How to Pressure-Test the Seat-based SaaS Business Model Against the Impact of Gen AI

Estimated reading time: 11 minutes

Is the Seat Compression Thesis Accurate?

One of the most consequential debates in enterprise software business right now surrounds this question: is the seat-based SaaS business model doomed by Gen AI? The core argument revolves around the idea of not needing as many seats for established types of enterprise software solutions. For example, with the introduction of agents into customer service or human capital management, organizations will not need as many employees and can reduce seat purchases as agents allow the remaining humans to scale productivity. This topic has been adopted and attacked intensely with most of the commentary falling into two camps: the bears who extrapolate a straight line to zero seats, and the bulls who hand-wave about “pricing evolution” without doing the math. This Research Note walks through both cases and builds an approach to modeling the outcomes.

Separately, MGI Research built the MGI Pricing Migration Model, an interactive online calculator to help companies estimate the impact of migrating from a seat-based approach to what MGI terms Multi-Mode Monetization (M3) – a hybrid pricing approach in which multiple pricing and selling models are brought under a single umbrella. Contact MGI Research to access the MGI Pricing Migration Model.

The Core Tension

The bear case is structurally simple: if an AI agent can do the work of three analysts, a company that previously bought thirty Salesforce seats now buys ten (or maybe even zero) seats and builds the capability themselves as the cost of code is approaching zero and the cost of inference is “expected to be negligible”. Doing this cowboy math across the install base translates to revenue compression even if retention stays high. It’s clean, intuitive, and partially correct – which is what makes it dangerous as an investment thesis.

The problem: it treats the enterprise software market as a closed system with a single pricing variable. This isn’t true now, never was, and will not be in the future. Moreover, the bears are conflating adoption of Gen AI with a greater emphasis on DIY software development. They see Gen AI adoption as a zero-sum game for software – code is free (or nearly so and will be shortly), therefore software vendors are not needed, and the seat compression argument is part of that larger debate.

A Framework for Decomposing the Argument

A more realistic scenario can be described through a simple reference model which breaks the seat analysis into five layers: Seat Elasticity, Migration Velocity, Workflow Expansion, Value Migration, and Competitive Dynamics. In practical terms, each of these layers needs independent assessment per vendor and market.

Layer 1: Seat Elasticity by Workflow Type

Not all seats are equally exposed. One needs to categorize a vendor’s revenue by the nature of the work the seat enables:

  • Human-in-the-Loop seats – where the software supports a decision-maker who isn’t going away (e.g., a CFO using a planning tool, a lawyer reviewing a contract in a CLM system, etc. And no, agents are not licensed to practice law). These seats are sticky. AI makes the person more productive, but the person and their tasks remain the unit of value.
  • Throughput seats – where the seat exists because a human was the only available processing unit (e.g., data entry clerks, first-pass document reviewers, tier-1 support agents). These are genuinely exposed to compression.
  • Collaboration seats – where the seat exists for visibility, communication, or approval workflows (Slack, Jira, lightweight CRM users). These are partially exposed but stickier than people think because the organizational graph doesn’t shrink as fast as the task graph. The value of these seats, in part, is connected to a network effect. As more seats and more data contained within them are connected, their value rises both directly and indirectly.

A vendor with 70% of revenue from human-judgment seats has a fundamentally different risk profile than one with 70% throughput seats. Most businesses and investors aren’t doing this decomposition.

Layer 2: The Pricing Model Migration Velocity Question

The “SaaS inertia” argument assumes vendors can’t or won’t evolve pricing. This deserves scrutiny on both sides.

The bear contention that pricing model transitions are brutally hard is right. Pricing model changes are painful and stress many parts of the organization, not just sales. Every enterprise software company that attempted a major pricing pivot (e.g., from perpetual licenses to subscriptions, from per-seat to consumption) endured a painful transition valley. Investors punish it. Customers scream and shout. Sales comp plans break. Channel partners revolt. The organizational muscle memory around enterprise deals and seat-based selling is deep, and the transition cost is both real and measurable in quarters of depressed GAAP growth. Many management teams will get re-staffed with individuals who have practical experience migrating pricing models and installed bases. Transitioning to a new model while carrying the legacy pricing models is mostly slower and more expensive than being pure play.

However, the bulls are correct in highlighting this change isn’t unprecedented. Recall Adobe’s transition to subscriptions as an example and consider the fact that this migration is already underway. MGI built an analytical model abstracting a dual-revenue architecture in which a vendor maintains seat-based pricing for the human-judgment layer while layering in consumption or outcome-based pricing for the AI-agent layer on top. This isn’t hypothetical – ServiceNow, Salesforce, Workday and others are already structuring AI agent pricing as a distinct SKU with usage-based or per-resolution economics. The question of migration velocity is valid and is a judgement on the quality of incumbent management teams.  So is the question about lack of management maturity and potential financial viability of most AI start-ups.

