A Performis white paper — September 2026
Executive Summary
Enterprise software has built its entire economics around one mechanism: charge for who is allowed to log in, not for what actually gets done once they do. A named seat costs the same whether the person behind it works eight hours a day or logged in once, months ago, and never came back. Independent research puts the resulting waste at tens of millions of dollars a year at the average large organization. And that's before the customization and integration fees that stack on top of the license every year after. Syspro's 2026 decision to end perpetual licensing for roughly 15,000 manufacturing customers is the clearest recent example, not an isolated one. SAP, Oracle, Microsoft, and Sage have each run a version of the same play in the last three years: a licensing-metric change here, an indirect-access claim there, a forced subscription migration, a new fee stacked on an old one. Different tactics. Same mechanism underneath.
Performis starts from a different premise: agents should execute the operational work directly, not sell a better rate on the same relationship. Profiling a system's data. Proposing how it maps onto a new one. Extracting the business rules buried in old scripts nobody remembers writing. Generating the code that moves the data, then running the reconciliation checks that prove it moved correctly. A person gets pulled in only for the decisions that are genuinely decisions: confirming a mapping an agent can't be certain about, signing off before a system goes live, approving something material enough that it needs a name attached to it. This isn't a roadmap slide. It's running today, across three separate engagements in three separate industries (legal, healthcare, manufacturing), each with agents and humans doing exactly this division of labor.
The rest of the software industry is only now arriving at the same conclusion: charging for outcomes beats charging for access. Salesforce, Intercom, and Sierra have each walked away from per-seat pricing on their AI products in the last eighteen months. Analysts at a16z, IDC, and Bain are independently describing the same shift. Performis got here first, and the reason is simple. Outcome-based pricing only makes sense once a real outcome is being produced by something other than the customer's own labor. A support ticket an AI agent actually resolves is a real, meterable event. A dashboard the customer still has to open, read, and act on themselves is not, no matter what the pricing page calls it.
Part 1: The tax nobody named until now
A tech tax is what a business pays when a piece of software's cost scales with how many people are permitted to access it, regardless of how much value that access produces on any given day. It's disconnected from outcomes by design. That's exactly why Syspro's installed base makes such a clean case study of the pattern. Performis has already published a full breakdown of what Syspro's 2026 licensing notice actually says and what it costs a typical shift-based manufacturer (see Your SYSPRO Renewal Isn't a Price Increase — It's a Tech Tax), along with a free Syspro Renewal Exposure Calculator that estimates a plant's real named-user exposure from its concurrent licenses and shift pattern in under a minute. What follows is the broader pattern Syspro's notice happens to make visible for one industry, in one quarter. It isn't unique to one ERP vendor.
The tax shows up in two layers. Both are real, sourced, and bigger than most finance teams assume the first time they hear the argument.
Layer one: seats you're paying for and not using
This isn't a new problem, and it didn't start with Syspro. Research firms whose business depends on getting the number right, not on flattering a vendor, have measured it independently and repeatedly.
Zylo's SaaS Management Index is built from real telemetry across more than 40 million licenses and $40 billion-plus in tracked spend. It found the average organization wastes $19.8–21 million a year on SaaS licenses nobody is using, running at only about 54% utilization of what's actually been purchased, across the 2025 and 2026 editions of the report. That waste scales with company size: roughly $3.8 million a year at organizations under 500 employees, more than $80 million a year at the largest enterprises. Zylo's own per-employee figure, north of $4,800 in annual SaaS spend per head, makes clear this is a structural line item, not a rounding error buried in an IT budget footnote.
Gartner's landmark CRM licensing study found that 42% of purchased CRM licenses went completely unused, over a billion dollars in waste against a roughly $3 billion market at the time. The study is decades old now. The mechanism it documented (seat-based pricing charging in full for access nobody exercises) hasn't gone anywhere; if anything, the category has only grown more entrenched.
And Flexera's State of ITAM research puts wasted spend at roughly a third of the total IT estate, consistently, across desktop software, data-center software, and SaaS alike. That consistency is the real point. It's what happens whenever software is priced on access instead of use, wherever that software runs. Not a SaaS quirk. Not a symptom of cloud migration moving too fast.
