// The Architecture Read
The 2026 Read · Edition OneThe Software Pricing Architecture Read
The state of software pricing architecture, from observed behavior.
What software companies actually did, not what they said. Built from the pattern library SPP has kept across decades of engagements, and a curated ledger of dated vendor pricing moves. No surveys. No respondents. No benchmark panel.
01 · The headline read
The record for the year is plain: software pricing architecture moved far less than the market’s noise implied, and decided far more. Across decades of patterns in our library, fewer than one SKU in five with a multi-year price history was ever repriced at all: 18.6 percent. And where list prices moved, they moved as batch calendar events, about once a year at most; continuous repricing was the rare exception.
When a raise did ship, the record is unambiguous: it either passed fully into net prices, or the field discounted deeper against the new list and the company booked higher discount percentages instead of higher revenue. Yet the trend runs against the fatalist read. Raises land clean more often now than they did through the inflation years, as repricing became a normal event rather than an ambush. That is the read for 2026: the companies that treat pricing as architecture are pulling away from the companies that treat it as an event.
And deeper in the year’s variance record sits the finding that inverts the org chart: in where realized price finally lands, the rep who closed the deal barely registers. The architecture decides. Section 04 shows the anatomy.
02 · The repricing record
Most software was never repriced at all.
Across decades of patterns in our library, fewer than one SKU in five with a multi-year price history was ever repriced: 18.6 percent. Where list prices moved, they moved as batch calendar events, about once a year at most; continuous repricing was the rare exception.
The cost of standing still does not appear on the list price; it appears underneath it. Where nobody maintains the pricebook, per-customer discounts drift deeper year over year. We have watched 92 percent discount norms emerge from sales floors that started at 70 percent. Where the pricebook is maintained, the same drift does not appear; the discipline is available, it just is not automatic.
03 · What happens when list moves
A raise passes clean, or it dies.
On list increases the record is unambiguous: a raise either passes fully into net prices, or the field discounts deeper against the new list and the company books higher discount percentages instead of higher revenue. There is almost no middle ground. How large the clean-pass share is tracks how disciplined the vendor’s list is, so across the modern record it runs from about two in five to half, not a fixed rate.
The trend runs against the fatalist read: raises land clean more often now than they did through the inflation years, as repricing became a normal event rather than an ambush. Whether a raise holds is not decided at the deal desk; it is decided by the architecture underneath it, long before the raise ships.
04 · The two disciplines are one muscle
Freezing list did not stabilize pricing. It moved the chaos.
Companies that stopped maintaining list prices did not stabilize pricing; they moved the chaos into the discount column. Across decades of patterns in our library, companies running a static list spread comparable customers across roughly twice the range of net prices for the same product in the same period, while pricebooks re-rated on an annual cycle held the same spread near 1.4x.
Maintaining the pricebook and managing the discount column are not two disciplines. They are one muscle, and where it goes unused on one side, the variance surfaces on the other.
From the pattern library
The anatomy of price variance.
The industry’s default story assigns the trajectory of realized price to the people closing the deals. The variance decomposition tells a different story.
Which product, on which pricebook
nearly nine tenths of realized price variance
Which customer is buying
about a tenth
Which rep closed the deal
barely registers
The rep barely registers.
In a world that manages the discount column rep by rep, the deal-closer explains almost none of where realized price lands. An organization coaching its reps on price is managing the sliver and inheriting the nine tenths it never touched. Pricing outcomes are architecture outcomes.
05 · The hidden free-goods economy
A packaging decision, executed through the discount column.
A hidden free-goods economy runs through the discount column. In our pattern library, up to 63 percent of a company’s total list value shipped at a 100 percent discount, almost entirely as carry-along lines inside multi-SKU deals.
At that point the reported discount rate is substantially a packaging artifact, not a price concession. The company is making packaging decisions, deal by deal, through the one column nobody governs, and reading the result as a discounting problem.
06 · The negotiation fingerprint
Discount magnitudes are socially constructed.
In the most negotiation-driven records in our pattern library, up to 87 percent of discounts land on a multiple of five, against a baseline near 12 percent when discounts are computed rather than negotiated. A heavy round-number fingerprint means price realization is a negotiation ritual, not an architecture output.
This one is a self-service diagnostic. Look at your own deal record: if discounts bunch on round numbers, the number was talked into existence rather than computed. That is not a sales problem, and it is not fixed in the deal room. Reps negotiate on the terrain the architecture gives them; a round-number fingerprint diagnoses missing pricing architecture, nothing else.
07 · The renewal pipes
Renewal uplift runs through two different pipes.
Companies that maintain the pricebook raise net prices through it at renewal. Static-list companies raise net with list flat, through silent discount retraction. And a third group leaves renewal pricing on the table entirely, with a large share of renewals booking dead flat.
Silent discount retraction is not malpractice; it is unpriced, rep-negotiated drift that a maintained pricebook would formalize. The question for any leadership team is simply which pipe its renewal revenue runs through, and whether anyone chose it.
08 · The market-move ledger
The year in observed pricing moves.
Dated vendor pricing moves from SPP’s curated ledger. Every entry passed a citability gate before it reached this page: primary or corroborated sourcing, verified faithful to the vendor’s own statement, not superseded. The lines here are frame-grade; the full reads live on the Observatory.
Opus 5 launched at Opus 4.8’s exact rates, $5 per million input and $25 per million output, with benchmarks approaching Fable 5 at half Fable’s price.
The readA new flagship-class model at the predecessor’s exact price is a price hold doing strategic work: near-Fable capability now sits at half Fable’s rate, undercutting the flagship from inside the house. When capability compounds while the price line holds, the effective price of the tier is falling.
