look honestly at the ceiling its own category is heading toward, and the appetite to move before the question becomes common knowledge rather than after it becomes safe. Those are not product properties. They are properties of a leadership team — and specifically of whether that team can tolerate being early in public.

My expectation is that most of the companies named above will read this, find something to agree with, and go back to optimising the metrics of the generation that made them. That is not a criticism. It is what the incentives reward right up until the moment they stop.

What follows

The subscription economy spent twenty-five years getting better at answering the question of its own generation. Each generation's tools worked. Each left a blind spot the next generation was built to fill. The chain ran forward reliably until it reached a blind spot with no successor — and that is where the market sits now.

The question is being asked out loud by the executives running large subscription businesses. Much of the data required to answer it already sits inside those companies, held by the vendors serving them. What I have not found is a platform organised around answering it as an ongoing management discipline — connecting subscriber economics, cohort change and the decisions that follow — or a vendor willing to say out loud that the metrics everyone optimises were built for a question that stopped being the important one.

The company that names the real question first is not competing for position in an existing category. It is defining the one that comes next — and if it also builds the environment rather than the workflow, it is defining one whose ceiling nobody has yet located.

If you hold one of these assets

The analysis above is not an opinion about subscriptions. It is the output of a method I run on categories.

Find the outcome the category has always promised to improve. Test whether that number has moved across generations of tooling — often it has not, and the flat line is the finding. Trace the chain of questions each generation was organised around, and what each one left untouched. Follow it forward until you reach a blind spot with no successor. Then apply two tests: whether a company is organised around the question its category inherited or the one underneath it, and whether what it builds is a workflow agents will replicate or an environment they have to operate inside.

The categories change. The chain does not.

For a board or an investor. If you hold an asset in this market — or in any category where a strong company is priced as a follower — the useful question is not whether it can grow faster. It is whether the category it competes in can be reset, whether this asset is the one positioned to reset it, and whether a missing half is better built, partnered for or acquired. That produces a written verdict: what the category's inherited question is, where the asset sits on both moves, what claiming the position would require, and what it would do to the multiple. It can come back negative, and a negative verdict is worth having before a process starts rather than during one.

For an adviser. The same map can expose a recombination thesis before a process starts: which adjacent assets hold the missing half, whether combining them closes only a product gap or also a time gap, and whether the result creates an environment rather than a larger bundle of features.

For an operating team. The same analysis, aimed at the roadmap and the narrative rather than the exit. Which question your product is actually organised around, and what the alternative is.

Either conversation starts the same way — tell me which category the asset competes in, and I will tell you whether the position is open.

mikebrady@g2gsaas.com