From Coarse Volume Tiers to a Smooth Pricing Surface: The Artifact AI Consumption Forces

TL;DR: In B2B software pricing, a pricing surface is a multi-input control surface that produces a net price for any commitment a customer might make, across volume, product, Customer Group, and channel. Coarse tiered pricing schedules are stepped surfaces with almost no fidelity, built for the seat era. At AI consumption volumes that coarseness leaks margin three ways: buyers argue the gaps between pricing tiers, fractions of a percent now move real money, and the flat-rate catch-all at the top cannot tell 10 million units from a billion while variable cost scales with every one of them. The fix is not a new label on the price. It is a better artifact: a smooth pricing surface with the cost slope engineered into the net price at every point.

Vendors reached for “outcome-based pricing” to calm buyers rattled by consumption pricing, and buyers looked under the hood anyway. Once a procurement team audits what the meter actually counts, the label stops doing any work and the real conversation begins: what does the next unit of commitment cost, and why. Most vendors cannot answer that question well, because the artifact they answer it with is a coarse tiered schedule, a pricing technology that predates the software industry and was never designed for the volumes AI products now produce.

This article makes the constructive argument our piece on whose outcome the label bills for tees up: the move that actually resolves consumption-pricing chaos is not better vocabulary. It is the move from coarse volume tiers to a smooth pricing surface.

What a pricing surface is

A pricing surface is a multi-input control surface that produces a net price for any commitment a customer might make, across volume, product, Customer Group, channel, and the other inputs that legitimately move price. Give it any commitment a deal could take, and it returns the net price your company intends for that commitment. Many surfaces are possible for the same product; the point is that one exists, deliberately constructed, rather than a stack of tables and exceptions pretending to be one.

The word choice matters. A curve plots one input against one output, volume against price, and it connotes something observed, like a demand curve you read from data. The pricing surface is the opposite of observed. It is engineered: you construct it, deploy it, and tune it as your understanding of willingness-to-pay improves. It balances volume, cost profile, Customer Group value-extraction rates, and product-portfolio interactions in one artifact. And it is smooth: net price changes continuously with commitment, with no cliff where the arithmetic jumps.

That distinguishes it from the two artifacts it is confused with. It is not a discount table, which is a stepped surface with a handful of hand-set entries. And it is not dynamic pricing, which varies price per transaction based on live market conditions. A pricing surface is deterministic and contract-pinned: the same commitment produces the same scheduled net price for any buyer on any day, and the surface changes only through deliberate central revision. That determinism is also what makes a surface safe to expose to the agent channel: a buyer-side agent can only act on a price that computes the same way every time it asks. The surface is one artifact inside a larger pricing architecture; the architecture decides what you meter and how you package, and the surface prices whatever commitment those decisions produce.

Where the coarse tiered schedule came from

The stepped schedule is old, and both of its lineages produced artifacts with almost no fidelity.

The professional-services lineage ran on the Lehman formula: a declining percentage taken on transaction value, stepped down in a few coarse bands. Every success-fee arrangement I priced early in my career reduced to that shape, a schedule with two or three entries standing in for an entire pricing model. It was a blunt percentage-of-value instrument. Nothing about it could vary price across the dimensions that actually drive value, because there were not enough entries in the artifact to express the variation. The companion piece covers where that era’s outcome fees failed on attribution; what matters here is where they failed on construction. There was no surface, only steps.

That artifact is not a museum piece. In one engagement, the client charged a declining percentage of their customers’ spend, a direct descendant of that Lehman-style schedule, and their own technology was optimizing down the very spend they took a percentage of. Moving them to a smooth pricing surface recaptured more than $2.2 million of upside in the first six months of the rollout. The gain did not come from customers changing how they committed; they spent as they always had. It came from the smoothing itself: discounts shrank, and the reduction was recaptured as margin. The client then embedded the surface, managed in LevelSetter, inside their own software so customers could plan future budgets against the same net-price function the sales team quotes from.

The manufacturing lineage gave software the volume-tranche discount table, price bands stepped by quantity, built for factories whose unit economics genuinely changed at production breakpoints. Software inherited the shape without the rationale, and margin-calibrated discounting covers that inheritance and the practice that replaces it in depth. The short version: the steps in a software discount table almost never correspond to anything real about the cost of serving the customer.

Two lineages, one property in common: a handful of table rows carrying the entire relationship between commitment and price.

Why volume tiers were good enough in the seat era

Coarse schedules survived four decades of software because seat-based pricing kept their weaknesses cheap.

Headcount bounds volume. A 5,000-employee company cannot buy 400,000 seats, so the schedule only ever had to price a narrow, predictable range of commitments, and the catch-all band at the top rarely saw a real deal. The marginal cost of an extra seat was near zero, so no volume of usage could push an account below its margin floor.