It's the predictable output of a pricing model built to charge for access granted, not value delivered. A named seat can't tell the difference between someone using it productively all day and someone who logged in once, months ago, and never came back, and no amount of IT diligence changes that. Add up what a business actually pays across one renewal cycle: the vendor for the license, the systems integrator for the customization work, the platform itself simply for continuing to exist. Paying for what, exactly? The vendor controls what happens next, not the customer. That control is the actual product being sold, whether or not it's the one named on the invoice.
Layer two: the customization tax that compounds every year after
The license fee is only the entry ticket. Panorama Consulting's independent ERP research, one of the few research bodies tracking ERP total cost of ownership without a vendor relationship to protect, finds that professional services spend (implementation, customization, integration) typically runs 100–200% of the annual license fee, on top of the license itself. That figure isn't a one-time cost. It recurs every time a business process changes, a new module gets bolted on, or a version upgrade forces existing customizations to be rebuilt from scratch. Panorama's own manufacturing-specific survey found only 7% of ERP customers run their system "as-is" out of the box. The other 93% customize to some degree, and every one of those customizations is a small, recurring toll charged by whoever controls the underlying platform — a professional-services invoice, a module license, a "certified partner" day rate, take your pick.
It's valuable enough that private equity has begun consolidating the implementation-partner layer itself, acquiring ERP and EPM consultancies specifically for their ability to customize a platform and deepen a client's dependence on it. A firm that customizes a system more thoroughly bills more hours for the work and makes the client harder to leave in the process. That stickiness is now a real, financially exploited business model, not an unfortunate byproduct of technical debt. In a growing number of cases, deepening the customer's dependence on that specific platform is the actual investment thesis. Not the implementer acting as a neutral party helping a customer get more from their software.
This is the deeper reason Syspro's notice matters beyond Syspro's own installed base. A concurrent-to-named licensing change is one visible, sudden mechanism for extracting more revenue from an existing customer base. Customization fees, module upsells, and integration costs are quieter, steadier versions of the same extraction, spread out so they never trigger the attention a single formal notice does. Both come from the same root cause:
The vendor is paid for the platform existing, not for the outcome it produces.
The same playbook, five times over
Syspro isn't an outlier, and this isn't the first time it's happened this decade. It's the latest instance of a pattern that's run, in slightly different form, at nearly every major enterprise software vendor in recent memory. The specific mechanism differs vendor to vendor, and flattening four distinct playbooks into one would be a mistake, but the underlying behavior is identical every time: monetize access by counting people, then redefine what counts as a person, a seat, or a user, exactly when it's time to grow revenue from a customer base that hasn't changed what it actually does with the software.
SAP built an entire body of contract law around this move. In 2017, the UK High Court ruled against Diageo in a dispute over "indirect access," SAP's position that when a Salesforce integration fed data into Diageo's SAP system, everyone touching that integration, not just people logging into SAP directly, counted as a licensable named user. Diageo's reported exposure ran past £54 million for usage that hadn't changed at all. Only SAP's definition of "user" had. Anheuser-Busch InBev faced a similar claim, reportedly seeking more than $600 million before it was settled confidentially. The complexity hasn't eased since: in 2025, SAP retired its "RISE Premium" cloud tier for a new structure Info-Tech Research Group's Scott Bickley called a "stealth price increase" in industry coverage, describing the customer experience in a line that could serve as this paper's epigraph: "SAP licensing feels like a world where the rug is constantly being pulled from under one's feet."
Oracle runs a quieter version of the same play with a different mechanic. Its Named User Plus licensing requires a minimum of 25 licenses per processor regardless of how many people actually touch the system, a tax floor that exists independent of real usage. In January 2023, Oracle went further with Java SE specifically, replacing per-user licensing with an "Employee" metric that counts a company's entire headcount, whether or not most employees ever open Java. Reported bill increases ran as high as 30x for some customers, 2–5x being more typical. By 2025, roughly 90% of Oracle Java customers were actively evaluating a move to open-source alternatives, according to industry surveys.
Microsoft takes the quietest approach of the four: stacking, not redefining. Microsoft 365 Copilot launched in 2023 at $30 per user per month, not instead of an existing per-seat subscription, but layered directly on top of one a customer was already paying. The tax doesn't get restructured. It just accumulates.
