Your SaaS Product Isn't the Problem. Your Business Model Is.

Introduction

Last month, Thomson Reuters dropped 18%. LexisNexis fell 14.4%. Wolters Kluwer shed 13%. This happened within days of Anthropic launching a browser-based AI tool that automates the kinds of workflow tasks these companies charge per seat for.

Those weren't the outliers. Over $2 trillion has been wiped from software company valuations in the months since. The software sector's weight in the S&P 500 dropped from 12% to 8.4%. Figma is down 65% from its peak. Duolingo lost 68% of its market cap in six months. SaaS indices are down 22% while the broader market gained 17.6%.

If you're a SaaS founder reading this, the instinct is to dismiss it as public market noise. Your product is solid, your customers are still paying. But the uncomfortable truth is that most commentators have the story wrong. This isn't an AI disruption story. It's a business model crisis that AI has simply made impossible to ignore. And for founders running businesses with 10 to 40 employees, the decisions made in the next six months will determine whether your company adapts or gets caught in the downdraft.

This article examines what's actually breaking in the SaaS model, where value is shifting, and what you can do about it before the market makes the decision for you.

A Trend That Didn't Start With AI

SaaS growth has been declining every single quarter since 2021. Before ChatGPT, before Claude, before anyone used the word "agent" outside a spy novel.

What changed was the budget. AI is now consuming the IT spend that used to buy SaaS seats. Meta is spending $135 billion a year on AI infrastructure. Microsoft, $75 billion. The hyperscalers collectively will pour $470 billion into AI in 2026. Every one of those pounds is a pound that never reaches another Salesforce seat or Workday module.

Meanwhile, the "best-of-breed" era is collapsing. CIOs don't want five point solutions for marketing, sales, support, finance, and operations anymore. They want fewer vendors, less overhead, less risk. UK businesses in particular are consolidating their software stacks, not expanding them.

And much of what passed for "growth" over the past three years wasn't genuine growth at all. It was price hikes on captive customers, expansion revenue from existing accounts, and contract uplifts with no connection to real product adoption. Strip those away, and net-new customer acquisition, the metric that actually matters, is the weakest line in public earnings.

AI didn't cause this structural crisis. It turned up the volume.

Why One Seat No Longer Means One Pound

The classic SaaS formula, seats multiplied by ARR per seat multiplied by growth multiple, assumed that each employee remained the unit of work. AI agents have severed that assumption and the charging model along with it.

An AI-driven sales development agent can replace an £80,000-a-year SDR and a £1,800-a-year licence with a few hundred pounds of compute each month. That's an 80 to 90% drop in revenue per account, while the customer's productivity actually increases. The vendor's revenue collapses while the customer's output rises, and the entire model breaks.

About 35% of SaaS companies have responded by hiking seat prices. This is late-stage extraction, squeezing more rent from a dying revenue stream. It's self-defeating: the higher you push seat prices, the faster CFOs approve the AI replacement project.

The remaining 65% are experimenting with hybrid pricing, layering seat-based base fees with a usage or outcome component. It's a sensible transitional move, but the sales teams, compensation plans, and billing systems are all built around seat counts. Re-engineering those structures is an operational problem, not just a product one.

Salesforce illustrates the contradiction in real time. They launched Agentforce at $125 to $550 per user per month, on top of already-elevated base pricing, at the exact moment their enterprise customers are cutting headcount because AI is handling more work. The CFOs noticed. Per-seat pricing is becoming a liability, not an asset.

What AI Has Commoditised, and What It Can't Touch

I've spent the past six months testing native AI workflows: browser automation, skills-based systems, and agent orchestration. Over that time, the pattern has become clear.

Anything probabilistic is in danger. Pattern matching, content generation, recommendations, simple workflow automation. If a foundation model can replicate 90% of your core value at 1% of the cost, your moat is gone. Workflow automation is the number one category being replaced right now. Retool's 2026 report says 35% of teams have already swapped out at least one SaaS tool for a custom-built solution. 78% plan to build more this year.

Most founders miss something important here: AI agents don't use your interface. They call APIs directly. Your UI, the thing your design team spent two years perfecting, becomes optional when the user isn't a human clicking buttons. One founder told me he chose between CRM platforms purely because the APIs were more robust and feature-rich. The interface was irrelevant. This is a fundamental shift in how organisations should think about integrating AI into their operations.

But deterministic systems of record are holding up. Platforms that store proprietary customer data, enforce compliance, or run mission-critical logic where a single error has material consequences. AI can't easily replicate that function. Toast owns every menu item, pricing change, and supplier relationship for thousands of restaurants. Procore owns the project data for construction sites. That depth of integration is the moat.