The one visible weakness was the buyer who landed awkwardly between price breaks, and the industry priced that weakness in as tolerance: in our experience the gap between what a coarse schedule charged and what a well-tuned price would have charged typically ran on the order of fourteen percent, and both sides learned to shrug it off.

I never made peace with that shrug. An executive at a software company, formerly of private equity, once told me fourteen percent was close enough. I reminded him that at $44 million, fourteen percent is over $6 million. It always amazed me that close enough in software was measured in millions. I started my own software company in the late nineties and scrambled for every last penny of profit; close enough was never an error rate I could afford.

The flat band even had academic support: peer-reviewed work on quantity discounts found that all-unit structures, one rate inside each band, typically outperform incremental schedules when marginal costs are constant. Constant marginal cost is the seat era in one phrase. Software met the condition for four decades, and consumption pricing removed it.

Under seat-era conditions, a stepped schedule with six rows still passed for a serviceable artifact. Every one of those conditions, the bounded volume, the near-zero marginal cost, the survivable gap, is now false for consumption-priced software, and the next three sections take the failures one at a time.

Leg 1: coarse volume tiers invite the between-tier argument

Put a stepped schedule in front of a competent buyer and watch what the steps teach them to do.

A buyer whose expected volume lands between two price breaks does not accept the arithmetic. They argue the special scenario: we are almost at the next discount tier, our growth will carry us there by Q3, surely you can extend that rate now. The vendor caves or splits the difference, and the net price that results is out of tune with the buyer, produced by negotiating leverage rather than by anything the pricing was designed to express. The mechanism is the same one that breaks financial-outcome metrics: hand the buyer a price they can legitimately dispute and they will dispute it, every cycle. Designing outcome-based pricing treats that dispute as something you engineer around in the metric; the coarse schedule manufactures the same dispute in the discount structure, deal after deal.

The steps also distort what buyers commit to. The schedule hands a buyer approaching a price break from below an economically rational move: stop at the break, buying less than the workload actually needs, because the next unit past the boundary is priced against a cliff. We call the pattern commitment-hedging, and a tiered schedule aggravates it structurally: the buyer under-provisions, their time-to-value stretches because they bought less than the workload required, and your average selling price collapses toward the breakpoints.

Crossing the boundary does not end the distortion; it changes shape. The units past the break come at the discounted rate, so the buyer argues that reaching the break earned that rate for every unit beneath it too, collapsing what was built as a tax-bracket calculation into one flat, in-that-tier discounted rate. Or the buyer runs the arithmetic the other way: the per-unit price offers no incentive to push that far into the next tier, so they lop those units off the commitment to cap their financial risk. Either way, the schedule is steering the commitment instead of the use case. The job of the pricing surface is the opposite: make it rational for the buyer to come all in with the volume of units their use case actually needs.

The schedule is nudging customers to buy less and argue more, and that is not speculation. We ran an internal study of thousands of tiered schedules crossing thirty years: not one was underpinned by solid mathematical structure. Every one was hand-created, with oddly bizarre unit economics, and every one was changing buying behavior in ways the seller never chose. Somebody designed every one of these schedules. Nobody designed them with any science, which is the point. And the science was sitting there the whole time: the operations-research literature has treated discount schedules as mathematically optimizable objects for decades. If thirty years of schedules show one durable pattern, it is this: software companies love manufacturing pretty-looking tier schedules in Excel, and looking pretty and maximizing revenue turn out to be two very different things.

Leg 2: at consumption scale, fractions of a percent are real money

The seat-era tolerance had arithmetic behind it: when the coarse schedule mispriced a deal by fourteen percent and the deal was bounded by headcount, the absolute dollars were survivable. Consumption pricing removes the bound. Metered volumes on AI products run to hundreds of millions of units on a single account, and they grow between renewals without a salesperson touching the deal.

At those volumes the arithmetic inverts. A tenth of a percent of net price on a large consumption account is more absolute margin than entire seat-era deals carried. Where a rate sits between price breaks, how fast discount accumulates from one commitment level to the next, where the accumulation flattens: at scale, each of those decisions moves real money, and a schedule with six rows has made almost none of them. The step function does not have enough entries to be wrong precisely. It is wrong in bulk.

There is an old fraud story that makes the scale point better than any schedule can. A programmer rounded the fractions of a penny off every transaction into a separate bank account, confident nobody would notice amounts that small. Nobody did. What he had not planned for was the volume: the fractions piled up so fast that he had to keep withdrawing money to keep the account from drawing attention, and the withdrawals were what drew notice. Fractions of a penny were never the risk. Fractions of a penny at volume were. A coarse schedule runs the same scheme against you: amounts too small to argue about on any single unit, accumulating at consumption volume into a number the income statement cannot miss.

