Usage Billing Review

Pricing Experiment Rollout Without Engineering Tickets

Companies can test new pricing tiers instantly when billing logic lives outside product code.

· 11 min read · Updated
Cover illustration for “Pricing Experiment Rollout Without Engineering Tickets”
Rating Engines and Pricing Rules · September 3, 2026 · 11 min read · 2,508 words

A product manager wants to test a new usage tier, so she writes a ticket. The ticket describes a discount rule, or a new credit rate, or a different way of counting what counts as a billable action. That ticket goes into a backlog, gets picked up in a sprint planning meeting, gets assigned to an engineer who has four other things due that week, and eventually produces a pull request, a code review, and a deploy. Weeks pass between the idea and the price actually changing for a customer. This happens because pricing logic lives inside the application code itself, so every hypothesis about a new tier, a discount rule, a credit rate, or a usage metric requires a code change before it can exist in the world.

The slow path from idea to live price is the predictable outcome of treating pricing as a feature of the product rather than as a layer that sits apart from the product's codebase. A feature needs a build cycle. A configuration does not. The companies stuck running pricing changes through sprint planning have, usually without deciding to, built pricing as the former when the business now needs it to behave like the latter.

This slowdown is visible in how rarely most companies actually test pricing. Only about a quarter of firms run regular pricing experiments, but Zylo's 2026 SaaS pricing strategy research found that the ones that do grow much faster than those that hold prices static. That gap doesn't come from a lack of ambition inside finance or product teams. It comes from friction: a pricing hypothesis that a team can debate and agree on in days can take weeks to actually ship once engineering has to own the deploy.

That friction is becoming more expensive with each passing quarter, because the rate of pricing change across the industry has sped up well past what a sprint-gated process can absorb. Kyle Poyar's 2026 State of B2B Monetization report covered 230 software and AI companies, and it found that roughly three in four vendors changed their pricing in the last year. A separate Poyar analysis found hundreds of pricing changes among top SaaS and AI companies with transparent pricing, just in 2025. A queue built for occasional, deliberate repricing cannot keep pace with a market where most vendors are repricing annually and the leaders are repricing continuously. The companies waiting on engineering tickets to test a new tier are competing against companies that no longer have to.

What is driving so many pricing changes at once

The move away from per-seat pricing is a direct consequence of AI agents breaking the link between the number of people on a team and the value a piece of software delivers, and that link is the entire economic logic per-seat pricing has always depended on. When AI agents let software do work that used to require more people, charging by the seat stops measuring anything real.

The mechanics of that breakdown hurt both sides of the relationship. When an AI agent absorbs work that used to require multiple human seats, a vendor charging per seat is effectively charging the customer for becoming less efficient, and if the customer resists by cutting seats, the vendor is punished for the very efficiency its product created. A Forbes tech council piece describes this as a seat apocalypse: teams shrinking from 50 people to 10 while the software consumption those 10 people generate keeps climbing, so the seat count and the value delivered are now moving in opposite directions. NxCode's 2026 pricing guide cites IDC forecasts that the majority of software vendors will abandon pure per-seat pricing by 2028, which places this shift on a timeline measured in a couple of years.

The dominant response so far is hybrid pricing: a base subscription stays in place to anchor a predictable monthly number for finance teams to budget around, while a metered layer on top captures whatever AI-driven consumption the base fee can't fairly account for. Poyar's 2026 report found hybrid pricing's share of the market jumped from roughly a quarter to more than a third of companies in a short period, a pace of adoption that matches the overall acceleration in repricing activity.

Credit-based pricing has become the bridge companies use to make this transition without tearing down what they already have. Figma, HubSpot, and Salesforce all introduced credit-based components, with Figma and HubSpot doing so in 2025 and Salesforce following in 2026. Credits let a company start monetizing AI features without dismantling an existing subscription structure, but they bring a new layer of metering complexity that the original billing stack, built for a flat monthly charge, was never designed to carry. Outcome-based pricing, charging directly for a resolved ticket or a completed task rather than for access or consumption, is arriving at the frontier but is still limited: NxCode's 2026 guide notes only about 9% of companies, roughly one in twenty, has fully implemented outcome-based models. The direction of travel across all of these models points the same way: toward pricing structures with more moving parts, more dimensions, and more frequent adjustment, not fewer.