The survival line is simple: if your core value is probabilistic, you're exposed. If it's deterministic, you've got a foundation to build on. Be honest about which side you're on.

Five Questions Every SaaS Founder Should Answer Monday Morning

Before you touch your pricing page or rewrite your pitch deck, sit with these. Answer honestly. If you flinch, you've found the vulnerability.

1. "What would break if we disappeared tomorrow?" High switching costs and deep integration indicate a platform. If customers can find alternatives within a week, your offering is a feature, not a business.
2. "Are we the tool or the workflow?" Tools get replaced. Workflows become embedded. If your product is where customers think, decide, and store institutional knowledge, you own something AI can't easily replicate. If it's where they execute tasks that an agent could handle, you're exposed. Understanding where your value proposition sits in the hierarchy matters more than ever.
3. "Who would fight to keep us?" Communities create defensible advantages that AI cannot replicate. Users who actively contribute don't exist in a training set. If no customer would advocate for your product, it's a line item, not a partner.
4. "Can AI replicate our trust layer?" Governance, compliance, audit trails, human-in-the-loop oversight. With the EU AI Act enforcing in August, the companies that build clear accountability frameworks become the trusted layer between raw AI capability and real-world consequences.
5. "Is our core value probabilistic or deterministic?" This is the survival line. If your core is pattern matching and recommendations, you're in the danger zone. If it's deterministic systems of record, you have a foundation.

Any uncomfortable answer highlights an area that needs strengthening now, not next quarter.

The Business Model That Actually Works

Per-seat may be dying, but outcome-based pricing isn't plug-and-play either. When one AI-native SDR company offered customers both outcome-based and activity-based pricing, 90% chose activity-based. Defining "qualified lead" or "successful transaction" varies so widely across industries that negotiation friction kills deals.

Intercom's experience with their AI chatbot Fin offers another cautionary note. They launched with outcome-based pricing for AI chat resolution, then discovered that suggested prompts were consuming nearly half of the product's operating cost while seeing less than 1% usage. Even smart companies can get the model transition wrong if they don't understand their own unit economics first.

I've experienced this tension while launching Value by Design. My initial instinct was to charge a flat fee for the whole package. Clean, simple, seemingly right. But after prototyping, it became clear from the questions people asked that a tiered base with optional layers on top was what customers actually wanted.

The more practical path is a hybrid model. Charge a base platform fee that covers your deterministic core: data governance, the compliance layer, community infrastructure. That gives you predictable revenue. Then add a consumption-based AI layer on top, metered per inference, per API call, or per workflow execution, with transparent usage dashboards so customers control their spend.

This isn't a simple price tweak. It involves retraining sales teams, redesigning compensation, and rebuilding billing pipelines. The operational friction is real. But the founders who move first on hybrid pricing will capture the trust premium. The ones who wait will be forced into it on worse terms.

The Strategy Principles Haven't Changed

The technology shifted. The fundamentals of good strategy haven't. And if you are relying on AI to navigate this shift, it is worth understanding why LLMs default to trendy advice over context-specific strategy.

Know the real job your customer is hiring you for. If your product is the place where customers think, decide, and store institutional knowledge, you own something AI can't easily replicate. Design your model to capture that value, not the number of seats you sell. If your value proposition feels right but isn't converting, the pricing model might be the bottleneck, not the positioning.

Always test before you scale. If you're not devoting 10% of your budget to testing, you're not learning fast enough. Pilot a hybrid pricing tier with a handful of anchor accounts. Measure churn, lifetime value, and unit economics. If that works, roll out.

The per-seat era appears to be ending. But what replaces it rewards the same things that have always mattered: clarity about who you serve, depth of understanding about what they need, and the discipline to build something genuinely hard to copy. CharlieHR is a case in point: one segment, one outcome, no enterprise dilution. That kind of specificity is harder to commoditise than a feature set.

I wrote something recently that I keep returning to:

"The only thing that matters now is what can't be generated. That's where all the money is hiding."

What's Your Next Move?

If you're a SaaS founder working through these questions and wondering where the pivot is, I'd welcome the conversation. Whether you need help rethinking your pricing model, clarifying your value proposition, or building a strategy that accounts for how AI is reshaping your market, that's the work we do at Polything.

No sales pressure. Just a straight conversation about what can be done to help.

Book a call. No sales pressure, just a straight conversation about what's next.