That changes what pricing operations is. In the seat era, tuning the discount structure was bookkeeping, and it was rational to neglect it. At consumption scale, the surface is a P&L instrument: tuning its slope against demand response is one of the highest-return activities available to the company, and coarseness is not a simplification anymore. It is a standing leak.

Leg 3: the catch-all top tier goes margin-negative

The sharpest failure sits at the top of the schedule, in the row every tiered artifact ends with: the catch-all. “10M+ units” at one flat rate.

That row carries two assumptions software quietly inherited from manufacturing: a ceiling, and stable costs. A physical plant has finite capacity, and its cost per unit is planned in advance, computed through an MRP run against the bill of materials before the line ever starts. Against a fixed ceiling and a known, stable cost, a fixed tiered tariff made economic sense: run the line toward maximum capacity and spread fixed costs across as many units as possible. The catch-all top tier was the ceiling made explicit, one flat rate for volume at or near the limit.

Software holds neither assumption at consumption scale. It has other limits, funding, headcount, the size of the market, but no physical plant capacity, and it produces each additional unit of usage virtually, at will. Its costs are not stable either: compute and inference costs scale with every unit consumed and keep moving as models and workflows change. The artifact kept assuming a fixed ceiling and a stable cost long after software erased both, and that is why the catch-all now works against you as consumption volumes climb.

Run the arithmetic on a generic schedule. The catch-all rate was set the way those rates are always set, by extending the discount pattern one more step and rounding to something sales could quote. It prices the boundary case adequately: a customer at 10 or 12 million units, the volume the row’s author had in mind. Now land a real AI-era account inside it at 400 million units. The schedule bills every one of those units at the same flat rate, because a single row cannot distinguish 10 million from 100 million from a billion. The rate stopped being calibrated to anything the moment usage passed the last engineered breakpoint; past that line, the schedule is not pricing, it is extrapolating.

Meanwhile the cost side keeps moving. Consumption-priced products, AI products above all, carry real variable cost on every marginal unit: inference, compute, storage, the cost slope that has to be architected rather than assumed away. Seat software never had this problem twice over, once because headcount bounded the volume and once because the marginal seat cost nothing to serve. When the flat catch-all rate sits close to unit cost, and deep-in-the-tail usage shifts the workload mix against you, contribution per unit thins toward zero and crosses it. The loss then scales with exactly the number everyone is celebrating. The account that looks biggest on the revenue dashboard is the one compounding a negative margin, and the schedule cannot see it, because the schedule’s last deliberate decision happened at 10 million units.

This failure is specific to usage-based and outcome-based companies, because theirs are the value metrics that actually reach those quantities. It is not hypothetical and it is not rare; it is what a coarse schedule does by construction the first time real volume blows past its top row.

Relabeling the price does not fix the artifact

Set the three failures next to the market’s most popular response to consumption-pricing anxiety, which has been to rename it.

The outcome label spread because it soothes: buyers frightened by unpredictable metered bills are told they now pay for results. The companion piece walks through what AI vendors are calling outcomes and what the meters underneath actually count, and it ends where this article begins: a smart buyer audits the unit, the relabel comes off under inspection, and the vendor is back in the same negotiation with less trust than before.

Notice what the relabel never touched. The between-tier argument is still there, because the schedule still has steps. The uncalibrated catch-all is still there, because the schedule still ends in a flat row. Renaming the unit changed the noun on the invoice and left every structural property of the artifact intact. Buyers were never objecting to the noun. They were objecting to a price they could not predict and could too easily dispute, and both of those are properties of the artifact’s construction. The fix is fidelity, not vocabulary.

What a smooth pricing surface changes

Replace the stepped schedule with a smooth surface and each failure loses its mechanism.

The special-scenario argument dies because there is no gap in the schedule to argue from. Every commitment level sits on the surface and receives a deliberate net price, so there is no dead zone between price breaks for a negotiation to live in and no cliff for a rep to round-trip a deal through. Commitment-hedging loses its economic nudge the same way: with no break to stop at, the buyer sizes the commitment to the workload, not to the discount structure’s geometry.

The catch-all failure dies because the surface has no catch-all. The cost slope is engineered into the net-price function at every point, including the extreme tail, so a 400-million-unit commitment is priced with the same intent as a 10-thousand-unit one and no volume anywhere on the surface crosses below the margin floor. The schedule stops extrapolating because there is nothing left that extrapolates; every point is a decision.