Diagram: Pricing Model Adoption: Where the Industry Stands Today. Visualizes: Show three pricing model data points as a ranked or layered stat callout to illustrate the industry's current distribution and direction of travel.

What companies that repriced successfully did differently

The companies that have handled this transition well share one habit: they treat pricing as something to run experiments on continuously, a decision revisited constantly rather than once a year at renewal planning. The common thread across the clearest public examples is that each one changed its pricing mechanics multiple times in close succession, a pace that is only realistic when the billing layer isn't welded to the product codebase.

HubSpot's rollout of its Breeze credits offers the most fully documented example of this kind of iteration. It started with a March 2024 overhaul of its seat structure, then layered credits on top across several more adjustments, and each paid plan got a different amount of included credits every month. In April 2026, HubSpot changed its agent pricing again: the Breeze Customer Agent moved from a flat per-conversation rate to a lower rate charged per resolved conversation, effective April 14, 2026. That product was then folded into a new "Agent Hub" in a July 23, 2026 restructure, a move that also began phasing out the Breeze name for agents, while the Prospecting Agent shifted to a per-recommended-lead rate and was split into two separate versions in that same July relaunch. Through all of these changes, the base subscription on core seats never went away. The seat-based layer and the credit-based layer have coexisted the whole time. SaaStr's teardown at $3.5B ARR shows deals in HubSpot's larger tiers grew substantially year over year, driven by the credits SKU.

Lovable made the opposite bet instead, and moved faster because of it: it cut per-user charges out of its pricing. Its Free, Pro, and Business plans all allow unlimited users, and each one bills on credits consumed inside the workspace instead. Even that credit structure has kept moving: the Free plan went through a unification of its Cloud and AI balances in August 2026, followed by a September 2026 addition of temporary free chat credits valid through October 31, 2026.

monday.com took a different approach to the same problem: it ran a deliberate data-collection phase before it set any AI price at all. In early 2025, the company gave every paid tier, from Basic through Enterprise, an identical allowance of 500 free AI credits. That wasn't a revenue move. It was a way to learn what customers actually did with agents before monday.com committed to a meter or a rate. You can gather usage data first and design the pricing meter second only if the credit configuration sits apart from the core application, so it can be adjusted without engineering rebuilding anything.

GitHub Copilot ran a hybrid transition of its own on June 1, 2026, scrapping its "premium request units" in favor of AI Credits drawn against actual token consumption at published API rates. Seat prices for the Pro, Business, and Enterprise tiers didn't change the whole time. Business and Enterprise plans include credits equal to their seat price, while the individual plans, Pro, Pro+, and Max, include credits worth more than their price thanks to a flex allotment, and code completions remain free on every paid plan. The change touched the consumption metric, but it left the seat layer and the underlying product code untouched. New Relic stands as the clearest example of a full pivot to consumption pricing at scale: Aymane Boutbati's 2026 founders guide documents a large jump in annual revenue run-rate following that pivot, attributed to aligning what customers pay directly with the value they realize.

None of them had to rebuild the underlying product to do it.

The engineering cost that makes rapid iteration impossible when pricing lives in code

Usage-based and credit-based pricing create engineering problems that traditional subscription billing was never built to solve, and when those problems sit in application code instead of their own layer, they only get harder to manage.

The problem starts with event ingestion. Charging by the API call, the token, the resolved conversation, or the GPU-minute requires a pipeline that can capture usage events at millisecond speed, match each event against the correct pricing rule, and update a balance or accumulator in real time. A billing tool built to record a monthly invoice has no way to do any of this. It was designed to count seats and dates, leaving it unable to handle a stream of consumption events arriving continuously.

