Eight B2C industries, one pattern: the top ARPU quintile is pulling away. What the leaders are doing differently under the hood in 2026.
Every quarter we pull ARPU trajectories across eight B2C categories we watch closely, from streaming and telecom to fintech, travel and marketplaces. For four years the picture was noisy but broadly symmetric: leaders and laggards moved together, plus or minus a cycle. In the last six quarters that stopped being true. In every single category we track, the spread between the top quintile and the median is widening, and the spread between the top quintile and the bottom quintile is widening faster.
This is not a story about who has the best product. It is a story about who has turned decisioning into a compounding asset, and who is still running the same campaign calendar they ran in 2022.
Take any B2C industry with a mature customer base of at least a few million users. Rank companies by trailing-twelve-month ARPU growth. The top quintile is not doing 20 or 30 percent better than the median; it is doing 2 to 5 times better on the metric that actually matters, which is incremental ARPU per active customer per quarter. The absolute levels differ by industry, the ratios do not.
Analysts have been calling this a “post-pandemic normalization.” That framing is wrong. Normalization would mean the fan closes. What we see is the fan opening, monotonically, quarter after quarter, across categories that share almost nothing else in common. When a pattern shows up in eight unrelated industries at once, the cause is not sector-specific. It is infrastructural.
For clarity: ARPU spread is the ratio of top-quintile ARPU growth to bottom-quintile ARPU growth over the same twenty-four month window, controlled for base size. We ignore firms below a scale threshold because ARPU math breaks below a certain denominator, and we ignore firms that are actively acquiring customers below cost because the growth ARPU signal is contaminated by dilution.
The result is a clean comparison: operators of roughly comparable scale, in the same industry, subject to the same macro conditions, with different internal operating systems. Whatever explains the spread must be inside the operator, not around it.
There is no single cause, but three forces compound. Each of them has existed for years. What is new in 2026 is that they now reinforce each other faster than incumbents can absorb.
Until 2024, “1:1 personalization” was a slide-deck concept. In practice, most enterprises ran a small number of large segments with a few hundred variants per year. In 2026, a handful of operators are running hundreds of thousands of decisions per day, per customer, with the model being retrained on outcomes weekly. That is not a marginal improvement over segments. It is a different regime.
The uplift is not primarily in click-through or open rate, which are vanity signals. The uplift is in the tail: a small number of very high-value customers get routed to a very different offer than they would have received under a segment strategy, and the expected value of that route change is disproportionately large. Miss that tail and you look identical to your competitors on every dashboard while quietly losing the customers who compound.
Cheap capital hid a lot of bad decisions between 2020 and 2023. In 2026, boards are auditing contribution margin, not revenue growth. That has a specific consequence: retention actions that were justified on gut feel are being killed, and only the ones with provable incremental margin survive. Operators without a causal measurement stack cannot pass this audit. They quietly stop running the actions that were quietly working, and their ARPU stalls.
The single most under-appreciated shift of the last eighteen months is that the cost of running a causal experiment, from hypothesis to shipped variant to holdout to lift readout, has fallen roughly an order of magnitude. That means the top of the distribution is now shipping ten times more experiments per quarter than the median. Ten times more experiments does not produce ten times the lift. In our data it produces roughly three to four times the lift, because most experiments still fail. But three to four times the compounding lift, quarter over quarter, is exactly the shape of the spread we are measuring.
The mechanism looks the same across categories. The surface area differs. A short tour of what we are seeing in each of the eight industries we track.
The spread here is the widest we measure. Top-quintile operators have moved from a linear tier ladder to continuous offer construction: bundle, ad-supported switch, cadence and renewal price are decided per customer, per moment. Median operators still ship one price change per quarter and one bundle change per year.
Postpaid ARPU expansion is being won by operators who caught the overage-and-add-on flywheel early: predicting which customers are about to bump their data ceiling and pre-offering a step up at a margin the CFO will sign. This looks like retention. It is actually expansion.
Cross-sell inside fintech was for years a mess of pop-ups. The leaders now treat each product surface as an eligibility engine: which customer, on which day, is best served which second product, and how does routing this offer today affect the probability of a third product in six months. That last clause is the one nobody was modeling in 2023.
The spread here is the noisiest, because demand is exogenous. But even after controlling for occupancy, top-quintile operators are extracting materially more per booked customer through ancillary sequencing: which upsell, on which channel, at which point in the pre-trip window. The median operator still emails everyone the same lounge pass twelve days out.
The interesting spread is not between marketplaces. It is inside each marketplace, between how the top decile of sellers are being matched to demand and how the middle is. Leaders have moved matching from a static ranker to a continuously personalized objective. Median operators are still tuning a single ranking model quarterly.
Live services have always known this game, and it shows. The leaders spent 2025 rebuilding their offer engines on a causal substrate; the followers are still running LiveOps calendars. The spread here compounds especially fast because monetization cycles are days, not months.
Paywall optimization used to be a quarterly A/B. In the top quintile it is now a per-article, per-reader decision, and the subscription funnel is a byproduct of a much richer engagement model. The tail matters even more here, because a small share of loyal readers accounts for most of the LTV.
Loyalty programs are the visible artifact; the invisible one is the underlying decision layer choosing who sees which reward, with what expiry, at what redemption price. The leaders have stopped thinking of loyalty as a program at all. It is a decision surface.
Once you look past the industry-specific surface, the mechanism is the same. Every operator faces a stream of moments where a customer could be routed to one of many actions. In the old regime, most of those moments were served by defaults, or by a small set of hand-authored rules. In the new regime, each one is a decision, and each decision is a bet with a measurable outcome.
The moment you industrialize that loop, three things start to compound at the same time. Data quality improves because every decision leaves a trace. Model quality improves because the traces become training data. Organizational velocity improves because every team is arguing over evidence rather than opinion. Each of the three feeds the other two. This is why the spread widens instead of closing: the underlying process is exponential, and exponential processes do not converge.
We interviewed heads of growth, analytics and product at seventeen operators across these categories. Five practices show up in almost every top-quintile answer and in almost none of the median ones.
For the median operator, the honest read is uncomfortable. The gap is not going to close by working harder inside the current operating model. Another quarter of campaign planning, another round of segment refreshes, another consulting deliverable on “customer 360” will not move the ARPU curve. The current stack is a local maximum, and it is the wrong hill.
The good news is that the shift is smaller than it looks. Most median operators already have the raw data. They have the product surfaces. They have the customer relationships. What they lack is a decisioning layer that turns opportunities into experiments and experiments into shipped, measured actions, continuously, at a cadence humans cannot maintain by hand.
Three signals will tell us whether this trend accelerates or stabilizes over the next four quarters. First, whether the median operator in each category adopts an opportunity-first operating model, or continues to run planning against a calendar. Second, whether boards start writing ARPU compounding, rather than headline growth, into executive compensation. Third, whether the leaders start to publish their causal readouts, which would compress the informational edge and move the frontier forward for everyone.
Our current base case is that the spread continues to widen through 2027, and then bifurcates: a handful of categories where followers manage to catch up because acquisition economics force the issue, and a majority where the top quintile locks in the advantage for the rest of the decade. Either way, the operators who spend 2026 building the decisioning layer will spend 2027 collecting compound interest on it.
A longer version of this analysis, with the full industry benchmarks and the methodology, is available in The 2026 ARPU Report for B2C Enterprises.
Frequently asked
See it in the product
The same loops this note describes run 24/7 against your customer base. Watch the workspace decide, experiment and execute 1:1.