The arrival of Gen AI Is like a torpedo hitting midships but failing to explode and instead just generating shock. An event like this exposes incumbent software vendors’ weaknesses at every level. And amongst the disruptors, there are literally thousands of new “AI-native” start-ups. The number of good (not even great) management teams is an order of magnitude less. Then, there are AI-specific risks in energy, regulation (the bear argument exposes the ugly truth that good jobs will be lost), and a myriad of technical and operational issues. Will AI companies begin to implode because of these risks? There is then the inertia of enterprises to consider. AI will have a huge impact on SaaS, but its uncontested success is far from assured.

The key analytical question isn’t “will incumbents evolve?” – it’s “what’s the net revenue impact of the transition period, and how deep is the valley of despair?” A separate part of the assessment should focus on the skill set of the management team and its track record.

Layer 3: The Denominator Problem – Total Addressable Workflow Expansion

This is where the seat-compression thesis most often falls apart under scrutiny. It assumes a static denominator: the same workflows, use cases, and organizational scope.

Historically, every wave of productivity improvement in enterprise software has expanded the addressable workflow set. When spreadsheets automated manual calculations, we didn’t end up with fewer finance workers. We ended up with dramatically more analysis being done, by more people, in more contexts. When CRM moved to the cloud, the user base expanded from dedicated sales ops teams to every customer-facing employee.

The better question is: does AI-driven productivity create enough new addressable workflows and new categories of users to offset seat compression in existing ones? In CPQ, for example, the current penetration is approximately 15% of addressable enterprises. If AI makes CPQ dramatically easier to deploy and operate, the expansion of the buyer base could dwarf the seat reduction within existing accounts.

This isn’t a guaranteed offset. It’s an empirical question that varies by market maturity, current penetration, and deployment friction. To ignore it, as the simple seat-compression model does, is analytical malpractice.

Layer 4: The Value Capture Migration

Even in scenarios where seats genuinely compress, revenue doesn’t necessarily follow linearly. The reason is vendors have multiple levers for value capture:

  • Price per remaining seat increases – by adding AI to the core product, suppliers will now argue each remaining user is now 3x more productive because of embedded AI, and the willingness-to-pay per seat reflects that, going up over time. A vendor is no longer selling a seat to a data-entry clerk. It is empowering a human with an AI copilot and agents to help generate 3x the output.
  • Platform fees and AI consumption charges – the agent itself becomes a billable entity. Whether it’s per-transaction, per-resolution, per-document, or per-API-call, the AI work unit becomes a revenue line distinct from platform fees.
  • Outcome-based pricing – the most disruptive, the least likely to get adopted, but also potentially the highest-margin model. Charge based on business outcomes delivered (contracts processed, quotes generated, revenue billed or recognized) rather than inputs consumed.

The end result: lose 40% of seats, increase price by 50% on remaining seats, add a new AI-consumption line worth 30% of prior seat revenue, and you end up roughly flat to slightly up. The true gain being a better margin profile because the cost of serving has also dropped. MGI’s Pricing Migration Model takes this dynamic into account.

Layer 5: The Competitive Dynamics and Disruption Vector

The biggest risk for incumbent vendor seat compression is AI-native entrants building the same workflow capability with no seats at all. This is already under way and experiencing classic early-stage challenges of reliability, security, governance, etc. Today, these teething issues serve as a buffer for competition between incumbents and new entrants. A new CLM or CPQ vendor starting with an agent-first architecture and outcome-based pricing doesn’t have the transition rut problem. They just build the new model from scratch. The early SaaS-native cycle is now playing out again but with SaaS vendors on the defensive against disruption.

This is the actual existential risk for incumbents, and it’s separate from the seat-compression question. The seat-compression debate is really a proxy for the deeper question: do incumbents’ data moats, process and domain expertise, integration depth, partnerships, and customer relationships give them enough time to navigate the transition before AI-native competitors can reach feature and trust parity? If the deep logic contained within enterprise software really defensible?

The Realistic, Non-Dogmatic Answer

Seat compression is real but unevenly distributed. Throughput-heavy vendors in mature, high-penetration markets with limited pricing flexibility will face 15-30% seat-count headwinds over the next 3-5 years. This is notable and it would be foolish to dismiss.

The “seats go to zero” narrative dramatically overstates the case. This argument ignores workflow expansion, pricing migration (which is hard but not impossible), the potential for seat price increases, and the creation of entirely new billable units around AI agents. Subscriptions were once touted as a lower cost alternative to enterprise software agreements. In reality, most SaaS vendors, even the PLG-natives like Atlassian, have enterprise software agreements. As a result, growth in software markets only accelerated and so have margins (e.g. Microsoft has a 56% EBITDA margin).

The net revenue impact for well-managed vendors is likely modestly positive over a 5-year horizon, but with a transition rut of four to eight quarters of suppressed growth during the pricing model shift. The variance around the central case is wide, and not all vendors will make it across the transition valley.

The real risk isn’t ...