Credit drawdown looks simple from the outside and turns out to be one of the harder engineering problems in this entire stack. A working credit system needs real-time balance updates, atomic writes to prevent race conditions when concurrent API calls draw from the same wallet, graceful degradation when a balance hits zero, and edge-case handling for refunds, partial credits, and expiring balances. Teams that have built these systems from the ground up have spent weeks just on the initial implementation, then found that the edge cases they hadn't anticipated added significant additional time on top of that. Every new credit type or consumption rule a pricing team wants to test adds to that surface area, so each new experiment costs more than the last instead of costing the same.

Hybrid pricing multiplies this rather than simply adding to it. When a product bills across tokens, GPU-minutes, audio minutes, and a base seat fee on the same invoice, that isn't four separate billing problems running in parallel. The rating engine, the invoice renderer, and the revenue recognition logic all have to handle every possible combination of those dimensions at once. Companies that end up running separate, fragmented billing codepaths, one for subscriptions, one for usage, one for credits, routinely spend the first week of every month reconciling the three systems against each other, a finance cost that traces directly back to the same architectural decision to keep pricing logic scattered across the codebase.

No team has to buy a complete billing platform off the shelf. A middle path exists for teams that already run infrastructure like Kafka and ClickHouse for analytics and anomaly detection: keep ownership of the event pipeline, since it's already built, and buy the rating and invoicing layer that sits on top of it. For most teams without that infrastructure already in place, the economics favor buying the whole layer rather than building it.

Compliance adds a further layer of difficulty, and application code was never built to carry it. Under ASC 606, a hybrid contract that mixes a fixed platform fee with variable consumption charges requires a company to estimate the variable portion at the start of the contract and again at every reporting period, and that estimate has to be constrained to amounts where a significant revenue reversal isn't likely. A subscription billing tool that just records charges can't do this kind of revenue recognition, and a homegrown event pipeline built in-house can't either. The right-to-invoice practical expedient under ASC 606-10-55-18 is available for pure pay-as-you-go arrangements billed in arrears, but the moment a contract becomes hybrid, it leaves that safe harbor behind. Every new pricing dimension a company adds, whether it's a new credit type, a new usage meter, or a new hybrid bundle, adds another surface where this compliance work has to happen correctly.

What a configurable billing layer separates

If pricing experiments no longer need an engineering ticket, engineers don't disappear from billing because of it. It means moving the configuration of pricing out of the application codebase and into a system that product and finance teams can operate directly, without a deploy standing between a decision and its effect on a customer's invoice.

A configurable billing layer owns the pricing plan definitions themselves: the tier structures, the included quantities, the overage rates, the credit denominations, and which specific product actions draw down which meter. It owns entitlement rules, meaning what a customer on a given plan is allowed to access and at what rate, enforced inside the billing layer rather than hardcoded into the product. It owns segmentation and rollout controls, letting a team apply a new pricing model to one named cohort, whether that's new signups, one account tier, or a specific geographic region, without a code deploy. And it owns rollback: the ability to revert a failed experiment back to the prior configuration without an engineer touching the application at all.

What the layer does not own is the product itself. You still need engineering work to build a new AI capability, because pricing configuration has nothing to do with building features. The layer also doesn't own the event instrumentation that sends usage signals from the product to the billing system. Wiring up that instrumentation is engineering work, but you only do it once per usage metric, not once per pricing experiment. Once a team wires token consumption into the billing layer on day one, it can change what a token costs, how tokens bundle into a plan, or which plan a given rate applies to on any day after that, with no need to go back to engineering for any of it, that separation between instrumentation and configuration is the core premise behind Flexprice, a metered billing platform built for AI and SaaS companies that need to rate tokens, credits, API calls, and hybrid subscription-plus-consumption models inside a single engine.

The interface that makes this practical for non-engineers to use day to day is a no-code or low-code configuration tool. Product managers can create new plans, set up entitlements, and adjust credit rates through a visual interface, without writing SQL or asking for a deploy. Finance teams can configure invoice logic, apply discounts, and set custom contract terms for enterprise exceptions through that same interface, and they never have to route the request through an engineering team with its own backlog to manage. That is the actual shape of the separation: instrument the signal once, and then configure what it means and what it costs for as long as the product exists.

Sources

  1. SaaS Pricing Strategy and Models 2026: From Value-Based to Usage-Based Pricing

More in Rating Engines and Pricing Rules