Sage offers the closest real precedent to Syspro's own 2026 move. It stopped selling new perpetual licenses for Sage 50 at the start of 2023, using a security justification (deprecating older encryption protocols on a fixed deadline) to functionally force long-standing perpetual customers onto subscription pricing, whether they wanted the move or not. The trade press ran a multi-part investigation into the backlash, including one customer's account of losing a fifteen-year license with a decade of value still on the table, calling the vendor's compensation offer "insulting."
Four vendors, four different specific mechanisms: an indirect-access legal doctrine, an employee-count metric, fee-stacking, a forced perpetual-to-subscription conversion. Same category of behavior every time. A buyer who's actually negotiated a SAP or Oracle contract will rightly object to having their hard-won, vendor-specific knowledge flattened into someone else's story; these vendors don't run identical playbooks. But the pattern proves this was never a Syspro problem, or even an ERP problem. It's what happens whenever a vendor controls the mechanism its customer's cost is metered against.
The discount trap, and why your leverage disappears exactly when you need it most
Here's the part that never makes it into a sales deck, because it isn't in the vendor's interest to say out loud. The discount that got you into the relationship was the opening move in a negotiation that ends years later, on the vendor's terms. Not yours. It was never the favor it looked like at signing.
It plays out in almost the same shape across every vendor named above. A prospective customer gets a steep discount, a waived implementation fee, or a generous multi-year rate to sign in the first place, a rational trade for the vendor, since the value of a locked-in account justifies the cost of winning one. Syspro's own notice runs a version of this explicitly, crediting prior license payments toward the new subscription and "encouraging early conversion for maximum value." A discount dressed as a courtesy, with a deadline attached to make sure it's taken. Once the deal is signed, the real work begins: data gets loaded, processes get built around the system's specific quirks, integrations get wired to its API, employees learn workflows that exist nowhere else. None of that effort is wasted; it's exactly what a working system should produce. But every hour of it is also a small, permanent increase in switching cost. And switching cost is the only thing standing between a customer and whatever the vendor eventually decides to charge.
By the time the renewal conversation happens, the two sides aren't negotiating as equals, and both know it. The vendor has watched thousands of other customers run the same calculation and has a high-confidence estimate of exactly how expensive it would be for you to leave. The customer is often doing that math for the first time, on a deadline, with a system too deeply embedded to actually replace before the renewal date arrives. That asymmetry is the entire commercial logic of the model, not an accident of one bad negotiation: discount going in, because a locked-in account is worth the cost of winning it; extract going out, because the customer's leverage was already spent on the very success of the implementation. SAP's 2025 restructuring, Sage's forced migration, Syspro's own 2026 notice: three vendors, three different moments, the same script. The tax always arrives exactly when the customer is least able to say no to it.
Part 2: The Real Fix — Agents, Not Software You Configure
A different price on the same relationship doesn't fix a tech tax. It just changes its shape. A usage fee charged for logging into a dashboard is, functionally, a differently-shaped seat tax: the customer is still the one doing the work, only the billing mechanic changed. The actual fix has to sit upstream of pricing entirely: agents that execute the operational work directly, rather than a platform a customer's own people have to staff, configure, and operate regardless of what it produces that quarter.
Stated specifically rather than as a slogan, here's what agents do directly, today, without a human doing it first. Schema and data profiling against a customer's actual, live system. Candidate mapping proposals, each carrying a confidence score, when data needs to move from one system's model to another's. Business-rule extraction from existing scripts and stored procedures nobody currently understands well enough to document by hand. Transformation-code generation. Full-dataset reconciliation diffs, run continuously against production instead of checked once at go-live and assumed correct forever after. In an ongoing deployment, this same pattern extends to standing diagnostic work: reading a business's live data on a schedule and surfacing findings automatically, the way a good analyst would if one were watching the numbers every single day instead of once a month.