Breeze Agents billed through usage credits: $1 per lead recommended, $0.50 per resolved conversation, framed as pay when the task is complete.
The readA completed task is the vendor’s output, not the buyer’s business outcome. That is usage pricing on a per-task value metric, not outcome-based pricing.
Cost center support released for AI credit pools; budgets and usage caps attach to the cost center.
The readRather than change the pricing architecture, the vendor ships more tools for customers to manage their own spend. Expect the trend to hold until a breakout vendor absorbs the variability risk instead.
Agentforce Help Agent launched as a pre-packaged AI service agent with outcomes-based pricing.
The readThe label says outcomes; the meter bills workflow steps. A resolved call and a business outcome are two very different things, and only one of them is on the invoice.
09 · The architecture read
Every finding above is one architecture, seen from three decisions.
The repricing record, the discount column, the free-goods economy, the renewal pipes: none of these are separate problems. They are the visible behavior of three decisions every software company makes, deliberately or by default.
Licensing
Who may use the software, on what grant, measured by what unit. The value metric lives here, and the renewal conversation inherits whatever the grant left unsaid.
Packaging
What ships together and what ships separately. The hidden free-goods economy is packaging decided informally, deal by deal, through the discount column.
Pricing
What the units cost, on what pricebook, maintained at what cadence. The repricing record and the dispersion regimes are this decision’s fingerprint.
Phase 1Define
Make the three decisions deliberately, as one architecture, before the market makes them for you through the discount column.
Phase 2Deploy
Move the pricebook, the field, and the paper to the architecture. The transmission record shows a raise holds or dies on what sits underneath it.
Phase 3Defend
Maintain the pricebook on a cadence. A static list runs roughly twice the net-price spread of a pricebook re-rated annually; that gap is the observed difference between companies that defend the pricebook and companies that abandon it.
The direction of the market is visible in the record above. What any single company should do with it depends on its own licensing, packaging, and pricing, which is where the work leaves this page.
10 · Provenance and method note
Where this comes from.
SPP’s pattern library is the record we have kept for decades of how pricing decisions behave once they reach a deal desk, drawn from patterns across more than $480 billion in software transactions priced, negotiated, or repriced. It was built while software pricing barely existed as a discipline, decades before the category’s other advisors arrived, and it cannot be assembled after the fact: early observations cannot be back-dated, which is why no comparable record exists.
The record spans the breadth of B2B software along every axis: infrastructure to vertical applications, self-serve to enterprise-negotiated, seat-licensed to consumption-metered, founder-led startups to public companies. Patterns published here draw on that breadth, not one corner of the market.
It also runs through every era of commercial AI, beginning with pricing work for a Kurzweil company, in the direct lineage of one of the first commercial AI products ever sold, through the predictive-analytics wave, to the generative and agentic era this Read covers. Pricing AI was not a pivot for this record; the library has carried AI product pricing since the technology first had a price.
Every read on this page is written and signed by a practitioner, not generated by a dashboard. Chris Mele, SPP’s CEO and #1 on OpenView’s list of B2B SaaS pricing experts, leads the Architecture Read the way SPP has run pricing architecture work for decades: licensing and packaging decided before pricing carries the number.
The discipline, then and now
The method this Read runs on is not a recent invention. The earliest engagement paper in the record already shows it, element for element:
2002
This Read
Structure read before any price level moved: complexity, packaging fit, and channel needs first.
The trifecta discipline: licensing and packaging decided before pricing carries the number.
Maintenance terms treated as part of the architecture, never an afterthought.
Renewal and maintenance economics scored as architecture in every read.
Repricing implemented to preserve the reseller channel that sold the product.
Legacy conversion as doctrine: new logos first, the installed base on a modeled schedule.
Volume pricing designed so negotiated prices land near list; concessions small by construction.
The scheduled net price: discounts live in the architecture, not in the rep’s discretion.
Priced on economic tradeoffs and market evidence, never opinion.
Published at the precision the record supports, and independently verified before it prints.
What this is
- The year’s read: judgment applied to what the record shows, signed by the firm that keeps it.
- Built on the pattern library: a record kept across decades of how pricing decisions behave once they reach a deal desk, published at the precision the record supports.
- And on the year’s move ledger: dated vendor pricing events, each verified faithful to the vendor’s own statement before it enters, kept live in the AI Pricing Observatory and distilled here once a year.
What this is not
- Not a survey. Nobody was asked what they intend, believe, or self-report. The record is what companies did.
- Not a benchmark panel. Findings describe observed patterns and ceilings, never a number to price against.
- Not a dataset. Nothing on this page is downloadable, queryable, or shareable, and no client is identifiable in it.
How the record is read
- Variables are separated before they are credited. A finding about products is computed with the product held fixed; a finding about reps is computed with product and customer held fixed first. Nothing is attributed to a factor that was never isolated.
- Comparisons are matched: same product, same period, same context. Where a comparison cannot be matched, it is not made.
- Where attribution cannot be fully separated, we publish the ceiling, never the point.
- Where the record cannot support a finding, it sits out of that finding. No basis is ever widened to strengthen a number.
- A pattern is tested across eras before it is called durable.
- Dated market events are independently verified against the vendor’s own statement before they enter the record, and re-verified on a standing cadence after.
11 · Questions
Frequently asked questions
12 · What continues from here
The Read is the snapshot. The Observatory is the pulse.
The market-move ledger you saw above is a glimpse of something larger: the AI Pricing Observatory, the living continuation of this page, now live. Dated vendor moves, each passed through the citability gate, each carrying a full read on the mechanic underneath the vendor’s label, refreshed weekly rather than annually.
And when the question stops being what the market did and becomes what your company should do, that is a working conversation, not a page.