And the company gains something the stepped artifact could never provide: one central object to tune. When transaction data shows the slope is too generous between two commitment levels, the surface is revised once and every quote, every channel, and every territory prices off the revision immediately. Margin-Calibrated Discounting is our name for the practice that engineers and operates the surface this way, with sales compensation tied to where deals land on it; the practice has its own article, linked above, and this one stays at the level of the artifact. At portfolio scale, keeping dozens of surfaces tuned across products and channels is infrastructure work, and it is what our experts use LevelSetter for; the design judgment about where the slope should sit stays a human decision.

Stable architecture, tuned surface

The market has produced two reflexes to pricing volatility in the AI era, and the surface answers the anxiety underneath both.

The first reflex says pricing will now change constantly, so buy infrastructure that lets you swap pricing models quickly. That is the billing-platform pitch, and it mistakes motion for control. A company that churns its pricing model every two quarters is not iterating; it is broadcasting that it never committed to an architecture, and each swap resets whatever the market had learned about how to buy the product. The second reflex is the relabel, covered above: keep the mechanics, change the story.

Our answer is neither. Keep the architecture stable and tune the surface. Commit to one right value metric, the unit that tracks the value your customers actually receive, and let it stand long enough for buyers, reps, and your own transaction data to accumulate meaning around it. Then do the continuous work where it belongs: on the surface, adjusting slope and level as evidence arrives, without ripping out the model underneath. When we see a company running several pricing models simultaneously, we usually find an architecture that never committed, with the model menu standing in for the metric decision nobody made. Stability of architecture and fluidity of price are not in tension; the surface is precisely the artifact that lets you have both, which is why it is the enabling artifact of continuous monetization. Until a software company has a pricing surface, list-price adjustments are largely fictional, because chaotic discounting absorbs them before they reach a net price.

How to move from coarse volume tiers to a pricing surface

The migration is seller-side design work, and its shape follows from the failures above.

1. Map the cost slope, not the price list, first. For each metered SKU, the company has to know how variable cost actually behaves across the full commitment range, including volumes ten times anything yet sold, because that slope is what the surface will carry.
2. Set the margin floor at the extreme tail. The catch-all row is where the old artifact was lying most, so the tail is where calibration begins, working backward toward the volumes where the schedule was merely imprecise.
3. Engineer the accumulation slope between anchor commitments. How fast net price improves as commitment deepens is the surface’s central lever, and it is engineered against margin targets and demand response rather than by extending last year’s discount pattern one more row. The demand side of that slope is measurable, not guesswork: peer-reviewed subscription-pricing research found willingness to pay decays rapidly with quantity. The slope should follow the decay your own transaction data shows.
4. Tie sales compensation to where deals land on the surface. A rep defending the surface is then defending their own number; that alignment, and the operating discipline around it, is the substance of the margin-calibrated discounting practice.

One caution about method. The popularized shortcut is to run a regression through historical deals in Excel and call the fitted curve a pricing surface. Regression teases out correlation: it describes the discounts your reps happened to give, including every leak the study above catalogs. That is definitively not a pricing surface. The surfaces we build are structurally optimized from a pattern library of how deals actually land across SMB, mid-market, and enterprise and across verticals, specialized to willingness-to-pay patterns proven in transaction data, then calibrated to your cost slope and margin targets. The optimization itself is programmatic: millions of candidate surfaces evaluated against the stated margin and revenue targets, converging on one best answer rather than a defensible-looking curve. Our experts run that optimization in LevelSetter; the judgment about which outcome to optimize for stays human.

The design space is also wide. Many surfaces satisfy the same margin targets, and they are not interchangeable: some of the mathematically cleanest candidates deliver abnormally large price increases to legacy customers, which converts a margin repair into a churn event. Legacy transitions are therefore a designed input to the optimization, not an afterthought: the surface search includes how to structure legacy-customer transitions so existing accounts move forward without breaking.

What we typically find when this work starts is the pattern that study of thousands of schedules shows: thresholds fabricated deal by deal, discounts leaking everywhere. Putting the pricing surface in place on today’s packaging and licensing, before touching anything else, drives millions of upside on its own. Addressing packaging and licensing behind it is where growth compounds, provided the product-market fit is real. Across our engagement work, pricing remains the single biggest friction point in the software business model.

From there the surface lives as an operated artifact, tuned on transaction evidence rather than renegotiated in a renewal-cycle crisis. The demand-response patterns we tune against come from the $481B+ in B2B software transactions we have analyzed since 1982, and one pattern recurs throughout that data: the gap between what a coarse schedule charged and what the buyer would have committed to under a deliberate price widens with volume. At AI consumption scale, that gap is the difference between your biggest account being your best one or your most expensive one.

If your schedule ends in a flat catch-all row and your metered volumes are climbing toward it, talk to an expert: describe your value metric, your volume schedule, and where usage is heading, and a pricing expert will reply with a read on where the surface work should start. Our approach describes how we build the architecture the surface sits inside.

FAQs