What stays deliberately human is just as specific, and for good reason. Ratifying a semantic mapping between two systems' data models is a judgment call, not a computation. An agent can propose with high confidence that one system's "Customer Status" field corresponds to another's "Account Type," but only a person with real domain context can confirm the mapping doesn't quietly break an edge case the agent has no way of knowing matters (a status value that means something different for one specific customer segment, say, for reasons buried in a decade of institutional history no schema documents). Signing off a system cutover, approving a recommendation large enough to matter, owning the actual client relationship: these sit in the same category. Agents genuinely compress hours of manual excavation into a decision a person can review in seconds. That's a claim about surfacing a decision-ready option quickly. It is not a claim about removing the judgment itself. This is a deliberate, named boundary, not a hedge, and it answers the question anyone should ask when a vendor says "our platform is really a team": if something goes wrong, who's actually accountable, and for what? The answer here is a named list of human responsibilities. Not an assertion that the team is "mostly agentic" and therefore doesn't need one.
And here's what this claim is not, because the honest version of this pitch is more credible than a cleaner-sounding one. This is not a claim that implementation, migration, hosting, and ongoing support become free. They don't. A real system migration carries real cost and genuine risk; data-conversion errors and reconciliation failures at cutover are a named, serious risk in any migration, not a footnote to wave past. The accurate way to state the economics: Performis replaces a seat tax with a smaller, front-loaded implementation cost, followed by ongoing pricing tied to usage. Not a claim of "near-$0" total cost, and not a rip-and-replace of a customer's entire technology stack. What actually changes is which costs are real and which are artificial. The cost of migrating and operating a system honestly reflects the work involved in doing that migration and operation well. The cost of a named-user tax reflects nothing but headcount, whether or not that headcount is creating any value for the business on the day it's being billed for.
Proven across three verticals, not one
None of this is a plan for some future engagement. It's the live operating model behind three real, currently running engagements in three genuinely different industries, each arriving independently at the same division of labor.
A multi-office litigation practice. Agents read incoming case filings and derive the jurisdiction-specific procedural rules that determine deadlines and required filings, automatically, across courts with meaningfully different rule sets. A person ratifies the genuinely ambiguous edge cases and signs off before anything reaches a firm calendar or a client-facing deadline; a missed procedural deadline in litigation carries malpractice exposure, not just an error to correct and move past. The firm's senior partner frames it as more than a calendaring fix. Agents that raise the quality and consistency of scheduling work also raise the job satisfaction of the clerks and assistants who used to own it by hand, and he's already looking at extending the same model across the firm's other operational workflows. Not just deadlines.
A payer-contracting system in healthcare. Agents extract and structure contract terms and negotiated rate data out of payer agreements at a volume and consistency no human team reviews manually, the kind of document-heavy, detail-dense work that used to consume analyst-hours by the week. A person ratifies the final negotiating position before it's ever used in an actual conversation with a payer, because that's a business decision with real financial consequences, not a data-extraction task.
A discrete manufacturer running Syspro ERP. Agents profile the ERP's schema and monitor operational risk (cycle-time deviation, capacity constraints) continuously rather than on a reporting cadence. A person ratifies every data-migration mapping and signs off the system cutover itself, the same boundary described above, applied to a live manufacturing environment.
Three industries, three structurally different workflows, the same division of labor holding in every one: agents doing the specific, named technical work, humans making the specific, named judgment calls. That consistency across unrelated domains is itself evidence this is a real operating model. Not a framing applied after the fact to whatever happened to get built.
Part 3: The market is already moving here — Performis got there first
None of Part 2's argument depends on taking Performis's word for it that agentic execution changes what's fair to charge for. The wider software market, including companies with every commercial incentive to defend the per-seat model that built their existing revenue, is independently arriving at the same conclusion, recently enough that the shift is still genuinely newsworthy rather than settled history.
Salesforce, arguably the company most responsible for popularizing per-seat SaaS pricing in the first place, launched its Agentforce AI product in late 2024 priced per conversation, revised that model to Flex Credits sold in bulk, and has since moved again to pure outcome-based pricing: $2 per resolved case, with an explicit rule that a case escalated to a human isn't billed at all. Three pricing iterations in under two years, from the company that arguably wrote the per-seat SaaS playbook everyone else copied. That's a fast, public admission the old model doesn't fit an AI-agent product.
Intercom's Fin charges $0.99 per resolved outcome, and is explicitly marketed as capable of running on top of a competitor's seat-priced helpdesk platform "with no seats required." That's as direct a statement as a vendor can make that seats and outcomes are now understood as two separate, competing units of value in the same market.
Sierra, the AI customer-agent company founded by Bret Taylor, Salesforce's own former co-CEO, states its positioning publicly and without much hedging: legacy customer-experience vendors are "trapped in seat-based revenue models" that actively work against deploying effective AI agents, and outcome-based pricing "eliminates shelfware" by construction, since there's no such thing as an unused outcome the way there's an unused seat. Worth noting: the person making this argument most publicly used to run the company most associated with the model he's now arguing against.
Even inside ERP specifically, the exact category this paper opened with, Acumatica has built genuine, durable market share on unlimited-user, consumption-based pricing rather than per-seat or per-named-user licensing. That proves a usage-priced alternative to concurrent or named-user ERP licensing is a viable, currently-competing commercial model today. Not a theoretical future one.
Analysts are describing the same inflection point from outside any single vendor's incentives. a16z's enterprise newsletter argued in December 2024 that per-seat pricing is "no longer the natural atomic unit" once AI agents absorb the underlying work a seat used to represent. Industry coverage of IDC's FutureScape 2026 predictions forecasts pure seat-based pricing will be functionally obsolete by 2028, with most software vendors expected to refactor their pricing models around new value metrics, a view Bain & Company echoes, framing it as customers increasingly expecting to pay based on outcomes achieved, "not log-ons."
Our own reasoning for choosing this model is simpler than a trend argument. An outcome-based fee only works if the vendor's incentives actually match the customer's, and tying payment to the outcome itself is the cleanest way to prove that alignment is real. When the outcome doesn't happen, we don't get paid for it. That's the whole mechanism. Not a marketing claim layered on top of the pricing; the actual thing the pricing is built on.
What this section proves, and what it doesn't, is worth stating plainly. It proves the pricing shift described in this paper is real, already underway, and driven by sophisticated, well-capitalized companies with no reason to chase a fad; this is not a fringe idea Performis invented to sound current. What it doesn't prove, on its own, is that Performis is meaningfully different from Acumatica specifically, because Acumatica already sells usage-based ERP pricing today, full stop. If "we price on usage, not seats" were the entirety of Performis's pitch, it would be a feature already shared with an established, credible competitor. Not a category difference worth a white paper. The actual difference is everything in Part 2: Performis's claim is backed by a live, checkable mechanism running across three real engagements in three real industries. Not a rate card structured differently from a competitor's.
Part 4: A world without the tax
Strip the tax out entirely and picture what's actually left.
Adding a new person to a growing team doesn't trigger a conversation about whether there's a spare license, which tier of "user" they should be provisioned as, or whether the budget can absorb one more seat this quarter. Access was never the thing being charged for, so it isn't a constraint. A finance team stops treating its annual renewal as a bracing-for-impact exercise, because the fallback of walking away was never lost to begin with: there's no proprietary system holding years of switching cost hostage, the way there was for Diageo, for Sage's perpetual-license holdouts, for every Syspro customer reading this year's notice. The conversation with a software provider shifts from "how many seats do we need this year" to "what did the system actually do for us this year," a question every vendor named in this paper would rather not have to answer in public, because it's a much harder question for a seat-metered product to answer well.
This isn't a hypothetical Performis is inventing from nothing. a16z's own December 2024 analysis makes the mechanism explicit: Zendesk currently charges roughly $115 a month per support-agent seat regardless of how many tickets that agent actually resolves. Once an AI agent does the resolving instead of a human, "the natural pricing metric becomes successful outcomes," not seats. That's precisely why AI-native companies building customer-facing agents from scratch (Decagon, pricing by resolved conversation, among others) are skipping seat-based pricing entirely rather than retrofitting it onto a new product.
The honest caveat, stated plainly rather than implied: none of SAP, Oracle, Microsoft, or Sage show real evidence of moving their core ERP or platform pricing this direction yet. If anything, the pattern from Part 1 shows the opposite: incumbents doubling down on seat- and headcount-based metering at the exact moment the rest of the software industry is visibly moving toward outcomes. That's the argument, not a weakness in it. The vendors with the most to lose from a world without the tax are, unsurprisingly, the last ones who'll build it. Someone else has to. That's the whole reason Part 5 exists as a real, worked example rather than a thought experiment.
Part 5: A manufacturer's full transition, start to finish
To make this concrete rather than abstract, here's what the pattern looks like end to end, for one manufacturer.
A Midwest-based manufacturer of vehicular lighting systems supplies five original-equipment channels (trailer, work truck, marine, emergency, and transit) from a single facility running Syspro ERP across more than 4,000 active part numbers, on a three-shift operating schedule.
Diagnose. When Syspro's 2026 licensing notice arrived, the plant's IT team had exactly one number to work with: a concurrent-license count on the existing invoice. Nobody had an answer for how many distinct people actually touched the system across a full day's rotation of shifts, a number that simply never had a commercial reason to exist before. Performis's Renewal Exposure assessment, run as a read-only connection against the plant's own Syspro database, took an afternoon to produce the real figure: 52 concurrent licenses translated to just under 290 distinct named identities once every shift, every warehouse login, and roughly a dozen legacy e.net integrations were actually counted individually, the way the new subscription model would count them. At list-rate pricing, that gap represented a real exposure of roughly $390,000 a year, more than four times what the plant had originally budgeted for the renewal.
Bridge, then migrate. Rather than negotiate against Syspro with no real number of their own, the plant used that assessment to secure a 12-month bridge term with a capped renewal uplift clause, buying room to make a deliberate decision instead of signing a multi-year agreement under deadline pressure, exactly the trap a forced renewal deadline is designed to create. Over the following two quarters, a certified open-source implementation team migrated the plant's item master, bill-of-materials structure, and open work orders onto an ERPNext foundation, while Performis's reconciliation agents ran the full trial-balance, inventory-valuation, and BOM-explosion checks (the same category of audit-grade verification described in Part 2) that let the plant's CFO sign off the cutover on the strength of evidence. Not faith in the implementation team's assurance that everything had gone correctly.
Run the agentic layer. Two agentic workflows went live alongside the new system. One tracks mold-tool cycle-time deviation against engineering standard continuously across the plant's entire press fleet, instead of surfacing deviations only when someone happens to review a monthly report. The other models work-center capacity thirty days out, specifically to flag revenue-at-risk from an emerging bottleneck before it actually causes a missed delivery. Not after a customer has already noticed.
The results, within two quarters of running on the new platform: the projected $390,000 licensing exposure was eliminated entirely, replaced by a usage-priced agentic layer running at roughly a quarter of that annual cost. $180,000 in previously invisible lost capacity was surfaced at a single constrained press station running at 97% utilization, caught by the capacity-monitoring agent before it caused a missed shipment, not discovered afterward in a post-mortem. And unplanned mold-tool downtime fell by an estimated 30%, driven by flagging cycle-time deviation as it happened rather than discovering it weeks later in a scheduled report.
The plant's CEO frames all of it as a direction, not a single project. Every routine, manual step still running on judgment and spreadsheets is a candidate for replacement, starting with supplier shipment scheduling and extending outward from there. Each one is another hidden cost or software tax extracted from the business and converted into a measurable, compounding gain.
What this means
A perpetual software license was a negotiating position more than it was ever really about the software itself: concrete proof that a customer didn't have to accept whatever a vendor decided to charge at the next renewal, because the fallback of simply not renewing and continuing to run what they already owned was real and available. Named-user subscription pricing, layered on top of a proprietary system a customer doesn't ultimately control, removes that fallback permanently, for every customer it touches. Syspro's installed base is simply the clearest, most recent, most visible place to watch that removal happen in real time.
The alternative described here is a genuinely different relationship, not a cheaper version of the same one dressed up in new pricing language: agents that do real, specific, accountable work (already proven, today, across legal, healthcare, and manufacturing engagements) running on a foundation no single vendor can unilaterally re-tax at the next renewal, priced against the usage and value that work actually creates instead of against how many people happen to be logged in that month.
See where your own exposure sits first: the Syspro Renewal Exposure Calculator estimates your named-user count and annual subscription cost from your concurrent licenses and shift pattern, free, no sign-up, in seconds. Every operation is different, and a calculator can only ever be a starting estimate. When you want to walk through what this actually looks like for your own facility, talk to Performis directly.